Built site for gh-pages

2025-03-05 15:03:12 +00:00
parent 9270933a28
commit e6453035a7
43 changed files with 2748 additions and 1364 deletions
--- a/.nojekyll
+++ b/.nojekyll
@@ -1 +1 @@
-8c1dbc37
+8418d357
--- a/FAQS.html
+++ b/FAQS.html
@@ -206,6 +206,12 @@ ul.task-list li input[type="checkbox"] {
      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
          <li class="sidebar-item">
  <div class="sidebar-item-container"> 
+  <a href="./docs/docker.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Docker</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
  <a href="./docs/multi-gpu.html" class="sidebar-item-text sidebar-link">
 <span class="menu-text">Multi-GPU</span></a>
  </div>
--- a/TODO.html
+++ b/TODO.html
@@ -206,6 +206,12 @@ ul.task-list li input[type="checkbox"] {
      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
          <li class="sidebar-item">
  <div class="sidebar-item-container"> 
+  <a href="./docs/docker.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Docker</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
  <a href="./docs/multi-gpu.html" class="sidebar-item-text sidebar-link">
 <span class="menu-text">Multi-GPU</span></a>
  </div>
--- a/docs/amd_hpc.html
+++ b/docs/amd_hpc.html
@@ -130,7 +130,7 @@ pre > code.sourceCode > span > a:first-child::before { text-decoration: underlin
      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
        <i class="bi bi-layout-text-sidebar-reverse"></i>
      </button>
-        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../docs/multi-gpu.html">Deployments</a></li><li class="breadcrumb-item"><a href="../docs/amd_hpc.html">AMD GPUs on HPC Systems</a></li></ol></nav>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../docs/docker.html">Deployments</a></li><li class="breadcrumb-item"><a href="../docs/amd_hpc.html">AMD GPUs on HPC Systems</a></li></ol></nav>
        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
        </a>
    </div>
@@ -241,6 +241,12 @@ pre > code.sourceCode > span > a:first-child::before { text-decoration: underlin
      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
          <li class="sidebar-item">
  <div class="sidebar-item-container"> 
+  <a href="../docs/docker.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Docker</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
  <a href="../docs/multi-gpu.html" class="sidebar-item-text sidebar-link">
 <span class="menu-text">Multi-GPU</span></a>
  </div>
@@ -453,7 +459,7 @@ pre > code.sourceCode > span > a:first-child::before { text-decoration: underlin
 <!-- main -->
 <main class="content" id="quarto-document-content">

-<header id="title-block-header" class="quarto-title-block default"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../docs/multi-gpu.html">Deployments</a></li><li class="breadcrumb-item"><a href="../docs/amd_hpc.html">AMD GPUs on HPC Systems</a></li></ol></nav>
+<header id="title-block-header" class="quarto-title-block default"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../docs/docker.html">Deployments</a></li><li class="breadcrumb-item"><a href="../docs/amd_hpc.html">AMD GPUs on HPC Systems</a></li></ol></nav>
 <div class="quarto-title">
 <h1 class="title">AMD GPUs on HPC Systems</h1>
 </div>
--- a/docs/batch_vs_grad.html
+++ b/docs/batch_vs_grad.html
@@ -207,6 +207,12 @@ ul.task-list li input[type="checkbox"] {
      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
          <li class="sidebar-item">
  <div class="sidebar-item-container"> 
+  <a href="../docs/docker.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Docker</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
  <a href="../docs/multi-gpu.html" class="sidebar-item-text sidebar-link">
 <span class="menu-text">Multi-GPU</span></a>
  </div>
--- a/docs/cli.html
+++ b/docs/cli.html
@@ -240,6 +240,12 @@ pre > code.sourceCode > span > a:first-child::before { text-decoration: underlin
      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
          <li class="sidebar-item">
  <div class="sidebar-item-container"> 
+  <a href="../docs/docker.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Docker</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
  <a href="../docs/multi-gpu.html" class="sidebar-item-text sidebar-link">
 <span class="menu-text">Multi-GPU</span></a>
  </div>
--- a/docs/config.html
+++ b/docs/config.html
@@ -241,6 +241,12 @@ pre > code.sourceCode > span > a:first-child::before { text-decoration: underlin
      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
          <li class="sidebar-item">
  <div class="sidebar-item-container"> 
+  <a href="../docs/docker.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Docker</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
  <a href="../docs/multi-gpu.html" class="sidebar-item-text sidebar-link">
 <span class="menu-text">Multi-GPU</span></a>
  </div>
@@ -616,417 +622,425 @@ pre > code.sourceCode > span > a:first-child::before { text-decoration: underlin
 <span id="cb1-157"><a href="#cb1-157" aria-hidden="true" tabindex="-1"></a><span class="at">      </span><span class="fu">system</span><span class="kw">:</span><span class="at"> </span><span class="kw">[</span><span class="st">"system"</span><span class="kw">]</span></span>
 <span id="cb1-158"><a href="#cb1-158" aria-hidden="true" tabindex="-1"></a><span class="at">      </span><span class="fu">tool</span><span class="kw">:</span><span class="at"> </span><span class="kw">[</span><span class="st">"tool"</span><span class="kw">]</span></span>
 <span id="cb1-159"><a href="#cb1-159" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb1-160"><a href="#cb1-160" aria-hidden="true" tabindex="-1"></a><span class="co">    # IMPORTANT: The following fields determine which parts of the conversation to train on.</span></span>
-<span id="cb1-161"><a href="#cb1-161" aria-hidden="true" tabindex="-1"></a><span class="co">    # Priority order: message_field_training &gt; message_field_training_detail &gt; train_on_inputs or role in roles_to_train</span></span>
-<span id="cb1-162"><a href="#cb1-162" aria-hidden="true" tabindex="-1"></a><span class="co">    # See examples at `docs/dataset-formats/conversation.qmd`</span></span>
-<span id="cb1-163"><a href="#cb1-163" aria-hidden="true" tabindex="-1"></a><span class="co">    # Note: If the below 4 fields are set to empty, defaults to training only on the last message.</span></span>
-<span id="cb1-164"><a href="#cb1-164" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb1-165"><a href="#cb1-165" aria-hidden="true" tabindex="-1"></a><span class="co">    # Optional[List[str]]. Roles to train on. The tokens from these roles will be considered for the loss.</span></span>
-<span id="cb1-166"><a href="#cb1-166" aria-hidden="true" tabindex="-1"></a><span class="at">    </span><span class="fu">roles_to_train</span><span class="kw">:</span><span class="at"> </span><span class="kw">[</span><span class="st">"assistant"</span><span class="kw">]</span><span class="co">  # default</span></span>
-<span id="cb1-167"><a href="#cb1-167" aria-hidden="true" tabindex="-1"></a><span class="co">    # Optional[str]. Which EOS tokens to train on in the conversation. Possible values are:</span></span>
-<span id="cb1-168"><a href="#cb1-168" aria-hidden="true" tabindex="-1"></a><span class="co">    # - all: train on all EOS tokens</span></span>
-<span id="cb1-169"><a href="#cb1-169" aria-hidden="true" tabindex="-1"></a><span class="co">    # - turn (default): train on the EOS token at the end of each trainable turn</span></span>
-<span id="cb1-170"><a href="#cb1-170" aria-hidden="true" tabindex="-1"></a><span class="co">    # - last: train on the last EOS token in the conversation</span></span>
-<span id="cb1-171"><a href="#cb1-171" aria-hidden="true" tabindex="-1"></a><span class="co">    # TIP: Please make sure that your `tokenizer.eos_token` is same as EOS/EOT token in template. Otherwise, set `eos_token` under `special_tokens`.</span></span>
-<span id="cb1-172"><a href="#cb1-172" aria-hidden="true" tabindex="-1"></a><span class="at">    </span><span class="fu">train_on_eos</span><span class="kw">:</span><span class="at"> last</span></span>
-<span id="cb1-173"><a href="#cb1-173" aria-hidden="true" tabindex="-1"></a><span class="co">    # The key in the message turn that indicates via boolean whether tokens of a turn should be considered for training. Useful to selectively train on certain turns besides the `roles_to_train`.</span></span>
-<span id="cb1-174"><a href="#cb1-174" aria-hidden="true" tabindex="-1"></a><span class="at">    </span><span class="fu">message_field_training</span><span class="kw">:</span><span class="at"> training</span></span>
-<span id="cb1-175"><a href="#cb1-175" aria-hidden="true" tabindex="-1"></a><span class="co">    # The key in the message turn that contains the training details. Useful to selectively train on certain tokens in a turn.</span></span>
-<span id="cb1-176"><a href="#cb1-176" aria-hidden="true" tabindex="-1"></a><span class="co">    # The value of the key is a List[Dict] containing `begin_offset` (start character index in content), `end_offset` (end character index in content), and `train` (boolean whether to train).</span></span>
-<span id="cb1-177"><a href="#cb1-177" aria-hidden="true" tabindex="-1"></a><span class="at">    </span><span class="fu">message_field_training_detail</span><span class="kw">:</span><span class="at"> train_detail</span></span>
-<span id="cb1-178"><a href="#cb1-178" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb1-179"><a href="#cb1-179" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb1-180"><a href="#cb1-180" aria-hidden="true" tabindex="-1"></a><span class="co"># If false, the datasets will not be shuffled and will keep their original order in `datasets`.</span></span>
-<span id="cb1-181"><a href="#cb1-181" aria-hidden="true" tabindex="-1"></a><span class="co"># The same applies to the `test_datasets` option and the `pretraining_dataset` option. Default is true.</span></span>
-<span id="cb1-182"><a href="#cb1-182" aria-hidden="true" tabindex="-1"></a><span class="fu">shuffle_merged_datasets</span><span class="kw">:</span><span class="at"> </span><span class="ch">true</span></span>
-<span id="cb1-183"><a href="#cb1-183" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb1-184"><a href="#cb1-184" aria-hidden="true" tabindex="-1"></a><span class="at">Deduplicates datasets and test_datasets with identical entries.</span></span>
-<span id="cb1-185"><a href="#cb1-185" aria-hidden="true" tabindex="-1"></a><span class="fu">dataset_exact_deduplication</span><span class="kw">:</span><span class="at"> </span><span class="ch">true</span></span>
-<span id="cb1-186"><a href="#cb1-186" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb1-187"><a href="#cb1-187" aria-hidden="true" tabindex="-1"></a><span class="co"># A list of one or more datasets to eval the model with.</span></span>
-<span id="cb1-188"><a href="#cb1-188" aria-hidden="true" tabindex="-1"></a><span class="co"># You can use either test_datasets, or val_set_size, but not both.</span></span>
-<span id="cb1-189"><a href="#cb1-189" aria-hidden="true" tabindex="-1"></a><span class="fu">test_datasets</span><span class="kw">:</span></span>
-<span id="cb1-190"><a href="#cb1-190" aria-hidden="true" tabindex="-1"></a><span class="at">  </span><span class="kw">-</span><span class="at"> </span><span class="fu">path</span><span class="kw">:</span><span class="at"> /workspace/data/eval.jsonl</span></span>
-<span id="cb1-191"><a href="#cb1-191" aria-hidden="true" tabindex="-1"></a><span class="at">    </span><span class="fu">ds_type</span><span class="kw">:</span><span class="at"> json</span></span>
-<span id="cb1-192"><a href="#cb1-192" aria-hidden="true" tabindex="-1"></a><span class="co">    # You need to specify a split. For "json" datasets the default split is called "train".</span></span>
-<span id="cb1-193"><a href="#cb1-193" aria-hidden="true" tabindex="-1"></a><span class="at">    </span><span class="fu">split</span><span class="kw">:</span><span class="at"> train</span></span>
-<span id="cb1-194"><a href="#cb1-194" aria-hidden="true" tabindex="-1"></a><span class="at">    </span><span class="fu">type</span><span class="kw">:</span><span class="at"> completion</span></span>
-<span id="cb1-195"><a href="#cb1-195" aria-hidden="true" tabindex="-1"></a><span class="at">    </span><span class="fu">data_files</span><span class="kw">:</span></span>
-<span id="cb1-196"><a href="#cb1-196" aria-hidden="true" tabindex="-1"></a><span class="at">      </span><span class="kw">-</span><span class="at"> /workspace/data/eval.jsonl</span></span>
-<span id="cb1-197"><a href="#cb1-197" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb1-198"><a href="#cb1-198" aria-hidden="true" tabindex="-1"></a><span class="co"># use RL training: 'dpo', 'ipo', 'kto'</span></span>
-<span id="cb1-199"><a href="#cb1-199" aria-hidden="true" tabindex="-1"></a><span class="fu">rl</span><span class="kw">:</span></span>
-<span id="cb1-200"><a href="#cb1-200" aria-hidden="true" tabindex="-1"></a><span class="co"># whether to perform weighting if doing DPO training. Boolean.</span></span>
-<span id="cb1-201"><a href="#cb1-201" aria-hidden="true" tabindex="-1"></a><span class="fu">dpo_use_weighting</span><span class="kw">:</span></span>
-<span id="cb1-202"><a href="#cb1-202" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb1-203"><a href="#cb1-203" aria-hidden="true" tabindex="-1"></a><span class="co"># reward modelling: `True` or `False`</span></span>
-<span id="cb1-204"><a href="#cb1-204" aria-hidden="true" tabindex="-1"></a><span class="fu">reward_model</span><span class="kw">:</span></span>
-<span id="cb1-205"><a href="#cb1-205" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb1-206"><a href="#cb1-206" aria-hidden="true" tabindex="-1"></a><span class="co"># process reward modelling: `True` or `False`</span></span>
-<span id="cb1-207"><a href="#cb1-207" aria-hidden="true" tabindex="-1"></a><span class="fu">process_reward_model</span><span class="kw">:</span></span>
+<span id="cb1-160"><a href="#cb1-160" aria-hidden="true" tabindex="-1"></a><span class="co">    # Optional[bool]. Whether to drop the system turn from the dataset. Only works with chat_template.</span></span>
+<span id="cb1-161"><a href="#cb1-161" aria-hidden="true" tabindex="-1"></a><span class="co">    # This does not drop the default system message from chat_template if it exists. If you wish to,</span></span>
+<span id="cb1-162"><a href="#cb1-162" aria-hidden="true" tabindex="-1"></a><span class="co">    # we recommend using a custom jinja template with the default system message removed or</span></span>
+<span id="cb1-163"><a href="#cb1-163" aria-hidden="true" tabindex="-1"></a><span class="co">    # adding a system turn with empty content.</span></span>
+<span id="cb1-164"><a href="#cb1-164" aria-hidden="true" tabindex="-1"></a><span class="at">    </span><span class="fu">drop_system_message</span><span class="kw">:</span></span>
+<span id="cb1-165"><a href="#cb1-165" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb1-166"><a href="#cb1-166" aria-hidden="true" tabindex="-1"></a><span class="co">    # IMPORTANT: The following fields determine which parts of the conversation to train on.</span></span>
+<span id="cb1-167"><a href="#cb1-167" aria-hidden="true" tabindex="-1"></a><span class="co">    # Priority order: message_field_training &gt; message_field_training_detail &gt; train_on_inputs or role in roles_to_train</span></span>
+<span id="cb1-168"><a href="#cb1-168" aria-hidden="true" tabindex="-1"></a><span class="co">    # See examples at `docs/dataset-formats/conversation.qmd`</span></span>
+<span id="cb1-169"><a href="#cb1-169" aria-hidden="true" tabindex="-1"></a><span class="co">    # Note: If the below 4 fields are set to empty, defaults to training only on the last message.</span></span>
+<span id="cb1-170"><a href="#cb1-170" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb1-171"><a href="#cb1-171" aria-hidden="true" tabindex="-1"></a><span class="co">    # Optional[List[str]]. Roles to train on. The tokens from these roles will be considered for the loss.</span></span>
+<span id="cb1-172"><a href="#cb1-172" aria-hidden="true" tabindex="-1"></a><span class="at">    </span><span class="fu">roles_to_train</span><span class="kw">:</span><span class="at"> </span><span class="kw">[</span><span class="st">"assistant"</span><span class="kw">]</span><span class="co">  # default</span></span>
+<span id="cb1-173"><a href="#cb1-173" aria-hidden="true" tabindex="-1"></a><span class="co">    # Optional[str]. Which EOS tokens to train on in the conversation. Possible values are:</span></span>
+<span id="cb1-174"><a href="#cb1-174" aria-hidden="true" tabindex="-1"></a><span class="co">    # - all: train on all EOS tokens</span></span>
+<span id="cb1-175"><a href="#cb1-175" aria-hidden="true" tabindex="-1"></a><span class="co">    # - turn (default): train on the EOS token at the end of each trainable turn</span></span>
+<span id="cb1-176"><a href="#cb1-176" aria-hidden="true" tabindex="-1"></a><span class="co">    # - last: train on the last EOS token in the conversation</span></span>
+<span id="cb1-177"><a href="#cb1-177" aria-hidden="true" tabindex="-1"></a><span class="co">    # TIP: Please make sure that your `tokenizer.eos_token` is same as EOS/EOT token in template. Otherwise, set `eos_token` under `special_tokens`.</span></span>
+<span id="cb1-178"><a href="#cb1-178" aria-hidden="true" tabindex="-1"></a><span class="at">    </span><span class="fu">train_on_eos</span><span class="kw">:</span><span class="at"> last</span></span>
+<span id="cb1-179"><a href="#cb1-179" aria-hidden="true" tabindex="-1"></a><span class="co">    # The key in the message turn that indicates via boolean whether tokens of a turn should be considered for training. Useful to selectively train on certain turns besides the `roles_to_train`.</span></span>
+<span id="cb1-180"><a href="#cb1-180" aria-hidden="true" tabindex="-1"></a><span class="at">    </span><span class="fu">message_field_training</span><span class="kw">:</span><span class="at"> training</span></span>
+<span id="cb1-181"><a href="#cb1-181" aria-hidden="true" tabindex="-1"></a><span class="co">    # The key in the message turn that contains the training details. Useful to selectively train on certain tokens in a turn.</span></span>
+<span id="cb1-182"><a href="#cb1-182" aria-hidden="true" tabindex="-1"></a><span class="co">    # The value of the key is a List[Dict] containing `begin_offset` (start character index in content), `end_offset` (end character index in content), and `train` (boolean whether to train).</span></span>
+<span id="cb1-183"><a href="#cb1-183" aria-hidden="true" tabindex="-1"></a><span class="at">    </span><span class="fu">message_field_training_detail</span><span class="kw">:</span><span class="at"> train_detail</span></span>
+<span id="cb1-184"><a href="#cb1-184" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb1-185"><a href="#cb1-185" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb1-186"><a href="#cb1-186" aria-hidden="true" tabindex="-1"></a><span class="co"># If false, the datasets will not be shuffled and will keep their original order in `datasets`.</span></span>
+<span id="cb1-187"><a href="#cb1-187" aria-hidden="true" tabindex="-1"></a><span class="co"># The same applies to the `test_datasets` option and the `pretraining_dataset` option. Default is true.</span></span>
+<span id="cb1-188"><a href="#cb1-188" aria-hidden="true" tabindex="-1"></a><span class="fu">shuffle_merged_datasets</span><span class="kw">:</span><span class="at"> </span><span class="ch">true</span></span>
+<span id="cb1-189"><a href="#cb1-189" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb1-190"><a href="#cb1-190" aria-hidden="true" tabindex="-1"></a><span class="at">Deduplicates datasets and test_datasets with identical entries.</span></span>
+<span id="cb1-191"><a href="#cb1-191" aria-hidden="true" tabindex="-1"></a><span class="fu">dataset_exact_deduplication</span><span class="kw">:</span><span class="at"> </span><span class="ch">true</span></span>
+<span id="cb1-192"><a href="#cb1-192" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb1-193"><a href="#cb1-193" aria-hidden="true" tabindex="-1"></a><span class="co"># A list of one or more datasets to eval the model with.</span></span>
+<span id="cb1-194"><a href="#cb1-194" aria-hidden="true" tabindex="-1"></a><span class="co"># You can use either test_datasets, or val_set_size, but not both.</span></span>
+<span id="cb1-195"><a href="#cb1-195" aria-hidden="true" tabindex="-1"></a><span class="fu">test_datasets</span><span class="kw">:</span></span>
+<span id="cb1-196"><a href="#cb1-196" aria-hidden="true" tabindex="-1"></a><span class="at">  </span><span class="kw">-</span><span class="at"> </span><span class="fu">path</span><span class="kw">:</span><span class="at"> /workspace/data/eval.jsonl</span></span>
+<span id="cb1-197"><a href="#cb1-197" aria-hidden="true" tabindex="-1"></a><span class="at">    </span><span class="fu">ds_type</span><span class="kw">:</span><span class="at"> json</span></span>
+<span id="cb1-198"><a href="#cb1-198" aria-hidden="true" tabindex="-1"></a><span class="co">    # You need to specify a split. For "json" datasets the default split is called "train".</span></span>
+<span id="cb1-199"><a href="#cb1-199" aria-hidden="true" tabindex="-1"></a><span class="at">    </span><span class="fu">split</span><span class="kw">:</span><span class="at"> train</span></span>
+<span id="cb1-200"><a href="#cb1-200" aria-hidden="true" tabindex="-1"></a><span class="at">    </span><span class="fu">type</span><span class="kw">:</span><span class="at"> completion</span></span>
+<span id="cb1-201"><a href="#cb1-201" aria-hidden="true" tabindex="-1"></a><span class="at">    </span><span class="fu">data_files</span><span class="kw">:</span></span>
+<span id="cb1-202"><a href="#cb1-202" aria-hidden="true" tabindex="-1"></a><span class="at">      </span><span class="kw">-</span><span class="at"> /workspace/data/eval.jsonl</span></span>
+<span id="cb1-203"><a href="#cb1-203" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb1-204"><a href="#cb1-204" aria-hidden="true" tabindex="-1"></a><span class="co"># use RL training: 'dpo', 'ipo', 'kto'</span></span>
+<span id="cb1-205"><a href="#cb1-205" aria-hidden="true" tabindex="-1"></a><span class="fu">rl</span><span class="kw">:</span></span>
+<span id="cb1-206"><a href="#cb1-206" aria-hidden="true" tabindex="-1"></a><span class="co"># whether to perform weighting if doing DPO training. Boolean.</span></span>
+<span id="cb1-207"><a href="#cb1-207" aria-hidden="true" tabindex="-1"></a><span class="fu">dpo_use_weighting</span><span class="kw">:</span></span>
 <span id="cb1-208"><a href="#cb1-208" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb1-209"><a href="#cb1-209" aria-hidden="true" tabindex="-1"></a><span class="co"># The name of the chat template to use for training, following values are supported:</span></span>
-<span id="cb1-210"><a href="#cb1-210" aria-hidden="true" tabindex="-1"></a><span class="co"># - tokenizer_default: Uses the chat template that is available in the tokenizer_config.json. If the chat template is not available in the tokenizer, it will raise an error. This is the default value.</span></span>
-<span id="cb1-211"><a href="#cb1-211" aria-hidden="true" tabindex="-1"></a><span class="co"># - alpaca/inst/chatml/gemma/cohere/llama3/phi_3/deepseek_v2/jamba: These chat templates are available in the axolotl codebase at src/axolotl/utils/chat_templates.py</span></span>
-<span id="cb1-212"><a href="#cb1-212" aria-hidden="true" tabindex="-1"></a><span class="co"># - tokenizer_default_fallback_*: where * is the name of the chat template to fallback to. E.g. tokenizer_default_fallback_chatml. This is useful when the chat template is not available in the tokenizer.</span></span>
-<span id="cb1-213"><a href="#cb1-213" aria-hidden="true" tabindex="-1"></a><span class="co"># - jinja: Uses a custom jinja template for the chat template. The custom jinja template should be provided in the chat_template_jinja field.</span></span>
-<span id="cb1-214"><a href="#cb1-214" aria-hidden="true" tabindex="-1"></a><span class="co"># The selected chat template will be saved to the tokenizer_config.json for easier inferencing</span></span>
-<span id="cb1-215"><a href="#cb1-215" aria-hidden="true" tabindex="-1"></a><span class="co"># Note: It is recommended to set train_on_inputs to true when using a chat template that is different from the model's default chat template.</span></span>
-<span id="cb1-216"><a href="#cb1-216" aria-hidden="true" tabindex="-1"></a><span class="fu">chat_template</span><span class="kw">:</span><span class="at"> tokenizer_default</span></span>
-<span id="cb1-217"><a href="#cb1-217" aria-hidden="true" tabindex="-1"></a><span class="co"># custom jinja template for chat template. This will be only used if chat_template is set to `jinja` or `null` (in which case chat_template is automatically set to `jinja`). Default is null.</span></span>
-<span id="cb1-218"><a href="#cb1-218" aria-hidden="true" tabindex="-1"></a><span class="fu">chat_template_jinja</span><span class="kw">:</span><span class="at"> </span><span class="ch">null</span></span>
-<span id="cb1-219"><a href="#cb1-219" aria-hidden="true" tabindex="-1"></a><span class="co"># Changes the default system message</span></span>
-<span id="cb1-220"><a href="#cb1-220" aria-hidden="true" tabindex="-1"></a><span class="fu">default_system_message</span><span class="kw">:</span><span class="at"> You are a helpful assistant. Please give a long and detailed answer.</span><span class="co"> # Currently only supports chatml.</span></span>
-<span id="cb1-221"><a href="#cb1-221" aria-hidden="true" tabindex="-1"></a><span class="co"># Axolotl attempts to save the dataset as an arrow after packing the data together so</span></span>
-<span id="cb1-222"><a href="#cb1-222" aria-hidden="true" tabindex="-1"></a><span class="co"># subsequent training attempts load faster, relative path</span></span>
-<span id="cb1-223"><a href="#cb1-223" aria-hidden="true" tabindex="-1"></a><span class="fu">dataset_prepared_path</span><span class="kw">:</span><span class="at"> data/last_run_prepared</span></span>
-<span id="cb1-224"><a href="#cb1-224" aria-hidden="true" tabindex="-1"></a><span class="co"># Push prepared dataset to hub</span></span>
-<span id="cb1-225"><a href="#cb1-225" aria-hidden="true" tabindex="-1"></a><span class="fu">push_dataset_to_hub</span><span class="kw">:</span><span class="co"> # repo path</span></span>
-<span id="cb1-226"><a href="#cb1-226" aria-hidden="true" tabindex="-1"></a><span class="co"># The maximum number of processes to use while preprocessing your input dataset. This defaults to `os.cpu_count()`</span></span>
-<span id="cb1-227"><a href="#cb1-227" aria-hidden="true" tabindex="-1"></a><span class="co"># if not set.</span></span>
-<span id="cb1-228"><a href="#cb1-228" aria-hidden="true" tabindex="-1"></a><span class="fu">dataset_processes</span><span class="kw">:</span><span class="co"> # defaults to os.cpu_count() if not set</span></span>
-<span id="cb1-229"><a href="#cb1-229" aria-hidden="true" tabindex="-1"></a><span class="co"># Keep dataset in memory while preprocessing</span></span>
-<span id="cb1-230"><a href="#cb1-230" aria-hidden="true" tabindex="-1"></a><span class="co"># Only needed if cached dataset is taking too much storage</span></span>
-<span id="cb1-231"><a href="#cb1-231" aria-hidden="true" tabindex="-1"></a><span class="fu">dataset_keep_in_memory</span><span class="kw">:</span></span>
-<span id="cb1-232"><a href="#cb1-232" aria-hidden="true" tabindex="-1"></a><span class="co"># push checkpoints to hub</span></span>
-<span id="cb1-233"><a href="#cb1-233" aria-hidden="true" tabindex="-1"></a><span class="fu">hub_model_id</span><span class="kw">:</span><span class="co"> # private repo path to push finetuned model</span></span>
-<span id="cb1-234"><a href="#cb1-234" aria-hidden="true" tabindex="-1"></a><span class="co"># how to push checkpoints to hub</span></span>
-<span id="cb1-235"><a href="#cb1-235" aria-hidden="true" tabindex="-1"></a><span class="co"># https://huggingface.co/docs/transformers/v4.31.0/en/main_classes/trainer#transformers.TrainingArguments.hub_strategy</span></span>
-<span id="cb1-236"><a href="#cb1-236" aria-hidden="true" tabindex="-1"></a><span class="fu">hub_strategy</span><span class="kw">:</span></span>
-<span id="cb1-237"><a href="#cb1-237" aria-hidden="true" tabindex="-1"></a><span class="co"># Whether to use hf `use_auth_token` for loading datasets. Useful for fetching private datasets</span></span>
-<span id="cb1-238"><a href="#cb1-238" aria-hidden="true" tabindex="-1"></a><span class="co"># Required to be true when used in combination with `push_dataset_to_hub`</span></span>
-<span id="cb1-239"><a href="#cb1-239" aria-hidden="true" tabindex="-1"></a><span class="fu">hf_use_auth_token</span><span class="kw">:</span><span class="co"> # boolean</span></span>
-<span id="cb1-240"><a href="#cb1-240" aria-hidden="true" tabindex="-1"></a><span class="co"># How much of the dataset to set aside as evaluation. 1 = 100%, 0.50 = 50%, etc. 0 for no eval.</span></span>
-<span id="cb1-241"><a href="#cb1-241" aria-hidden="true" tabindex="-1"></a><span class="fu">val_set_size</span><span class="kw">:</span><span class="at"> </span><span class="fl">0.04</span></span>
-<span id="cb1-242"><a href="#cb1-242" aria-hidden="true" tabindex="-1"></a><span class="co"># Num shards for whole dataset</span></span>
-<span id="cb1-243"><a href="#cb1-243" aria-hidden="true" tabindex="-1"></a><span class="fu">dataset_shard_num</span><span class="kw">:</span></span>
-<span id="cb1-244"><a href="#cb1-244" aria-hidden="true" tabindex="-1"></a><span class="co"># Index of shard to use for whole dataset</span></span>
-<span id="cb1-245"><a href="#cb1-245" aria-hidden="true" tabindex="-1"></a><span class="fu">dataset_shard_idx</span><span class="kw">:</span></span>
-<span id="cb1-246"><a href="#cb1-246" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb1-247"><a href="#cb1-247" aria-hidden="true" tabindex="-1"></a><span class="co"># The maximum length of an input to train with, this should typically be less than 2048</span></span>
-<span id="cb1-248"><a href="#cb1-248" aria-hidden="true" tabindex="-1"></a><span class="co"># as most models have a token/context limit of 2048</span></span>
-<span id="cb1-249"><a href="#cb1-249" aria-hidden="true" tabindex="-1"></a><span class="fu">sequence_len</span><span class="kw">:</span><span class="at"> </span><span class="dv">2048</span></span>
-<span id="cb1-250"><a href="#cb1-250" aria-hidden="true" tabindex="-1"></a><span class="co"># Pad inputs so each step uses constant sized buffers</span></span>
-<span id="cb1-251"><a href="#cb1-251" aria-hidden="true" tabindex="-1"></a><span class="co"># This will reduce memory fragmentation and may prevent OOMs, by re-using memory more efficiently</span></span>
-<span id="cb1-252"><a href="#cb1-252" aria-hidden="true" tabindex="-1"></a><span class="fu">pad_to_sequence_len</span><span class="kw">:</span></span>
-<span id="cb1-253"><a href="#cb1-253" aria-hidden="true" tabindex="-1"></a><span class="co"># Use efficient multi-packing with block diagonal attention and per sequence position_ids. Recommend set to 'true'</span></span>
-<span id="cb1-254"><a href="#cb1-254" aria-hidden="true" tabindex="-1"></a><span class="fu">sample_packing</span><span class="kw">:</span></span>
-<span id="cb1-255"><a href="#cb1-255" aria-hidden="true" tabindex="-1"></a><span class="co"># Set to 'false' if getting errors during eval with sample_packing on.</span></span>
-<span id="cb1-256"><a href="#cb1-256" aria-hidden="true" tabindex="-1"></a><span class="fu">eval_sample_packing</span><span class="kw">:</span></span>
-<span id="cb1-257"><a href="#cb1-257" aria-hidden="true" tabindex="-1"></a><span class="co"># You can set these packing optimizations AFTER starting a training at least once.</span></span>
-<span id="cb1-258"><a href="#cb1-258" aria-hidden="true" tabindex="-1"></a><span class="co"># The trainer will provide recommended values for these values.</span></span>
-<span id="cb1-259"><a href="#cb1-259" aria-hidden="true" tabindex="-1"></a><span class="fu">sample_packing_eff_est</span><span class="kw">:</span></span>
-<span id="cb1-260"><a href="#cb1-260" aria-hidden="true" tabindex="-1"></a><span class="fu">total_num_tokens</span><span class="kw">:</span></span>
-<span id="cb1-261"><a href="#cb1-261" aria-hidden="true" tabindex="-1"></a><span class="co"># Increasing the following values helps with packing, but usually only slightly (&lt;%1.)</span></span>
-<span id="cb1-262"><a href="#cb1-262" aria-hidden="true" tabindex="-1"></a><span class="co"># The number of samples packed at a time.</span></span>
-<span id="cb1-263"><a href="#cb1-263" aria-hidden="true" tabindex="-1"></a><span class="fu">sample_packing_group_size</span><span class="kw">:</span><span class="at"> </span><span class="dv">100000</span></span>
-<span id="cb1-264"><a href="#cb1-264" aria-hidden="true" tabindex="-1"></a><span class="co"># The number of samples which can be packed into one sequence. Increase if using a large sequence_len with many short samples.</span></span>
-<span id="cb1-265"><a href="#cb1-265" aria-hidden="true" tabindex="-1"></a><span class="fu">sample_packing_bin_size</span><span class="kw">:</span><span class="at"> </span><span class="dv">200</span></span>
-<span id="cb1-266"><a href="#cb1-266" aria-hidden="true" tabindex="-1"></a><span class="co"># whether to concatenate samples during pretraining</span></span>
-<span id="cb1-267"><a href="#cb1-267" aria-hidden="true" tabindex="-1"></a><span class="fu">pretraining_sample_concatenation</span><span class="kw">:</span></span>
-<span id="cb1-268"><a href="#cb1-268" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb1-269"><a href="#cb1-269" aria-hidden="true" tabindex="-1"></a><span class="co"># Use batch flattening for speedups when not using sample_packing</span></span>
-<span id="cb1-270"><a href="#cb1-270" aria-hidden="true" tabindex="-1"></a><span class="fu">batch_flattening</span><span class="kw">:</span></span>
-<span id="cb1-271"><a href="#cb1-271" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb1-272"><a href="#cb1-272" aria-hidden="true" tabindex="-1"></a><span class="co"># Passed through to transformers when loading the model when launched without accelerate</span></span>
-<span id="cb1-273"><a href="#cb1-273" aria-hidden="true" tabindex="-1"></a><span class="co"># Use `sequential` when training w/ model parallelism to limit memory</span></span>
-<span id="cb1-274"><a href="#cb1-274" aria-hidden="true" tabindex="-1"></a><span class="fu">device_map</span><span class="kw">:</span></span>
-<span id="cb1-275"><a href="#cb1-275" aria-hidden="true" tabindex="-1"></a><span class="co"># Defines the max memory usage per gpu on the system. Passed through to transformers when loading the model.</span></span>
-<span id="cb1-276"><a href="#cb1-276" aria-hidden="true" tabindex="-1"></a><span class="fu">max_memory</span><span class="kw">:</span></span>
+<span id="cb1-209"><a href="#cb1-209" aria-hidden="true" tabindex="-1"></a><span class="co"># reward modelling: `True` or `False`</span></span>
+<span id="cb1-210"><a href="#cb1-210" aria-hidden="true" tabindex="-1"></a><span class="fu">reward_model</span><span class="kw">:</span></span>
+<span id="cb1-211"><a href="#cb1-211" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb1-212"><a href="#cb1-212" aria-hidden="true" tabindex="-1"></a><span class="co"># process reward modelling: `True` or `False`</span></span>
+<span id="cb1-213"><a href="#cb1-213" aria-hidden="true" tabindex="-1"></a><span class="fu">process_reward_model</span><span class="kw">:</span></span>
+<span id="cb1-214"><a href="#cb1-214" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb1-215"><a href="#cb1-215" aria-hidden="true" tabindex="-1"></a><span class="co"># The name of the chat template to use for training, following values are supported:</span></span>
+<span id="cb1-216"><a href="#cb1-216" aria-hidden="true" tabindex="-1"></a><span class="co"># - tokenizer_default: Uses the chat template that is available in the tokenizer_config.json. If the chat template is not available in the tokenizer, it will raise an error. This is the default value.</span></span>
+<span id="cb1-217"><a href="#cb1-217" aria-hidden="true" tabindex="-1"></a><span class="co"># - alpaca/inst/chatml/gemma/cohere/llama3/phi_3/deepseek_v2/jamba: These chat templates are available in the axolotl codebase at src/axolotl/utils/chat_templates.py</span></span>
+<span id="cb1-218"><a href="#cb1-218" aria-hidden="true" tabindex="-1"></a><span class="co"># - tokenizer_default_fallback_*: where * is the name of the chat template to fallback to. E.g. tokenizer_default_fallback_chatml. This is useful when the chat template is not available in the tokenizer.</span></span>
+<span id="cb1-219"><a href="#cb1-219" aria-hidden="true" tabindex="-1"></a><span class="co"># - jinja: Uses a custom jinja template for the chat template. The custom jinja template should be provided in the chat_template_jinja field.</span></span>
+<span id="cb1-220"><a href="#cb1-220" aria-hidden="true" tabindex="-1"></a><span class="co"># The selected chat template will be saved to the tokenizer_config.json for easier inferencing</span></span>
+<span id="cb1-221"><a href="#cb1-221" aria-hidden="true" tabindex="-1"></a><span class="co"># Note: It is recommended to set train_on_inputs to true when using a chat template that is different from the model's default chat template.</span></span>
+<span id="cb1-222"><a href="#cb1-222" aria-hidden="true" tabindex="-1"></a><span class="fu">chat_template</span><span class="kw">:</span><span class="at"> tokenizer_default</span></span>
+<span id="cb1-223"><a href="#cb1-223" aria-hidden="true" tabindex="-1"></a><span class="co"># custom jinja template for chat template. This will be only used if chat_template is set to `jinja` or `null` (in which case chat_template is automatically set to `jinja`). Default is null.</span></span>
+<span id="cb1-224"><a href="#cb1-224" aria-hidden="true" tabindex="-1"></a><span class="fu">chat_template_jinja</span><span class="kw">:</span><span class="at"> </span><span class="ch">null</span></span>
+<span id="cb1-225"><a href="#cb1-225" aria-hidden="true" tabindex="-1"></a><span class="co"># Changes the default system message. Currently only supports chatml.</span></span>
+<span id="cb1-226"><a href="#cb1-226" aria-hidden="true" tabindex="-1"></a><span class="fu">default_system_message</span><span class="kw">:</span><span class="at"> You are a helpful assistant. Please give a long and detailed answer.</span></span>
+<span id="cb1-227"><a href="#cb1-227" aria-hidden="true" tabindex="-1"></a><span class="co"># Axolotl attempts to save the dataset as an arrow after packing the data together so</span></span>
+<span id="cb1-228"><a href="#cb1-228" aria-hidden="true" tabindex="-1"></a><span class="co"># subsequent training attempts load faster, relative path</span></span>
+<span id="cb1-229"><a href="#cb1-229" aria-hidden="true" tabindex="-1"></a><span class="fu">dataset_prepared_path</span><span class="kw">:</span><span class="at"> data/last_run_prepared</span></span>
+<span id="cb1-230"><a href="#cb1-230" aria-hidden="true" tabindex="-1"></a><span class="co"># Push prepared dataset to hub</span></span>
+<span id="cb1-231"><a href="#cb1-231" aria-hidden="true" tabindex="-1"></a><span class="fu">push_dataset_to_hub</span><span class="kw">:</span><span class="co"> # repo path</span></span>
+<span id="cb1-232"><a href="#cb1-232" aria-hidden="true" tabindex="-1"></a><span class="co"># The maximum number of processes to use while preprocessing your input dataset. This defaults to `os.cpu_count()`</span></span>
+<span id="cb1-233"><a href="#cb1-233" aria-hidden="true" tabindex="-1"></a><span class="co"># if not set.</span></span>
+<span id="cb1-234"><a href="#cb1-234" aria-hidden="true" tabindex="-1"></a><span class="fu">dataset_processes</span><span class="kw">:</span><span class="co"> # defaults to os.cpu_count() if not set</span></span>
+<span id="cb1-235"><a href="#cb1-235" aria-hidden="true" tabindex="-1"></a><span class="co"># Keep dataset in memory while preprocessing</span></span>
+<span id="cb1-236"><a href="#cb1-236" aria-hidden="true" tabindex="-1"></a><span class="co"># Only needed if cached dataset is taking too much storage</span></span>
+<span id="cb1-237"><a href="#cb1-237" aria-hidden="true" tabindex="-1"></a><span class="fu">dataset_keep_in_memory</span><span class="kw">:</span></span>
+<span id="cb1-238"><a href="#cb1-238" aria-hidden="true" tabindex="-1"></a><span class="co"># push checkpoints to hub</span></span>
+<span id="cb1-239"><a href="#cb1-239" aria-hidden="true" tabindex="-1"></a><span class="fu">hub_model_id</span><span class="kw">:</span><span class="co"> # private repo path to push finetuned model</span></span>
+<span id="cb1-240"><a href="#cb1-240" aria-hidden="true" tabindex="-1"></a><span class="co"># how to push checkpoints to hub</span></span>
+<span id="cb1-241"><a href="#cb1-241" aria-hidden="true" tabindex="-1"></a><span class="co"># https://huggingface.co/docs/transformers/v4.31.0/en/main_classes/trainer#transformers.TrainingArguments.hub_strategy</span></span>
+<span id="cb1-242"><a href="#cb1-242" aria-hidden="true" tabindex="-1"></a><span class="fu">hub_strategy</span><span class="kw">:</span></span>
+<span id="cb1-243"><a href="#cb1-243" aria-hidden="true" tabindex="-1"></a><span class="co"># Whether to use hf `use_auth_token` for loading datasets. Useful for fetching private datasets</span></span>
+<span id="cb1-244"><a href="#cb1-244" aria-hidden="true" tabindex="-1"></a><span class="co"># Required to be true when used in combination with `push_dataset_to_hub`</span></span>
+<span id="cb1-245"><a href="#cb1-245" aria-hidden="true" tabindex="-1"></a><span class="fu">hf_use_auth_token</span><span class="kw">:</span><span class="co"> # boolean</span></span>
+<span id="cb1-246"><a href="#cb1-246" aria-hidden="true" tabindex="-1"></a><span class="co"># How much of the dataset to set aside as evaluation. 1 = 100%, 0.50 = 50%, etc. 0 for no eval.</span></span>
+<span id="cb1-247"><a href="#cb1-247" aria-hidden="true" tabindex="-1"></a><span class="fu">val_set_size</span><span class="kw">:</span><span class="at"> </span><span class="fl">0.04</span></span>
+<span id="cb1-248"><a href="#cb1-248" aria-hidden="true" tabindex="-1"></a><span class="co"># Num shards for whole dataset</span></span>
+<span id="cb1-249"><a href="#cb1-249" aria-hidden="true" tabindex="-1"></a><span class="fu">dataset_shard_num</span><span class="kw">:</span></span>
+<span id="cb1-250"><a href="#cb1-250" aria-hidden="true" tabindex="-1"></a><span class="co"># Index of shard to use for whole dataset</span></span>
+<span id="cb1-251"><a href="#cb1-251" aria-hidden="true" tabindex="-1"></a><span class="fu">dataset_shard_idx</span><span class="kw">:</span></span>
+<span id="cb1-252"><a href="#cb1-252" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb1-253"><a href="#cb1-253" aria-hidden="true" tabindex="-1"></a><span class="co"># The maximum length of an input to train with, this should typically be less than 2048</span></span>
+<span id="cb1-254"><a href="#cb1-254" aria-hidden="true" tabindex="-1"></a><span class="co"># as most models have a token/context limit of 2048</span></span>
+<span id="cb1-255"><a href="#cb1-255" aria-hidden="true" tabindex="-1"></a><span class="fu">sequence_len</span><span class="kw">:</span><span class="at"> </span><span class="dv">2048</span></span>
+<span id="cb1-256"><a href="#cb1-256" aria-hidden="true" tabindex="-1"></a><span class="co"># Pad inputs so each step uses constant sized buffers</span></span>
+<span id="cb1-257"><a href="#cb1-257" aria-hidden="true" tabindex="-1"></a><span class="co"># This will reduce memory fragmentation and may prevent OOMs, by re-using memory more efficiently</span></span>
+<span id="cb1-258"><a href="#cb1-258" aria-hidden="true" tabindex="-1"></a><span class="fu">pad_to_sequence_len</span><span class="kw">:</span></span>
+<span id="cb1-259"><a href="#cb1-259" aria-hidden="true" tabindex="-1"></a><span class="co"># Use efficient multi-packing with block diagonal attention and per sequence position_ids. Recommend set to 'true'</span></span>
+<span id="cb1-260"><a href="#cb1-260" aria-hidden="true" tabindex="-1"></a><span class="fu">sample_packing</span><span class="kw">:</span></span>
+<span id="cb1-261"><a href="#cb1-261" aria-hidden="true" tabindex="-1"></a><span class="co"># Set to 'false' if getting errors during eval with sample_packing on.</span></span>
+<span id="cb1-262"><a href="#cb1-262" aria-hidden="true" tabindex="-1"></a><span class="fu">eval_sample_packing</span><span class="kw">:</span></span>
+<span id="cb1-263"><a href="#cb1-263" aria-hidden="true" tabindex="-1"></a><span class="co"># You can set these packing optimizations AFTER starting a training at least once.</span></span>
+<span id="cb1-264"><a href="#cb1-264" aria-hidden="true" tabindex="-1"></a><span class="co"># The trainer will provide recommended values for these values.</span></span>
+<span id="cb1-265"><a href="#cb1-265" aria-hidden="true" tabindex="-1"></a><span class="fu">sample_packing_eff_est</span><span class="kw">:</span></span>
+<span id="cb1-266"><a href="#cb1-266" aria-hidden="true" tabindex="-1"></a><span class="fu">total_num_tokens</span><span class="kw">:</span></span>
+<span id="cb1-267"><a href="#cb1-267" aria-hidden="true" tabindex="-1"></a><span class="co"># Increasing the following values helps with packing, but usually only slightly (&lt;%1.)</span></span>
+<span id="cb1-268"><a href="#cb1-268" aria-hidden="true" tabindex="-1"></a><span class="co"># The number of samples packed at a time.</span></span>
+<span id="cb1-269"><a href="#cb1-269" aria-hidden="true" tabindex="-1"></a><span class="fu">sample_packing_group_size</span><span class="kw">:</span><span class="at"> </span><span class="dv">100000</span></span>
+<span id="cb1-270"><a href="#cb1-270" aria-hidden="true" tabindex="-1"></a><span class="co"># The number of samples which can be packed into one sequence. Increase if using a large sequence_len with many short samples.</span></span>
+<span id="cb1-271"><a href="#cb1-271" aria-hidden="true" tabindex="-1"></a><span class="fu">sample_packing_bin_size</span><span class="kw">:</span><span class="at"> </span><span class="dv">200</span></span>
+<span id="cb1-272"><a href="#cb1-272" aria-hidden="true" tabindex="-1"></a><span class="co"># whether to concatenate samples during pretraining</span></span>
+<span id="cb1-273"><a href="#cb1-273" aria-hidden="true" tabindex="-1"></a><span class="fu">pretraining_sample_concatenation</span><span class="kw">:</span></span>
+<span id="cb1-274"><a href="#cb1-274" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb1-275"><a href="#cb1-275" aria-hidden="true" tabindex="-1"></a><span class="co"># Use batch flattening for speedups when not using sample_packing</span></span>
+<span id="cb1-276"><a href="#cb1-276" aria-hidden="true" tabindex="-1"></a><span class="fu">batch_flattening</span><span class="kw">:</span></span>
 <span id="cb1-277"><a href="#cb1-277" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb1-278"><a href="#cb1-278" aria-hidden="true" tabindex="-1"></a><span class="co"># If you want to use 'lora' or 'qlora' or leave blank to train all parameters in original model</span></span>
-<span id="cb1-279"><a href="#cb1-279" aria-hidden="true" tabindex="-1"></a><span class="fu">adapter</span><span class="kw">:</span><span class="at"> lora</span></span>
-<span id="cb1-280"><a href="#cb1-280" aria-hidden="true" tabindex="-1"></a><span class="co"># If you already have a lora model trained that you want to load, put that here.</span></span>
-<span id="cb1-281"><a href="#cb1-281" aria-hidden="true" tabindex="-1"></a><span class="co"># This means after training, if you want to test the model, you should set this to the value of `output_dir`.</span></span>
-<span id="cb1-282"><a href="#cb1-282" aria-hidden="true" tabindex="-1"></a><span class="co"># Note that if you merge an adapter to the base model, a new subdirectory `merged` will be created under the `output_dir`.</span></span>
-<span id="cb1-283"><a href="#cb1-283" aria-hidden="true" tabindex="-1"></a><span class="fu">lora_model_dir</span><span class="kw">:</span></span>
-<span id="cb1-284"><a href="#cb1-284" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb1-285"><a href="#cb1-285" aria-hidden="true" tabindex="-1"></a><span class="co"># LoRA hyperparameters</span></span>
-<span id="cb1-286"><a href="#cb1-286" aria-hidden="true" tabindex="-1"></a><span class="co"># For more details about the following options, see:</span></span>
-<span id="cb1-287"><a href="#cb1-287" aria-hidden="true" tabindex="-1"></a><span class="co"># https://www.anyscale.com/blog/fine-tuning-llms-lora-or-full-parameter-an-in-depth-analysis-with-llama-2</span></span>
-<span id="cb1-288"><a href="#cb1-288" aria-hidden="true" tabindex="-1"></a><span class="fu">lora_r</span><span class="kw">:</span><span class="at"> </span><span class="dv">8</span></span>
-<span id="cb1-289"><a href="#cb1-289" aria-hidden="true" tabindex="-1"></a><span class="fu">lora_alpha</span><span class="kw">:</span><span class="at"> </span><span class="dv">16</span></span>
-<span id="cb1-290"><a href="#cb1-290" aria-hidden="true" tabindex="-1"></a><span class="fu">lora_dropout</span><span class="kw">:</span><span class="at"> </span><span class="fl">0.05</span></span>
-<span id="cb1-291"><a href="#cb1-291" aria-hidden="true" tabindex="-1"></a><span class="fu">lora_target_modules</span><span class="kw">:</span></span>
-<span id="cb1-292"><a href="#cb1-292" aria-hidden="true" tabindex="-1"></a><span class="at">  </span><span class="kw">-</span><span class="at"> q_proj</span></span>
-<span id="cb1-293"><a href="#cb1-293" aria-hidden="true" tabindex="-1"></a><span class="at">  </span><span class="kw">-</span><span class="at"> v_proj</span></span>
-<span id="cb1-294"><a href="#cb1-294" aria-hidden="true" tabindex="-1"></a><span class="co">#  - k_proj</span></span>
-<span id="cb1-295"><a href="#cb1-295" aria-hidden="true" tabindex="-1"></a><span class="co">#  - o_proj</span></span>
-<span id="cb1-296"><a href="#cb1-296" aria-hidden="true" tabindex="-1"></a><span class="co">#  - gate_proj</span></span>
-<span id="cb1-297"><a href="#cb1-297" aria-hidden="true" tabindex="-1"></a><span class="co">#  - down_proj</span></span>
-<span id="cb1-298"><a href="#cb1-298" aria-hidden="true" tabindex="-1"></a><span class="co">#  - up_proj</span></span>
-<span id="cb1-299"><a href="#cb1-299" aria-hidden="true" tabindex="-1"></a><span class="fu">lora_target_linear</span><span class="kw">:</span><span class="co"> # If true, will target all linear modules</span></span>
-<span id="cb1-300"><a href="#cb1-300" aria-hidden="true" tabindex="-1"></a><span class="fu">peft_layers_to_transform</span><span class="kw">:</span><span class="co"> # The layer indices to transform, otherwise, apply to all layers</span></span>
-<span id="cb1-301"><a href="#cb1-301" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb1-302"><a href="#cb1-302" aria-hidden="true" tabindex="-1"></a><span class="co"># If you added new tokens to the tokenizer, you may need to save some LoRA modules because they need to know the new tokens.</span></span>
-<span id="cb1-303"><a href="#cb1-303" aria-hidden="true" tabindex="-1"></a><span class="co"># For LLaMA and Mistral, you need to save `embed_tokens` and `lm_head`. It may vary for other models.</span></span>
-<span id="cb1-304"><a href="#cb1-304" aria-hidden="true" tabindex="-1"></a><span class="co"># `embed_tokens` converts tokens to embeddings, and `lm_head` converts embeddings to token probabilities.</span></span>
-<span id="cb1-305"><a href="#cb1-305" aria-hidden="true" tabindex="-1"></a><span class="co"># https://github.com/huggingface/peft/issues/334#issuecomment-1561727994</span></span>
-<span id="cb1-306"><a href="#cb1-306" aria-hidden="true" tabindex="-1"></a><span class="fu">lora_modules_to_save</span><span class="kw">:</span></span>
-<span id="cb1-307"><a href="#cb1-307" aria-hidden="true" tabindex="-1"></a><span class="co">#  - embed_tokens</span></span>
-<span id="cb1-308"><a href="#cb1-308" aria-hidden="true" tabindex="-1"></a><span class="co">#  - lm_head</span></span>
-<span id="cb1-309"><a href="#cb1-309" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb1-310"><a href="#cb1-310" aria-hidden="true" tabindex="-1"></a><span class="fu">lora_fan_in_fan_out</span><span class="kw">:</span><span class="at"> </span><span class="ch">false</span></span>
-<span id="cb1-311"><a href="#cb1-311" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb1-312"><a href="#cb1-312" aria-hidden="true" tabindex="-1"></a><span class="co"># Apply custom LoRA autograd functions and activation function Triton kernels for</span></span>
-<span id="cb1-313"><a href="#cb1-313" aria-hidden="true" tabindex="-1"></a><span class="co"># speed and memory savings</span></span>
-<span id="cb1-314"><a href="#cb1-314" aria-hidden="true" tabindex="-1"></a><span class="co"># See: https://axolotl-ai-cloud.github.io/axolotl/docs/lora_optims.html</span></span>
-<span id="cb1-315"><a href="#cb1-315" aria-hidden="true" tabindex="-1"></a><span class="fu">lora_mlp_kernel</span><span class="kw">:</span><span class="at"> </span><span class="ch">true</span></span>
-<span id="cb1-316"><a href="#cb1-316" aria-hidden="true" tabindex="-1"></a><span class="fu">lora_qkv_kernel</span><span class="kw">:</span><span class="at"> </span><span class="ch">true</span></span>
-<span id="cb1-317"><a href="#cb1-317" aria-hidden="true" tabindex="-1"></a><span class="fu">lora_o_kernel</span><span class="kw">:</span><span class="at"> </span><span class="ch">true</span></span>
-<span id="cb1-318"><a href="#cb1-318" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb1-319"><a href="#cb1-319" aria-hidden="true" tabindex="-1"></a><span class="co"># LoRA+ hyperparameters</span></span>
-<span id="cb1-320"><a href="#cb1-320" aria-hidden="true" tabindex="-1"></a><span class="co"># For more details about the following options, see:</span></span>
-<span id="cb1-321"><a href="#cb1-321" aria-hidden="true" tabindex="-1"></a><span class="co"># https://arxiv.org/abs/2402.12354  and `src/axolotl/core/train_builder.py`</span></span>
-<span id="cb1-322"><a href="#cb1-322" aria-hidden="true" tabindex="-1"></a><span class="fu">loraplus_lr_ratio</span><span class="kw">:</span><span class="co"> # loraplus learning rate ratio lr_B / lr_A. Recommended value is 2^4.</span></span>
-<span id="cb1-323"><a href="#cb1-323" aria-hidden="true" tabindex="-1"></a><span class="fu">loraplus_lr_embedding</span><span class="kw">:</span><span class="co"> #  loraplus learning rate for lora embedding layers. Default value is 1e-6.</span></span>
+<span id="cb1-278"><a href="#cb1-278" aria-hidden="true" tabindex="-1"></a><span class="co"># Passed through to transformers when loading the model when launched without accelerate</span></span>
+<span id="cb1-279"><a href="#cb1-279" aria-hidden="true" tabindex="-1"></a><span class="co"># Use `sequential` when training w/ model parallelism to limit memory</span></span>
+<span id="cb1-280"><a href="#cb1-280" aria-hidden="true" tabindex="-1"></a><span class="fu">device_map</span><span class="kw">:</span></span>
+<span id="cb1-281"><a href="#cb1-281" aria-hidden="true" tabindex="-1"></a><span class="co"># Defines the max memory usage per gpu on the system. Passed through to transformers when loading the model.</span></span>
+<span id="cb1-282"><a href="#cb1-282" aria-hidden="true" tabindex="-1"></a><span class="fu">max_memory</span><span class="kw">:</span></span>
+<span id="cb1-283"><a href="#cb1-283" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb1-284"><a href="#cb1-284" aria-hidden="true" tabindex="-1"></a><span class="co"># If you want to use 'lora' or 'qlora' or leave blank to train all parameters in original model</span></span>
+<span id="cb1-285"><a href="#cb1-285" aria-hidden="true" tabindex="-1"></a><span class="fu">adapter</span><span class="kw">:</span><span class="at"> lora</span></span>
+<span id="cb1-286"><a href="#cb1-286" aria-hidden="true" tabindex="-1"></a><span class="co"># If you already have a lora model trained that you want to load, put that here.</span></span>
+<span id="cb1-287"><a href="#cb1-287" aria-hidden="true" tabindex="-1"></a><span class="co"># This means after training, if you want to test the model, you should set this to the value of `output_dir`.</span></span>
+<span id="cb1-288"><a href="#cb1-288" aria-hidden="true" tabindex="-1"></a><span class="co"># Note that if you merge an adapter to the base model, a new subdirectory `merged` will be created under the `output_dir`.</span></span>
+<span id="cb1-289"><a href="#cb1-289" aria-hidden="true" tabindex="-1"></a><span class="fu">lora_model_dir</span><span class="kw">:</span></span>
+<span id="cb1-290"><a href="#cb1-290" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb1-291"><a href="#cb1-291" aria-hidden="true" tabindex="-1"></a><span class="co"># LoRA hyperparameters</span></span>
+<span id="cb1-292"><a href="#cb1-292" aria-hidden="true" tabindex="-1"></a><span class="co"># For more details about the following options, see:</span></span>
+<span id="cb1-293"><a href="#cb1-293" aria-hidden="true" tabindex="-1"></a><span class="co"># https://www.anyscale.com/blog/fine-tuning-llms-lora-or-full-parameter-an-in-depth-analysis-with-llama-2</span></span>
+<span id="cb1-294"><a href="#cb1-294" aria-hidden="true" tabindex="-1"></a><span class="fu">lora_r</span><span class="kw">:</span><span class="at"> </span><span class="dv">8</span></span>
+<span id="cb1-295"><a href="#cb1-295" aria-hidden="true" tabindex="-1"></a><span class="fu">lora_alpha</span><span class="kw">:</span><span class="at"> </span><span class="dv">16</span></span>
+<span id="cb1-296"><a href="#cb1-296" aria-hidden="true" tabindex="-1"></a><span class="fu">lora_dropout</span><span class="kw">:</span><span class="at"> </span><span class="fl">0.05</span></span>
+<span id="cb1-297"><a href="#cb1-297" aria-hidden="true" tabindex="-1"></a><span class="fu">lora_target_modules</span><span class="kw">:</span></span>
+<span id="cb1-298"><a href="#cb1-298" aria-hidden="true" tabindex="-1"></a><span class="at">  </span><span class="kw">-</span><span class="at"> q_proj</span></span>
+<span id="cb1-299"><a href="#cb1-299" aria-hidden="true" tabindex="-1"></a><span class="at">  </span><span class="kw">-</span><span class="at"> v_proj</span></span>
+<span id="cb1-300"><a href="#cb1-300" aria-hidden="true" tabindex="-1"></a><span class="co">#  - k_proj</span></span>
+<span id="cb1-301"><a href="#cb1-301" aria-hidden="true" tabindex="-1"></a><span class="co">#  - o_proj</span></span>
+<span id="cb1-302"><a href="#cb1-302" aria-hidden="true" tabindex="-1"></a><span class="co">#  - gate_proj</span></span>
+<span id="cb1-303"><a href="#cb1-303" aria-hidden="true" tabindex="-1"></a><span class="co">#  - down_proj</span></span>
+<span id="cb1-304"><a href="#cb1-304" aria-hidden="true" tabindex="-1"></a><span class="co">#  - up_proj</span></span>
+<span id="cb1-305"><a href="#cb1-305" aria-hidden="true" tabindex="-1"></a><span class="fu">lora_target_linear</span><span class="kw">:</span><span class="co"> # If true, will target all linear modules</span></span>
+<span id="cb1-306"><a href="#cb1-306" aria-hidden="true" tabindex="-1"></a><span class="fu">peft_layers_to_transform</span><span class="kw">:</span><span class="co"> # The layer indices to transform, otherwise, apply to all layers</span></span>
+<span id="cb1-307"><a href="#cb1-307" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb1-308"><a href="#cb1-308" aria-hidden="true" tabindex="-1"></a><span class="co"># If you added new tokens to the tokenizer, you may need to save some LoRA modules because they need to know the new tokens.</span></span>
+<span id="cb1-309"><a href="#cb1-309" aria-hidden="true" tabindex="-1"></a><span class="co"># For LLaMA and Mistral, you need to save `embed_tokens` and `lm_head`. It may vary for other models.</span></span>
+<span id="cb1-310"><a href="#cb1-310" aria-hidden="true" tabindex="-1"></a><span class="co"># `embed_tokens` converts tokens to embeddings, and `lm_head` converts embeddings to token probabilities.</span></span>
+<span id="cb1-311"><a href="#cb1-311" aria-hidden="true" tabindex="-1"></a><span class="co"># https://github.com/huggingface/peft/issues/334#issuecomment-1561727994</span></span>
+<span id="cb1-312"><a href="#cb1-312" aria-hidden="true" tabindex="-1"></a><span class="fu">lora_modules_to_save</span><span class="kw">:</span></span>
+<span id="cb1-313"><a href="#cb1-313" aria-hidden="true" tabindex="-1"></a><span class="co">#  - embed_tokens</span></span>
+<span id="cb1-314"><a href="#cb1-314" aria-hidden="true" tabindex="-1"></a><span class="co">#  - lm_head</span></span>
+<span id="cb1-315"><a href="#cb1-315" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb1-316"><a href="#cb1-316" aria-hidden="true" tabindex="-1"></a><span class="fu">lora_fan_in_fan_out</span><span class="kw">:</span><span class="at"> </span><span class="ch">false</span></span>
+<span id="cb1-317"><a href="#cb1-317" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb1-318"><a href="#cb1-318" aria-hidden="true" tabindex="-1"></a><span class="co"># Apply custom LoRA autograd functions and activation function Triton kernels for</span></span>
+<span id="cb1-319"><a href="#cb1-319" aria-hidden="true" tabindex="-1"></a><span class="co"># speed and memory savings</span></span>
+<span id="cb1-320"><a href="#cb1-320" aria-hidden="true" tabindex="-1"></a><span class="co"># See: https://axolotl-ai-cloud.github.io/axolotl/docs/lora_optims.html</span></span>
+<span id="cb1-321"><a href="#cb1-321" aria-hidden="true" tabindex="-1"></a><span class="fu">lora_mlp_kernel</span><span class="kw">:</span><span class="at"> </span><span class="ch">true</span></span>
+<span id="cb1-322"><a href="#cb1-322" aria-hidden="true" tabindex="-1"></a><span class="fu">lora_qkv_kernel</span><span class="kw">:</span><span class="at"> </span><span class="ch">true</span></span>
+<span id="cb1-323"><a href="#cb1-323" aria-hidden="true" tabindex="-1"></a><span class="fu">lora_o_kernel</span><span class="kw">:</span><span class="at"> </span><span class="ch">true</span></span>
 <span id="cb1-324"><a href="#cb1-324" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb1-325"><a href="#cb1-325" aria-hidden="true" tabindex="-1"></a><span class="fu">peft</span><span class="kw">:</span></span>
-<span id="cb1-326"><a href="#cb1-326" aria-hidden="true" tabindex="-1"></a><span class="co">  # Configuration options for loftq initialization for LoRA</span></span>
-<span id="cb1-327"><a href="#cb1-327" aria-hidden="true" tabindex="-1"></a><span class="co">  # https://huggingface.co/docs/peft/developer_guides/quantization#loftq-initialization</span></span>
-<span id="cb1-328"><a href="#cb1-328" aria-hidden="true" tabindex="-1"></a><span class="at">  </span><span class="fu">loftq_config</span><span class="kw">:</span></span>
-<span id="cb1-329"><a href="#cb1-329" aria-hidden="true" tabindex="-1"></a><span class="at">    </span><span class="fu">loftq_bits</span><span class="kw">:</span><span class="co">  # typically 4 bits</span></span>
+<span id="cb1-325"><a href="#cb1-325" aria-hidden="true" tabindex="-1"></a><span class="co"># LoRA+ hyperparameters</span></span>
+<span id="cb1-326"><a href="#cb1-326" aria-hidden="true" tabindex="-1"></a><span class="co"># For more details about the following options, see:</span></span>
+<span id="cb1-327"><a href="#cb1-327" aria-hidden="true" tabindex="-1"></a><span class="co"># https://arxiv.org/abs/2402.12354  and `src/axolotl/core/train_builder.py`</span></span>
+<span id="cb1-328"><a href="#cb1-328" aria-hidden="true" tabindex="-1"></a><span class="fu">loraplus_lr_ratio</span><span class="kw">:</span><span class="co"> # loraplus learning rate ratio lr_B / lr_A. Recommended value is 2^4.</span></span>
+<span id="cb1-329"><a href="#cb1-329" aria-hidden="true" tabindex="-1"></a><span class="fu">loraplus_lr_embedding</span><span class="kw">:</span><span class="co"> #  loraplus learning rate for lora embedding layers. Default value is 1e-6.</span></span>
 <span id="cb1-330"><a href="#cb1-330" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb1-331"><a href="#cb1-331" aria-hidden="true" tabindex="-1"></a><span class="co"># ReLoRA configuration</span></span>
-<span id="cb1-332"><a href="#cb1-332" aria-hidden="true" tabindex="-1"></a><span class="co"># Must use either 'lora' or 'qlora' adapter, and does not support fsdp or deepspeed</span></span>
-<span id="cb1-333"><a href="#cb1-333" aria-hidden="true" tabindex="-1"></a><span class="fu">relora_steps</span><span class="kw">:</span><span class="co"> # Number of steps per ReLoRA restart</span></span>
-<span id="cb1-334"><a href="#cb1-334" aria-hidden="true" tabindex="-1"></a><span class="fu">relora_warmup_steps</span><span class="kw">:</span><span class="co"> # Number of per-restart warmup steps</span></span>
-<span id="cb1-335"><a href="#cb1-335" aria-hidden="true" tabindex="-1"></a><span class="fu">relora_anneal_steps</span><span class="kw">:</span><span class="co"> # Number of anneal steps for each relora cycle</span></span>
-<span id="cb1-336"><a href="#cb1-336" aria-hidden="true" tabindex="-1"></a><span class="fu">relora_prune_ratio</span><span class="kw">:</span><span class="co"> # threshold for optimizer magnitude when pruning</span></span>
-<span id="cb1-337"><a href="#cb1-337" aria-hidden="true" tabindex="-1"></a><span class="fu">relora_cpu_offload</span><span class="kw">:</span><span class="co"> # True to perform lora weight merges on cpu during restarts, for modest gpu memory savings</span></span>
-<span id="cb1-338"><a href="#cb1-338" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb1-339"><a href="#cb1-339" aria-hidden="true" tabindex="-1"></a><span class="co"># wandb configuration if you're using it</span></span>
-<span id="cb1-340"><a href="#cb1-340" aria-hidden="true" tabindex="-1"></a><span class="co"># Make sure your `WANDB_API_KEY` environment variable is set (recommended) or you login to wandb with `wandb login`.</span></span>
-<span id="cb1-341"><a href="#cb1-341" aria-hidden="true" tabindex="-1"></a><span class="fu">wandb_mode</span><span class="kw">:</span><span class="co"> # "offline" to save run metadata locally and not sync to the server, "disabled" to turn off wandb</span></span>
-<span id="cb1-342"><a href="#cb1-342" aria-hidden="true" tabindex="-1"></a><span class="fu">wandb_project</span><span class="kw">:</span><span class="co"> # Your wandb project name</span></span>
-<span id="cb1-343"><a href="#cb1-343" aria-hidden="true" tabindex="-1"></a><span class="fu">wandb_entity</span><span class="kw">:</span><span class="co"> # A wandb Team name if using a Team</span></span>
-<span id="cb1-344"><a href="#cb1-344" aria-hidden="true" tabindex="-1"></a><span class="fu">wandb_watch</span><span class="kw">:</span></span>
-<span id="cb1-345"><a href="#cb1-345" aria-hidden="true" tabindex="-1"></a><span class="fu">wandb_name</span><span class="kw">:</span><span class="co"> # Set the name of your wandb run</span></span>
-<span id="cb1-346"><a href="#cb1-346" aria-hidden="true" tabindex="-1"></a><span class="fu">wandb_run_id</span><span class="kw">:</span><span class="co"> # Set the ID of your wandb run</span></span>
-<span id="cb1-347"><a href="#cb1-347" aria-hidden="true" tabindex="-1"></a><span class="fu">wandb_log_model</span><span class="kw">:</span><span class="co"> # "checkpoint" to log model to wandb Artifacts every `save_steps` or "end" to log only at the end of training</span></span>
-<span id="cb1-348"><a href="#cb1-348" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb1-349"><a href="#cb1-349" aria-hidden="true" tabindex="-1"></a><span class="co"># mlflow configuration if you're using it</span></span>
-<span id="cb1-350"><a href="#cb1-350" aria-hidden="true" tabindex="-1"></a><span class="fu">mlflow_tracking_uri</span><span class="kw">:</span><span class="co"> # URI to mlflow</span></span>
-<span id="cb1-351"><a href="#cb1-351" aria-hidden="true" tabindex="-1"></a><span class="fu">mlflow_experiment_name</span><span class="kw">:</span><span class="co"> # Your experiment name</span></span>
-<span id="cb1-352"><a href="#cb1-352" aria-hidden="true" tabindex="-1"></a><span class="fu">mlflow_run_name</span><span class="kw">:</span><span class="co"> # Your run name</span></span>
-<span id="cb1-353"><a href="#cb1-353" aria-hidden="true" tabindex="-1"></a><span class="fu">hf_mlflow_log_artifacts</span><span class="kw">:</span><span class="co">  # set to true to copy each saved checkpoint on each save to mlflow artifact registry</span></span>
+<span id="cb1-331"><a href="#cb1-331" aria-hidden="true" tabindex="-1"></a><span class="fu">peft</span><span class="kw">:</span></span>
+<span id="cb1-332"><a href="#cb1-332" aria-hidden="true" tabindex="-1"></a><span class="co">  # Configuration options for loftq initialization for LoRA</span></span>
+<span id="cb1-333"><a href="#cb1-333" aria-hidden="true" tabindex="-1"></a><span class="co">  # https://huggingface.co/docs/peft/developer_guides/quantization#loftq-initialization</span></span>
+<span id="cb1-334"><a href="#cb1-334" aria-hidden="true" tabindex="-1"></a><span class="at">  </span><span class="fu">loftq_config</span><span class="kw">:</span></span>
+<span id="cb1-335"><a href="#cb1-335" aria-hidden="true" tabindex="-1"></a><span class="at">    </span><span class="fu">loftq_bits</span><span class="kw">:</span><span class="co">  # typically 4 bits</span></span>
+<span id="cb1-336"><a href="#cb1-336" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb1-337"><a href="#cb1-337" aria-hidden="true" tabindex="-1"></a><span class="co"># ReLoRA configuration</span></span>
+<span id="cb1-338"><a href="#cb1-338" aria-hidden="true" tabindex="-1"></a><span class="co"># Must use either 'lora' or 'qlora' adapter, and does not support fsdp or deepspeed</span></span>
+<span id="cb1-339"><a href="#cb1-339" aria-hidden="true" tabindex="-1"></a><span class="fu">relora_steps</span><span class="kw">:</span><span class="co"> # Number of steps per ReLoRA restart</span></span>
+<span id="cb1-340"><a href="#cb1-340" aria-hidden="true" tabindex="-1"></a><span class="fu">relora_warmup_steps</span><span class="kw">:</span><span class="co"> # Number of per-restart warmup steps</span></span>
+<span id="cb1-341"><a href="#cb1-341" aria-hidden="true" tabindex="-1"></a><span class="fu">relora_anneal_steps</span><span class="kw">:</span><span class="co"> # Number of anneal steps for each relora cycle</span></span>
+<span id="cb1-342"><a href="#cb1-342" aria-hidden="true" tabindex="-1"></a><span class="fu">relora_prune_ratio</span><span class="kw">:</span><span class="co"> # threshold for optimizer magnitude when pruning</span></span>
+<span id="cb1-343"><a href="#cb1-343" aria-hidden="true" tabindex="-1"></a><span class="fu">relora_cpu_offload</span><span class="kw">:</span><span class="co"> # True to perform lora weight merges on cpu during restarts, for modest gpu memory savings</span></span>
+<span id="cb1-344"><a href="#cb1-344" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb1-345"><a href="#cb1-345" aria-hidden="true" tabindex="-1"></a><span class="co"># wandb configuration if you're using it</span></span>
+<span id="cb1-346"><a href="#cb1-346" aria-hidden="true" tabindex="-1"></a><span class="co"># Make sure your `WANDB_API_KEY` environment variable is set (recommended) or you login to wandb with `wandb login`.</span></span>
+<span id="cb1-347"><a href="#cb1-347" aria-hidden="true" tabindex="-1"></a><span class="fu">wandb_mode</span><span class="kw">:</span><span class="co"> # "offline" to save run metadata locally and not sync to the server, "disabled" to turn off wandb</span></span>
+<span id="cb1-348"><a href="#cb1-348" aria-hidden="true" tabindex="-1"></a><span class="fu">wandb_project</span><span class="kw">:</span><span class="co"> # Your wandb project name</span></span>
+<span id="cb1-349"><a href="#cb1-349" aria-hidden="true" tabindex="-1"></a><span class="fu">wandb_entity</span><span class="kw">:</span><span class="co"> # A wandb Team name if using a Team</span></span>
+<span id="cb1-350"><a href="#cb1-350" aria-hidden="true" tabindex="-1"></a><span class="fu">wandb_watch</span><span class="kw">:</span></span>
+<span id="cb1-351"><a href="#cb1-351" aria-hidden="true" tabindex="-1"></a><span class="fu">wandb_name</span><span class="kw">:</span><span class="co"> # Set the name of your wandb run</span></span>
+<span id="cb1-352"><a href="#cb1-352" aria-hidden="true" tabindex="-1"></a><span class="fu">wandb_run_id</span><span class="kw">:</span><span class="co"> # Set the ID of your wandb run</span></span>
+<span id="cb1-353"><a href="#cb1-353" aria-hidden="true" tabindex="-1"></a><span class="fu">wandb_log_model</span><span class="kw">:</span><span class="co"> # "checkpoint" to log model to wandb Artifacts every `save_steps` or "end" to log only at the end of training</span></span>
 <span id="cb1-354"><a href="#cb1-354" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb1-355"><a href="#cb1-355" aria-hidden="true" tabindex="-1"></a><span class="co"># Comet configuration if you're using it</span></span>
-<span id="cb1-356"><a href="#cb1-356" aria-hidden="true" tabindex="-1"></a><span class="co"># Make sure your `COMET_API_KEY` environment variable is set (recommended) or you login to Comet with `comet login`.</span></span>
-<span id="cb1-357"><a href="#cb1-357" aria-hidden="true" tabindex="-1"></a><span class="co"># Check out our documentation for more details https://www.comet.com/docs/v2/api-and-sdk/python-sdk/reference/Experiment-Creation/#comet_ml.start</span></span>
-<span id="cb1-358"><a href="#cb1-358" aria-hidden="true" tabindex="-1"></a><span class="fu">use_comet</span><span class="kw">:</span><span class="co"> # Enable or disable Comet integration.</span></span>
-<span id="cb1-359"><a href="#cb1-359" aria-hidden="true" tabindex="-1"></a><span class="fu">comet_api_key</span><span class="kw">:</span><span class="co"> # API key for Comet. Recommended to set via `comet login`.</span></span>
-<span id="cb1-360"><a href="#cb1-360" aria-hidden="true" tabindex="-1"></a><span class="fu">comet_workspace</span><span class="kw">:</span><span class="co"> # Workspace name in Comet. Defaults to the user's default workspace.</span></span>
-<span id="cb1-361"><a href="#cb1-361" aria-hidden="true" tabindex="-1"></a><span class="fu">comet_project_name</span><span class="kw">:</span><span class="co"> # Project name in Comet. Defaults to Uncategorized.</span></span>
-<span id="cb1-362"><a href="#cb1-362" aria-hidden="true" tabindex="-1"></a><span class="fu">comet_experiment_key</span><span class="kw">:</span><span class="co"> # Identifier for the experiment. Used to append data to an existing experiment or control the key of new experiments. Default to a random key.</span></span>
-<span id="cb1-363"><a href="#cb1-363" aria-hidden="true" tabindex="-1"></a><span class="fu">comet_mode</span><span class="kw">:</span><span class="co"> # Create a new experiment ("create") or log to an existing one ("get"). Default ("get_or_create") auto-selects based on configuration.</span></span>
-<span id="cb1-364"><a href="#cb1-364" aria-hidden="true" tabindex="-1"></a><span class="fu">comet_online</span><span class="kw">:</span><span class="co"> # Set to True to log data to Comet server, or False for offline storage. Default is True.</span></span>
-<span id="cb1-365"><a href="#cb1-365" aria-hidden="true" tabindex="-1"></a><span class="fu">comet_experiment_config</span><span class="kw">:</span><span class="co"> # Dictionary for additional configuration settings, see the doc for more details.</span></span>
-<span id="cb1-366"><a href="#cb1-366" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb1-367"><a href="#cb1-367" aria-hidden="true" tabindex="-1"></a><span class="co"># Tensorboard</span></span>
-<span id="cb1-368"><a href="#cb1-368" aria-hidden="true" tabindex="-1"></a><span class="fu">use_tensorboard</span><span class="kw">:</span><span class="co"> # Optional[bool]</span></span>
-<span id="cb1-369"><a href="#cb1-369" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb1-370"><a href="#cb1-370" aria-hidden="true" tabindex="-1"></a><span class="co"># Where to save the full-finetuned model to</span></span>
-<span id="cb1-371"><a href="#cb1-371" aria-hidden="true" tabindex="-1"></a><span class="fu">output_dir</span><span class="kw">:</span><span class="at"> ./completed-model</span></span>
+<span id="cb1-355"><a href="#cb1-355" aria-hidden="true" tabindex="-1"></a><span class="co"># mlflow configuration if you're using it</span></span>
+<span id="cb1-356"><a href="#cb1-356" aria-hidden="true" tabindex="-1"></a><span class="fu">mlflow_tracking_uri</span><span class="kw">:</span><span class="co"> # URI to mlflow</span></span>
+<span id="cb1-357"><a href="#cb1-357" aria-hidden="true" tabindex="-1"></a><span class="fu">mlflow_experiment_name</span><span class="kw">:</span><span class="co"> # Your experiment name</span></span>
+<span id="cb1-358"><a href="#cb1-358" aria-hidden="true" tabindex="-1"></a><span class="fu">mlflow_run_name</span><span class="kw">:</span><span class="co"> # Your run name</span></span>
+<span id="cb1-359"><a href="#cb1-359" aria-hidden="true" tabindex="-1"></a><span class="fu">hf_mlflow_log_artifacts</span><span class="kw">:</span><span class="co">  # set to true to copy each saved checkpoint on each save to mlflow artifact registry</span></span>
+<span id="cb1-360"><a href="#cb1-360" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb1-361"><a href="#cb1-361" aria-hidden="true" tabindex="-1"></a><span class="co"># Comet configuration if you're using it</span></span>
+<span id="cb1-362"><a href="#cb1-362" aria-hidden="true" tabindex="-1"></a><span class="co"># Make sure your `COMET_API_KEY` environment variable is set (recommended) or you login to Comet with `comet login`.</span></span>
+<span id="cb1-363"><a href="#cb1-363" aria-hidden="true" tabindex="-1"></a><span class="co"># Check out our documentation for more details https://www.comet.com/docs/v2/api-and-sdk/python-sdk/reference/Experiment-Creation/#comet_ml.start</span></span>
+<span id="cb1-364"><a href="#cb1-364" aria-hidden="true" tabindex="-1"></a><span class="fu">use_comet</span><span class="kw">:</span><span class="co"> # Enable or disable Comet integration.</span></span>
+<span id="cb1-365"><a href="#cb1-365" aria-hidden="true" tabindex="-1"></a><span class="fu">comet_api_key</span><span class="kw">:</span><span class="co"> # API key for Comet. Recommended to set via `comet login`.</span></span>
+<span id="cb1-366"><a href="#cb1-366" aria-hidden="true" tabindex="-1"></a><span class="fu">comet_workspace</span><span class="kw">:</span><span class="co"> # Workspace name in Comet. Defaults to the user's default workspace.</span></span>
+<span id="cb1-367"><a href="#cb1-367" aria-hidden="true" tabindex="-1"></a><span class="fu">comet_project_name</span><span class="kw">:</span><span class="co"> # Project name in Comet. Defaults to Uncategorized.</span></span>
+<span id="cb1-368"><a href="#cb1-368" aria-hidden="true" tabindex="-1"></a><span class="fu">comet_experiment_key</span><span class="kw">:</span><span class="co"> # Identifier for the experiment. Used to append data to an existing experiment or control the key of new experiments. Default to a random key.</span></span>
+<span id="cb1-369"><a href="#cb1-369" aria-hidden="true" tabindex="-1"></a><span class="fu">comet_mode</span><span class="kw">:</span><span class="co"> # Create a new experiment ("create") or log to an existing one ("get"). Default ("get_or_create") auto-selects based on configuration.</span></span>
+<span id="cb1-370"><a href="#cb1-370" aria-hidden="true" tabindex="-1"></a><span class="fu">comet_online</span><span class="kw">:</span><span class="co"> # Set to True to log data to Comet server, or False for offline storage. Default is True.</span></span>
+<span id="cb1-371"><a href="#cb1-371" aria-hidden="true" tabindex="-1"></a><span class="fu">comet_experiment_config</span><span class="kw">:</span><span class="co"> # Dictionary for additional configuration settings, see the doc for more details.</span></span>
 <span id="cb1-372"><a href="#cb1-372" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb1-373"><a href="#cb1-373" aria-hidden="true" tabindex="-1"></a><span class="co"># Whether to use torch.compile and which backend to use</span></span>
-<span id="cb1-374"><a href="#cb1-374" aria-hidden="true" tabindex="-1"></a><span class="co"># setting to `auto` will enable torch compile when torch&gt;=2.5.1</span></span>
-<span id="cb1-375"><a href="#cb1-375" aria-hidden="true" tabindex="-1"></a><span class="fu">torch_compile</span><span class="kw">:</span><span class="co">  # Optional[Union[Literal["auto"], bool]]</span></span>
-<span id="cb1-376"><a href="#cb1-376" aria-hidden="true" tabindex="-1"></a><span class="fu">torch_compile_backend</span><span class="kw">:</span><span class="co">  # Optional[str]</span></span>
-<span id="cb1-377"><a href="#cb1-377" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb1-378"><a href="#cb1-378" aria-hidden="true" tabindex="-1"></a><span class="co"># Training hyperparameters</span></span>
-<span id="cb1-379"><a href="#cb1-379" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb1-380"><a href="#cb1-380" aria-hidden="true" tabindex="-1"></a><span class="co"># If greater than 1, backpropagation will be skipped and the gradients will be accumulated for the given number of steps.</span></span>
-<span id="cb1-381"><a href="#cb1-381" aria-hidden="true" tabindex="-1"></a><span class="fu">gradient_accumulation_steps</span><span class="kw">:</span><span class="at"> </span><span class="dv">1</span></span>
-<span id="cb1-382"><a href="#cb1-382" aria-hidden="true" tabindex="-1"></a><span class="co"># The number of samples to include in each batch. This is the number of samples sent to each GPU.</span></span>
-<span id="cb1-383"><a href="#cb1-383" aria-hidden="true" tabindex="-1"></a><span class="co"># Batch size per gpu = micro_batch_size * gradient_accumulation_steps</span></span>
-<span id="cb1-384"><a href="#cb1-384" aria-hidden="true" tabindex="-1"></a><span class="fu">micro_batch_size</span><span class="kw">:</span><span class="at"> </span><span class="dv">2</span></span>
-<span id="cb1-385"><a href="#cb1-385" aria-hidden="true" tabindex="-1"></a><span class="fu">eval_batch_size</span><span class="kw">:</span></span>
-<span id="cb1-386"><a href="#cb1-386" aria-hidden="true" tabindex="-1"></a><span class="fu">num_epochs</span><span class="kw">:</span><span class="at"> </span><span class="dv">4</span></span>
-<span id="cb1-387"><a href="#cb1-387" aria-hidden="true" tabindex="-1"></a><span class="fu">warmup_steps</span><span class="kw">:</span><span class="at"> </span><span class="dv">100</span><span class="co">  # cannot use with warmup_ratio</span></span>
-<span id="cb1-388"><a href="#cb1-388" aria-hidden="true" tabindex="-1"></a><span class="fu">warmup_ratio</span><span class="kw">:</span><span class="at"> </span><span class="fl">0.05</span><span class="co">  # cannot use with warmup_steps</span></span>
-<span id="cb1-389"><a href="#cb1-389" aria-hidden="true" tabindex="-1"></a><span class="fu">learning_rate</span><span class="kw">:</span><span class="at"> </span><span class="fl">0.00003</span></span>
-<span id="cb1-390"><a href="#cb1-390" aria-hidden="true" tabindex="-1"></a><span class="fu">lr_quadratic_warmup</span><span class="kw">:</span></span>
-<span id="cb1-391"><a href="#cb1-391" aria-hidden="true" tabindex="-1"></a><span class="fu">logging_steps</span><span class="kw">:</span></span>
-<span id="cb1-392"><a href="#cb1-392" aria-hidden="true" tabindex="-1"></a><span class="fu">eval_steps</span><span class="kw">:</span><span class="co"> # Leave empty to eval at each epoch, integer for every N steps. float for fraction of total steps</span></span>
-<span id="cb1-393"><a href="#cb1-393" aria-hidden="true" tabindex="-1"></a><span class="fu">evals_per_epoch</span><span class="kw">:</span><span class="co"> # number of times per epoch to run evals, mutually exclusive with eval_steps</span></span>
-<span id="cb1-394"><a href="#cb1-394" aria-hidden="true" tabindex="-1"></a><span class="fu">eval_strategy</span><span class="kw">:</span><span class="co"> # Set to `"no"` to skip evaluation, `"epoch"` at end of each epoch, leave empty to infer from `eval_steps`.</span></span>
-<span id="cb1-395"><a href="#cb1-395" aria-hidden="true" tabindex="-1"></a><span class="fu">save_strategy</span><span class="kw">:</span><span class="co"> # Set to `"no"` to skip checkpoint saves, `"epoch"` at end of each epoch, `"best"` when better result is achieved, leave empty to infer from `save_steps`.</span></span>
-<span id="cb1-396"><a href="#cb1-396" aria-hidden="true" tabindex="-1"></a><span class="fu">save_steps</span><span class="kw">:</span><span class="co"> # Leave empty to save at each epoch, integer for every N steps. float for fraction of total steps</span></span>
-<span id="cb1-397"><a href="#cb1-397" aria-hidden="true" tabindex="-1"></a><span class="fu">saves_per_epoch</span><span class="kw">:</span><span class="co"> # number of times per epoch to save a checkpoint, mutually exclusive with save_steps</span></span>
-<span id="cb1-398"><a href="#cb1-398" aria-hidden="true" tabindex="-1"></a><span class="fu">save_total_limit</span><span class="kw">:</span><span class="co"> # Checkpoints saved at a time</span></span>
-<span id="cb1-399"><a href="#cb1-399" aria-hidden="true" tabindex="-1"></a><span class="co"># Maximum number of iterations to train for. It precedes num_epochs which means that</span></span>
-<span id="cb1-400"><a href="#cb1-400" aria-hidden="true" tabindex="-1"></a><span class="co"># if both are set, num_epochs will not be guaranteed.</span></span>
-<span id="cb1-401"><a href="#cb1-401" aria-hidden="true" tabindex="-1"></a><span class="co"># e.g., when 1 epoch is 1000 steps =&gt; `num_epochs: 2` and `max_steps: 100` will train for 100 steps</span></span>
-<span id="cb1-402"><a href="#cb1-402" aria-hidden="true" tabindex="-1"></a><span class="fu">max_steps</span><span class="kw">:</span></span>
-<span id="cb1-403"><a href="#cb1-403" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb1-404"><a href="#cb1-404" aria-hidden="true" tabindex="-1"></a><span class="co"># bool of whether to include tokens trainer per second in the training metrics. This iterates over the entire dataset once, so it takes some time.</span></span>
-<span id="cb1-405"><a href="#cb1-405" aria-hidden="true" tabindex="-1"></a><span class="fu">include_tokens_per_second</span><span class="kw">:</span><span class="co"> # Optional[bool]</span></span>
-<span id="cb1-406"><a href="#cb1-406" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb1-407"><a href="#cb1-407" aria-hidden="true" tabindex="-1"></a><span class="co"># whether to find batch size that fits in memory. Passed to underlying transformers Trainer</span></span>
-<span id="cb1-408"><a href="#cb1-408" aria-hidden="true" tabindex="-1"></a><span class="fu">auto_find_batch_size</span><span class="kw">:</span><span class="co"> # Optional[bool]</span></span>
+<span id="cb1-373"><a href="#cb1-373" aria-hidden="true" tabindex="-1"></a><span class="co"># Tensorboard</span></span>
+<span id="cb1-374"><a href="#cb1-374" aria-hidden="true" tabindex="-1"></a><span class="fu">use_tensorboard</span><span class="kw">:</span><span class="co"> # Optional[bool]</span></span>
+<span id="cb1-375"><a href="#cb1-375" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb1-376"><a href="#cb1-376" aria-hidden="true" tabindex="-1"></a><span class="co"># Where to save the full-finetuned model to</span></span>
+<span id="cb1-377"><a href="#cb1-377" aria-hidden="true" tabindex="-1"></a><span class="fu">output_dir</span><span class="kw">:</span><span class="at"> ./completed-model</span></span>
+<span id="cb1-378"><a href="#cb1-378" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb1-379"><a href="#cb1-379" aria-hidden="true" tabindex="-1"></a><span class="co"># Whether to use torch.compile and which backend to use</span></span>
+<span id="cb1-380"><a href="#cb1-380" aria-hidden="true" tabindex="-1"></a><span class="co"># setting to `auto` will enable torch compile when torch&gt;=2.5.1</span></span>
+<span id="cb1-381"><a href="#cb1-381" aria-hidden="true" tabindex="-1"></a><span class="fu">torch_compile</span><span class="kw">:</span><span class="co">  # Optional[Union[Literal["auto"], bool]]</span></span>
+<span id="cb1-382"><a href="#cb1-382" aria-hidden="true" tabindex="-1"></a><span class="fu">torch_compile_backend</span><span class="kw">:</span><span class="co">  # Optional[str]</span></span>
+<span id="cb1-383"><a href="#cb1-383" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb1-384"><a href="#cb1-384" aria-hidden="true" tabindex="-1"></a><span class="co"># Training hyperparameters</span></span>
+<span id="cb1-385"><a href="#cb1-385" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb1-386"><a href="#cb1-386" aria-hidden="true" tabindex="-1"></a><span class="co"># If greater than 1, backpropagation will be skipped and the gradients will be accumulated for the given number of steps.</span></span>
+<span id="cb1-387"><a href="#cb1-387" aria-hidden="true" tabindex="-1"></a><span class="fu">gradient_accumulation_steps</span><span class="kw">:</span><span class="at"> </span><span class="dv">1</span></span>
+<span id="cb1-388"><a href="#cb1-388" aria-hidden="true" tabindex="-1"></a><span class="co"># The number of samples to include in each batch. This is the number of samples sent to each GPU.</span></span>
+<span id="cb1-389"><a href="#cb1-389" aria-hidden="true" tabindex="-1"></a><span class="co"># Batch size per gpu = micro_batch_size * gradient_accumulation_steps</span></span>
+<span id="cb1-390"><a href="#cb1-390" aria-hidden="true" tabindex="-1"></a><span class="fu">micro_batch_size</span><span class="kw">:</span><span class="at"> </span><span class="dv">2</span></span>
+<span id="cb1-391"><a href="#cb1-391" aria-hidden="true" tabindex="-1"></a><span class="fu">eval_batch_size</span><span class="kw">:</span></span>
+<span id="cb1-392"><a href="#cb1-392" aria-hidden="true" tabindex="-1"></a><span class="fu">num_epochs</span><span class="kw">:</span><span class="at"> </span><span class="dv">4</span></span>
+<span id="cb1-393"><a href="#cb1-393" aria-hidden="true" tabindex="-1"></a><span class="fu">warmup_steps</span><span class="kw">:</span><span class="at"> </span><span class="dv">100</span><span class="co">  # cannot use with warmup_ratio</span></span>
+<span id="cb1-394"><a href="#cb1-394" aria-hidden="true" tabindex="-1"></a><span class="fu">warmup_ratio</span><span class="kw">:</span><span class="at"> </span><span class="fl">0.05</span><span class="co">  # cannot use with warmup_steps</span></span>
+<span id="cb1-395"><a href="#cb1-395" aria-hidden="true" tabindex="-1"></a><span class="fu">learning_rate</span><span class="kw">:</span><span class="at"> </span><span class="fl">0.00003</span></span>
+<span id="cb1-396"><a href="#cb1-396" aria-hidden="true" tabindex="-1"></a><span class="fu">lr_quadratic_warmup</span><span class="kw">:</span></span>
+<span id="cb1-397"><a href="#cb1-397" aria-hidden="true" tabindex="-1"></a><span class="fu">logging_steps</span><span class="kw">:</span></span>
+<span id="cb1-398"><a href="#cb1-398" aria-hidden="true" tabindex="-1"></a><span class="fu">eval_steps</span><span class="kw">:</span><span class="co"> # Leave empty to eval at each epoch, integer for every N steps. float for fraction of total steps</span></span>
+<span id="cb1-399"><a href="#cb1-399" aria-hidden="true" tabindex="-1"></a><span class="fu">evals_per_epoch</span><span class="kw">:</span><span class="co"> # number of times per epoch to run evals, mutually exclusive with eval_steps</span></span>
+<span id="cb1-400"><a href="#cb1-400" aria-hidden="true" tabindex="-1"></a><span class="fu">eval_strategy</span><span class="kw">:</span><span class="co"> # Set to `"no"` to skip evaluation, `"epoch"` at end of each epoch, leave empty to infer from `eval_steps`.</span></span>
+<span id="cb1-401"><a href="#cb1-401" aria-hidden="true" tabindex="-1"></a><span class="fu">save_strategy</span><span class="kw">:</span><span class="co"> # Set to `"no"` to skip checkpoint saves, `"epoch"` at end of each epoch, `"best"` when better result is achieved, leave empty to infer from `save_steps`.</span></span>
+<span id="cb1-402"><a href="#cb1-402" aria-hidden="true" tabindex="-1"></a><span class="fu">save_steps</span><span class="kw">:</span><span class="co"> # Leave empty to save at each epoch, integer for every N steps. float for fraction of total steps</span></span>
+<span id="cb1-403"><a href="#cb1-403" aria-hidden="true" tabindex="-1"></a><span class="fu">saves_per_epoch</span><span class="kw">:</span><span class="co"> # number of times per epoch to save a checkpoint, mutually exclusive with save_steps</span></span>
+<span id="cb1-404"><a href="#cb1-404" aria-hidden="true" tabindex="-1"></a><span class="fu">save_total_limit</span><span class="kw">:</span><span class="co"> # Checkpoints saved at a time</span></span>
+<span id="cb1-405"><a href="#cb1-405" aria-hidden="true" tabindex="-1"></a><span class="co"># Maximum number of iterations to train for. It precedes num_epochs which means that</span></span>
+<span id="cb1-406"><a href="#cb1-406" aria-hidden="true" tabindex="-1"></a><span class="co"># if both are set, num_epochs will not be guaranteed.</span></span>
+<span id="cb1-407"><a href="#cb1-407" aria-hidden="true" tabindex="-1"></a><span class="co"># e.g., when 1 epoch is 1000 steps =&gt; `num_epochs: 2` and `max_steps: 100` will train for 100 steps</span></span>
+<span id="cb1-408"><a href="#cb1-408" aria-hidden="true" tabindex="-1"></a><span class="fu">max_steps</span><span class="kw">:</span></span>
 <span id="cb1-409"><a href="#cb1-409" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb1-410"><a href="#cb1-410" aria-hidden="true" tabindex="-1"></a><span class="fu">eval_table_size</span><span class="kw">:</span><span class="co"> # Approximate number of predictions sent to wandb depending on batch size. Enabled above 0. Default is 0</span></span>
-<span id="cb1-411"><a href="#cb1-411" aria-hidden="true" tabindex="-1"></a><span class="fu">eval_max_new_tokens</span><span class="kw">:</span><span class="co"> # Total number of tokens generated for predictions sent to wandb. Default is 128</span></span>
-<span id="cb1-412"><a href="#cb1-412" aria-hidden="true" tabindex="-1"></a><span class="fu">eval_causal_lm_metrics</span><span class="kw">:</span><span class="co"> # HF evaluate metrics used during evaluation. Default is ["sacrebleu", "comet", "ter", "chrf", "perplexity"]</span></span>
-<span id="cb1-413"><a href="#cb1-413" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb1-414"><a href="#cb1-414" aria-hidden="true" tabindex="-1"></a><span class="fu">profiler_steps</span><span class="kw">:</span><span class="co"> # enable the pytorch profiler to capture the first N steps of training to the output_dir.</span></span>
-<span id="cb1-415"><a href="#cb1-415" aria-hidden="true" tabindex="-1"></a><span class="co">                # see https://pytorch.org/blog/understanding-gpu-memory-1/ for more information</span></span>
-<span id="cb1-416"><a href="#cb1-416" aria-hidden="true" tabindex="-1"></a><span class="co">                # snapshots can be visualized @ https://pytorch.org/memory_viz</span></span>
-<span id="cb1-417"><a href="#cb1-417" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb1-418"><a href="#cb1-418" aria-hidden="true" tabindex="-1"></a><span class="fu">loss_watchdog_threshold</span><span class="kw">:</span><span class="co"> # High loss value, indicating the learning has broken down (a good estimate is ~2 times the loss at the start of training)</span></span>
-<span id="cb1-419"><a href="#cb1-419" aria-hidden="true" tabindex="-1"></a><span class="fu">loss_watchdog_patience</span><span class="kw">:</span><span class="co"> # Number of high-loss steps in a row before the trainer aborts (default: 3)</span></span>
-<span id="cb1-420"><a href="#cb1-420" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb1-421"><a href="#cb1-421" aria-hidden="true" tabindex="-1"></a><span class="co"># Save model as safetensors (require safetensors package)</span></span>
-<span id="cb1-422"><a href="#cb1-422" aria-hidden="true" tabindex="-1"></a><span class="fu">save_safetensors</span><span class="kw">:</span></span>
+<span id="cb1-410"><a href="#cb1-410" aria-hidden="true" tabindex="-1"></a><span class="co"># bool of whether to include tokens trainer per second in the training metrics. This iterates over the entire dataset once, so it takes some time.</span></span>
+<span id="cb1-411"><a href="#cb1-411" aria-hidden="true" tabindex="-1"></a><span class="fu">include_tokens_per_second</span><span class="kw">:</span><span class="co"> # Optional[bool]</span></span>
+<span id="cb1-412"><a href="#cb1-412" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb1-413"><a href="#cb1-413" aria-hidden="true" tabindex="-1"></a><span class="co"># whether to find batch size that fits in memory. Passed to underlying transformers Trainer</span></span>
+<span id="cb1-414"><a href="#cb1-414" aria-hidden="true" tabindex="-1"></a><span class="fu">auto_find_batch_size</span><span class="kw">:</span><span class="co"> # Optional[bool]</span></span>
+<span id="cb1-415"><a href="#cb1-415" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb1-416"><a href="#cb1-416" aria-hidden="true" tabindex="-1"></a><span class="fu">eval_table_size</span><span class="kw">:</span><span class="co"> # Approximate number of predictions sent to wandb depending on batch size. Enabled above 0. Default is 0</span></span>
+<span id="cb1-417"><a href="#cb1-417" aria-hidden="true" tabindex="-1"></a><span class="fu">eval_max_new_tokens</span><span class="kw">:</span><span class="co"> # Total number of tokens generated for predictions sent to wandb. Default is 128</span></span>
+<span id="cb1-418"><a href="#cb1-418" aria-hidden="true" tabindex="-1"></a><span class="fu">eval_causal_lm_metrics</span><span class="kw">:</span><span class="co"> # HF evaluate metrics used during evaluation. Default is ["sacrebleu", "comet", "ter", "chrf", "perplexity"]</span></span>
+<span id="cb1-419"><a href="#cb1-419" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb1-420"><a href="#cb1-420" aria-hidden="true" tabindex="-1"></a><span class="fu">profiler_steps</span><span class="kw">:</span><span class="co"> # enable the pytorch profiler to capture the first N steps of training to the output_dir.</span></span>
+<span id="cb1-421"><a href="#cb1-421" aria-hidden="true" tabindex="-1"></a><span class="co">                # see https://pytorch.org/blog/understanding-gpu-memory-1/ for more information</span></span>
+<span id="cb1-422"><a href="#cb1-422" aria-hidden="true" tabindex="-1"></a><span class="co">                # snapshots can be visualized @ https://pytorch.org/memory_viz</span></span>
 <span id="cb1-423"><a href="#cb1-423" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb1-424"><a href="#cb1-424" aria-hidden="true" tabindex="-1"></a><span class="co"># Whether to mask out or include the human's prompt from the training labels</span></span>
-<span id="cb1-425"><a href="#cb1-425" aria-hidden="true" tabindex="-1"></a><span class="fu">train_on_inputs</span><span class="kw">:</span><span class="at"> </span><span class="ch">false</span></span>
-<span id="cb1-426"><a href="#cb1-426" aria-hidden="true" tabindex="-1"></a><span class="co"># Group similarly sized data to minimize padding.</span></span>
-<span id="cb1-427"><a href="#cb1-427" aria-hidden="true" tabindex="-1"></a><span class="co"># May be slower to start, as it must download and sort the entire dataset.</span></span>
-<span id="cb1-428"><a href="#cb1-428" aria-hidden="true" tabindex="-1"></a><span class="co"># Note that training loss may have an oscillating pattern with this enabled.</span></span>
-<span id="cb1-429"><a href="#cb1-429" aria-hidden="true" tabindex="-1"></a><span class="fu">group_by_length</span><span class="kw">:</span><span class="at"> </span><span class="ch">false</span></span>
-<span id="cb1-430"><a href="#cb1-430" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb1-431"><a href="#cb1-431" aria-hidden="true" tabindex="-1"></a><span class="co"># Whether to use gradient checkpointing https://huggingface.co/docs/transformers/v4.18.0/en/performance#gradient-checkpointing</span></span>
-<span id="cb1-432"><a href="#cb1-432" aria-hidden="true" tabindex="-1"></a><span class="fu">gradient_checkpointing</span><span class="kw">:</span><span class="at"> </span><span class="ch">false</span></span>
-<span id="cb1-433"><a href="#cb1-433" aria-hidden="true" tabindex="-1"></a><span class="co"># additional kwargs to pass to the trainer for gradient checkpointing</span></span>
-<span id="cb1-434"><a href="#cb1-434" aria-hidden="true" tabindex="-1"></a><span class="co"># gradient_checkpointing_kwargs:</span></span>
-<span id="cb1-435"><a href="#cb1-435" aria-hidden="true" tabindex="-1"></a><span class="co">#   use_reentrant: true</span></span>
+<span id="cb1-424"><a href="#cb1-424" aria-hidden="true" tabindex="-1"></a><span class="fu">loss_watchdog_threshold</span><span class="kw">:</span><span class="co"> # High loss value, indicating the learning has broken down (a good estimate is ~2 times the loss at the start of training)</span></span>
+<span id="cb1-425"><a href="#cb1-425" aria-hidden="true" tabindex="-1"></a><span class="fu">loss_watchdog_patience</span><span class="kw">:</span><span class="co"> # Number of high-loss steps in a row before the trainer aborts (default: 3)</span></span>
+<span id="cb1-426"><a href="#cb1-426" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb1-427"><a href="#cb1-427" aria-hidden="true" tabindex="-1"></a><span class="co"># Save model as safetensors (require safetensors package)</span></span>
+<span id="cb1-428"><a href="#cb1-428" aria-hidden="true" tabindex="-1"></a><span class="fu">save_safetensors</span><span class="kw">:</span></span>
+<span id="cb1-429"><a href="#cb1-429" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb1-430"><a href="#cb1-430" aria-hidden="true" tabindex="-1"></a><span class="co"># Whether to mask out or include the human's prompt from the training labels</span></span>
+<span id="cb1-431"><a href="#cb1-431" aria-hidden="true" tabindex="-1"></a><span class="fu">train_on_inputs</span><span class="kw">:</span><span class="at"> </span><span class="ch">false</span></span>
+<span id="cb1-432"><a href="#cb1-432" aria-hidden="true" tabindex="-1"></a><span class="co"># Group similarly sized data to minimize padding.</span></span>
+<span id="cb1-433"><a href="#cb1-433" aria-hidden="true" tabindex="-1"></a><span class="co"># May be slower to start, as it must download and sort the entire dataset.</span></span>
+<span id="cb1-434"><a href="#cb1-434" aria-hidden="true" tabindex="-1"></a><span class="co"># Note that training loss may have an oscillating pattern with this enabled.</span></span>
+<span id="cb1-435"><a href="#cb1-435" aria-hidden="true" tabindex="-1"></a><span class="fu">group_by_length</span><span class="kw">:</span><span class="at"> </span><span class="ch">false</span></span>
 <span id="cb1-436"><a href="#cb1-436" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb1-437"><a href="#cb1-437" aria-hidden="true" tabindex="-1"></a><span class="co"># Stop training after this many evaluation losses have increased in a row</span></span>
-<span id="cb1-438"><a href="#cb1-438" aria-hidden="true" tabindex="-1"></a><span class="co"># https://huggingface.co/transformers/v4.2.2/_modules/transformers/trainer_callback.html#EarlyStoppingCallback</span></span>
-<span id="cb1-439"><a href="#cb1-439" aria-hidden="true" tabindex="-1"></a><span class="fu">early_stopping_patience</span><span class="kw">:</span><span class="at"> </span><span class="dv">3</span></span>
-<span id="cb1-440"><a href="#cb1-440" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb1-441"><a href="#cb1-441" aria-hidden="true" tabindex="-1"></a><span class="co"># Specify a scheduler and kwargs to use with the optimizer</span></span>
-<span id="cb1-442"><a href="#cb1-442" aria-hidden="true" tabindex="-1"></a><span class="fu">lr_scheduler</span><span class="kw">:</span><span class="co"> # 'one_cycle' | 'log_sweep' | empty for cosine</span></span>
-<span id="cb1-443"><a href="#cb1-443" aria-hidden="true" tabindex="-1"></a><span class="fu">lr_scheduler_kwargs</span><span class="kw">:</span></span>
-<span id="cb1-444"><a href="#cb1-444" aria-hidden="true" tabindex="-1"></a><span class="fu">cosine_min_lr_ratio</span><span class="kw">:</span><span class="co"> # decay lr to some percentage of the peak lr, e.g. cosine_min_lr_ratio=0.1 for 10% of peak lr</span></span>
-<span id="cb1-445"><a href="#cb1-445" aria-hidden="true" tabindex="-1"></a><span class="fu">cosine_constant_lr_ratio</span><span class="kw">:</span><span class="co"> # freeze lr at some percentage of the step, e.g. cosine_constant_lr_ratio=0.8 means start cosine_min_lr at 80% of training step (https://arxiv.org/pdf/2308.04014.pdf)</span></span>
+<span id="cb1-437"><a href="#cb1-437" aria-hidden="true" tabindex="-1"></a><span class="co"># Whether to use gradient checkpointing https://huggingface.co/docs/transformers/v4.18.0/en/performance#gradient-checkpointing</span></span>
+<span id="cb1-438"><a href="#cb1-438" aria-hidden="true" tabindex="-1"></a><span class="fu">gradient_checkpointing</span><span class="kw">:</span><span class="at"> </span><span class="ch">false</span></span>
+<span id="cb1-439"><a href="#cb1-439" aria-hidden="true" tabindex="-1"></a><span class="co"># additional kwargs to pass to the trainer for gradient checkpointing</span></span>
+<span id="cb1-440"><a href="#cb1-440" aria-hidden="true" tabindex="-1"></a><span class="co"># gradient_checkpointing_kwargs:</span></span>
+<span id="cb1-441"><a href="#cb1-441" aria-hidden="true" tabindex="-1"></a><span class="co">#   use_reentrant: true</span></span>
+<span id="cb1-442"><a href="#cb1-442" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb1-443"><a href="#cb1-443" aria-hidden="true" tabindex="-1"></a><span class="co"># Stop training after this many evaluation losses have increased in a row</span></span>
+<span id="cb1-444"><a href="#cb1-444" aria-hidden="true" tabindex="-1"></a><span class="co"># https://huggingface.co/transformers/v4.2.2/_modules/transformers/trainer_callback.html#EarlyStoppingCallback</span></span>
+<span id="cb1-445"><a href="#cb1-445" aria-hidden="true" tabindex="-1"></a><span class="fu">early_stopping_patience</span><span class="kw">:</span><span class="at"> </span><span class="dv">3</span></span>
 <span id="cb1-446"><a href="#cb1-446" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb1-447"><a href="#cb1-447" aria-hidden="true" tabindex="-1"></a><span class="co"># For one_cycle optim</span></span>
-<span id="cb1-448"><a href="#cb1-448" aria-hidden="true" tabindex="-1"></a><span class="fu">lr_div_factor</span><span class="kw">:</span><span class="co"> # Learning rate div factor</span></span>
-<span id="cb1-449"><a href="#cb1-449" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb1-450"><a href="#cb1-450" aria-hidden="true" tabindex="-1"></a><span class="co"># Specify optimizer</span></span>
-<span id="cb1-451"><a href="#cb1-451" aria-hidden="true" tabindex="-1"></a><span class="co"># Valid values are driven by the Transformers OptimizerNames class, see:</span></span>
-<span id="cb1-452"><a href="#cb1-452" aria-hidden="true" tabindex="-1"></a><span class="co"># https://github.com/huggingface/transformers/blob/95b374952dc27d8511541d6f5a4e22c9ec11fb24/src/transformers/training_args.py#L134</span></span>
-<span id="cb1-453"><a href="#cb1-453" aria-hidden="true" tabindex="-1"></a><span class="co">#</span></span>
-<span id="cb1-454"><a href="#cb1-454" aria-hidden="true" tabindex="-1"></a><span class="co"># Note that not all optimizers may be available in your environment, ex: 'adamw_anyprecision' is part of</span></span>
-<span id="cb1-455"><a href="#cb1-455" aria-hidden="true" tabindex="-1"></a><span class="co"># torchdistx, 'adamw_bnb_8bit' is part of bnb.optim.Adam8bit, etc. When in doubt, it is recommended to start with the optimizer used</span></span>
-<span id="cb1-456"><a href="#cb1-456" aria-hidden="true" tabindex="-1"></a><span class="co"># in the examples/ for your model and fine-tuning use case.</span></span>
-<span id="cb1-457"><a href="#cb1-457" aria-hidden="true" tabindex="-1"></a><span class="co">#</span></span>
-<span id="cb1-458"><a href="#cb1-458" aria-hidden="true" tabindex="-1"></a><span class="co"># Valid values for 'optimizer' include:</span></span>
-<span id="cb1-459"><a href="#cb1-459" aria-hidden="true" tabindex="-1"></a><span class="co"># - adamw_hf</span></span>
-<span id="cb1-460"><a href="#cb1-460" aria-hidden="true" tabindex="-1"></a><span class="co"># - adamw_torch</span></span>
-<span id="cb1-461"><a href="#cb1-461" aria-hidden="true" tabindex="-1"></a><span class="co"># - adamw_torch_fused</span></span>
-<span id="cb1-462"><a href="#cb1-462" aria-hidden="true" tabindex="-1"></a><span class="co"># - adamw_torch_xla</span></span>
-<span id="cb1-463"><a href="#cb1-463" aria-hidden="true" tabindex="-1"></a><span class="co"># - adamw_apex_fused</span></span>
-<span id="cb1-464"><a href="#cb1-464" aria-hidden="true" tabindex="-1"></a><span class="co"># - adopt_adamw (an EXPERIMENTAL optimizer, only for torch version &gt;= 2.5.1)</span></span>
-<span id="cb1-465"><a href="#cb1-465" aria-hidden="true" tabindex="-1"></a><span class="co"># - adafactor</span></span>
-<span id="cb1-466"><a href="#cb1-466" aria-hidden="true" tabindex="-1"></a><span class="co"># - adamw_anyprecision</span></span>
-<span id="cb1-467"><a href="#cb1-467" aria-hidden="true" tabindex="-1"></a><span class="co"># - sgd</span></span>
-<span id="cb1-468"><a href="#cb1-468" aria-hidden="true" tabindex="-1"></a><span class="co"># - adagrad</span></span>
-<span id="cb1-469"><a href="#cb1-469" aria-hidden="true" tabindex="-1"></a><span class="co"># - adamw_bnb_8bit</span></span>
-<span id="cb1-470"><a href="#cb1-470" aria-hidden="true" tabindex="-1"></a><span class="co"># - lion_8bit</span></span>
-<span id="cb1-471"><a href="#cb1-471" aria-hidden="true" tabindex="-1"></a><span class="co"># - lion_32bit</span></span>
-<span id="cb1-472"><a href="#cb1-472" aria-hidden="true" tabindex="-1"></a><span class="co"># - paged_adamw_32bit</span></span>
-<span id="cb1-473"><a href="#cb1-473" aria-hidden="true" tabindex="-1"></a><span class="co"># - paged_adamw_8bit</span></span>
-<span id="cb1-474"><a href="#cb1-474" aria-hidden="true" tabindex="-1"></a><span class="co"># - paged_lion_32bit</span></span>
-<span id="cb1-475"><a href="#cb1-475" aria-hidden="true" tabindex="-1"></a><span class="co"># - paged_lion_8bit</span></span>
-<span id="cb1-476"><a href="#cb1-476" aria-hidden="true" tabindex="-1"></a><span class="co"># - galore_adamw</span></span>
-<span id="cb1-477"><a href="#cb1-477" aria-hidden="true" tabindex="-1"></a><span class="co"># - galore_adamw_8bit</span></span>
-<span id="cb1-478"><a href="#cb1-478" aria-hidden="true" tabindex="-1"></a><span class="co"># - galore_adafactor</span></span>
-<span id="cb1-479"><a href="#cb1-479" aria-hidden="true" tabindex="-1"></a><span class="co"># - galore_adamw_layerwise</span></span>
-<span id="cb1-480"><a href="#cb1-480" aria-hidden="true" tabindex="-1"></a><span class="co"># - galore_adamw_8bit_layerwise</span></span>
-<span id="cb1-481"><a href="#cb1-481" aria-hidden="true" tabindex="-1"></a><span class="co"># - galore_adafactor_layerwise</span></span>
-<span id="cb1-482"><a href="#cb1-482" aria-hidden="true" tabindex="-1"></a><span class="fu">optimizer</span><span class="kw">:</span></span>
-<span id="cb1-483"><a href="#cb1-483" aria-hidden="true" tabindex="-1"></a><span class="co"># Dictionary of arguments to pass to the optimizer</span></span>
-<span id="cb1-484"><a href="#cb1-484" aria-hidden="true" tabindex="-1"></a><span class="fu">optim_args</span><span class="kw">:</span></span>
-<span id="cb1-485"><a href="#cb1-485" aria-hidden="true" tabindex="-1"></a><span class="co"># For Galore Optimizers the following optim_args are available</span></span>
-<span id="cb1-486"><a href="#cb1-486" aria-hidden="true" tabindex="-1"></a><span class="co"># rank:  # type: int</span></span>
-<span id="cb1-487"><a href="#cb1-487" aria-hidden="true" tabindex="-1"></a><span class="co"># update_proj_gap  # type: int</span></span>
-<span id="cb1-488"><a href="#cb1-488" aria-hidden="true" tabindex="-1"></a><span class="co"># scale  # type: float</span></span>
-<span id="cb1-489"><a href="#cb1-489" aria-hidden="true" tabindex="-1"></a><span class="co"># proj_type:  # type: str, default = std</span></span>
-<span id="cb1-490"><a href="#cb1-490" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb1-491"><a href="#cb1-491" aria-hidden="true" tabindex="-1"></a><span class="co"># The target modules to optimize, i.e. the module names that you would like to train, right now this is used only for GaLore algorithm</span></span>
-<span id="cb1-492"><a href="#cb1-492" aria-hidden="true" tabindex="-1"></a><span class="fu">optim_target_modules</span><span class="kw">:</span></span>
-<span id="cb1-493"><a href="#cb1-493" aria-hidden="true" tabindex="-1"></a><span class="co"># - self_attn  # for llama</span></span>
-<span id="cb1-494"><a href="#cb1-494" aria-hidden="true" tabindex="-1"></a><span class="co"># - mlp</span></span>
-<span id="cb1-495"><a href="#cb1-495" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb1-496"><a href="#cb1-496" aria-hidden="true" tabindex="-1"></a><span class="co"># Specify weight decay</span></span>
-<span id="cb1-497"><a href="#cb1-497" aria-hidden="true" tabindex="-1"></a><span class="fu">weight_decay</span><span class="kw">:</span></span>
-<span id="cb1-498"><a href="#cb1-498" aria-hidden="true" tabindex="-1"></a><span class="co"># adamw hyperparams</span></span>
-<span id="cb1-499"><a href="#cb1-499" aria-hidden="true" tabindex="-1"></a><span class="fu">adam_beta1</span><span class="kw">:</span></span>
-<span id="cb1-500"><a href="#cb1-500" aria-hidden="true" tabindex="-1"></a><span class="fu">adam_beta2</span><span class="kw">:</span></span>
-<span id="cb1-501"><a href="#cb1-501" aria-hidden="true" tabindex="-1"></a><span class="fu">adam_epsilon</span><span class="kw">:</span></span>
-<span id="cb1-502"><a href="#cb1-502" aria-hidden="true" tabindex="-1"></a><span class="co"># Gradient clipping max norm</span></span>
-<span id="cb1-503"><a href="#cb1-503" aria-hidden="true" tabindex="-1"></a><span class="fu">max_grad_norm</span><span class="kw">:</span></span>
-<span id="cb1-504"><a href="#cb1-504" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb1-505"><a href="#cb1-505" aria-hidden="true" tabindex="-1"></a><span class="co"># Augmentation techniques</span></span>
-<span id="cb1-506"><a href="#cb1-506" aria-hidden="true" tabindex="-1"></a><span class="co"># NEFT https://arxiv.org/abs/2310.05914, set this to a number (paper default is 5) to add noise to embeddings</span></span>
-<span id="cb1-507"><a href="#cb1-507" aria-hidden="true" tabindex="-1"></a><span class="co"># currently only supported on Llama and Mistral</span></span>
-<span id="cb1-508"><a href="#cb1-508" aria-hidden="true" tabindex="-1"></a><span class="fu">neftune_noise_alpha</span><span class="kw">:</span></span>
-<span id="cb1-509"><a href="#cb1-509" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb1-510"><a href="#cb1-510" aria-hidden="true" tabindex="-1"></a><span class="co"># Whether to bettertransformers</span></span>
-<span id="cb1-511"><a href="#cb1-511" aria-hidden="true" tabindex="-1"></a><span class="fu">flash_optimum</span><span class="kw">:</span></span>
-<span id="cb1-512"><a href="#cb1-512" aria-hidden="true" tabindex="-1"></a><span class="co"># Whether to use xformers attention patch https://github.com/facebookresearch/xformers:</span></span>
-<span id="cb1-513"><a href="#cb1-513" aria-hidden="true" tabindex="-1"></a><span class="fu">xformers_attention</span><span class="kw">:</span></span>
-<span id="cb1-514"><a href="#cb1-514" aria-hidden="true" tabindex="-1"></a><span class="co"># Whether to use flash attention patch https://github.com/Dao-AILab/flash-attention:</span></span>
-<span id="cb1-515"><a href="#cb1-515" aria-hidden="true" tabindex="-1"></a><span class="fu">flash_attention</span><span class="kw">:</span></span>
-<span id="cb1-516"><a href="#cb1-516" aria-hidden="true" tabindex="-1"></a><span class="fu">flash_attn_cross_entropy</span><span class="kw">:</span><span class="co">  # Whether to use flash-attention cross entropy implementation - advanced use only</span></span>
-<span id="cb1-517"><a href="#cb1-517" aria-hidden="true" tabindex="-1"></a><span class="fu">flash_attn_rms_norm</span><span class="kw">:</span><span class="co">  # Whether to use flash-attention rms norm implementation - advanced use only</span></span>
-<span id="cb1-518"><a href="#cb1-518" aria-hidden="true" tabindex="-1"></a><span class="fu">flash_attn_fuse_qkv</span><span class="kw">:</span><span class="co"> # Whether to fuse QKV into a single operation</span></span>
-<span id="cb1-519"><a href="#cb1-519" aria-hidden="true" tabindex="-1"></a><span class="fu">flash_attn_fuse_mlp</span><span class="kw">:</span><span class="co"> # Whether to fuse part of the MLP into a single operation</span></span>
-<span id="cb1-520"><a href="#cb1-520" aria-hidden="true" tabindex="-1"></a><span class="co"># Whether to use scaled-dot-product attention</span></span>
-<span id="cb1-521"><a href="#cb1-521" aria-hidden="true" tabindex="-1"></a><span class="co"># https://pytorch.org/docs/stable/generated/torch.nn.functional.scaled_dot_product_attention.html</span></span>
-<span id="cb1-522"><a href="#cb1-522" aria-hidden="true" tabindex="-1"></a><span class="fu">sdp_attention</span><span class="kw">:</span></span>
-<span id="cb1-523"><a href="#cb1-523" aria-hidden="true" tabindex="-1"></a><span class="co"># Shifted-sparse attention (only llama) - https://arxiv.org/pdf/2309.12307.pdf</span></span>
-<span id="cb1-524"><a href="#cb1-524" aria-hidden="true" tabindex="-1"></a><span class="fu">s2_attention</span><span class="kw">:</span></span>
-<span id="cb1-525"><a href="#cb1-525" aria-hidden="true" tabindex="-1"></a><span class="co"># Resume from a specific checkpoint dir</span></span>
-<span id="cb1-526"><a href="#cb1-526" aria-hidden="true" tabindex="-1"></a><span class="fu">resume_from_checkpoint</span><span class="kw">:</span></span>
-<span id="cb1-527"><a href="#cb1-527" aria-hidden="true" tabindex="-1"></a><span class="co"># If resume_from_checkpoint isn't set and you simply want it to start where it left off.</span></span>
-<span id="cb1-528"><a href="#cb1-528" aria-hidden="true" tabindex="-1"></a><span class="co"># Be careful with this being turned on between different models.</span></span>
-<span id="cb1-529"><a href="#cb1-529" aria-hidden="true" tabindex="-1"></a><span class="fu">auto_resume_from_checkpoints</span><span class="kw">:</span><span class="at"> </span><span class="ch">false</span></span>
-<span id="cb1-530"><a href="#cb1-530" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb1-531"><a href="#cb1-531" aria-hidden="true" tabindex="-1"></a><span class="co"># Don't mess with this, it's here for accelerate and torchrun</span></span>
-<span id="cb1-532"><a href="#cb1-532" aria-hidden="true" tabindex="-1"></a><span class="fu">local_rank</span><span class="kw">:</span></span>
-<span id="cb1-533"><a href="#cb1-533" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb1-534"><a href="#cb1-534" aria-hidden="true" tabindex="-1"></a><span class="co"># Add or change special tokens.</span></span>
-<span id="cb1-535"><a href="#cb1-535" aria-hidden="true" tabindex="-1"></a><span class="co"># If you add tokens here, you don't need to add them to the `tokens` list.</span></span>
-<span id="cb1-536"><a href="#cb1-536" aria-hidden="true" tabindex="-1"></a><span class="fu">special_tokens</span><span class="kw">:</span></span>
-<span id="cb1-537"><a href="#cb1-537" aria-hidden="true" tabindex="-1"></a><span class="co">  # bos_token: "&lt;s&gt;"</span></span>
-<span id="cb1-538"><a href="#cb1-538" aria-hidden="true" tabindex="-1"></a><span class="co">  # eos_token: "&lt;/s&gt;"</span></span>
-<span id="cb1-539"><a href="#cb1-539" aria-hidden="true" tabindex="-1"></a><span class="co">  # unk_token: "&lt;unk&gt;"</span></span>
-<span id="cb1-540"><a href="#cb1-540" aria-hidden="true" tabindex="-1"></a><span class="co">  # pad_token: "[PAD]"</span></span>
+<span id="cb1-447"><a href="#cb1-447" aria-hidden="true" tabindex="-1"></a><span class="co"># Specify a scheduler and kwargs to use with the optimizer</span></span>
+<span id="cb1-448"><a href="#cb1-448" aria-hidden="true" tabindex="-1"></a><span class="fu">lr_scheduler</span><span class="kw">:</span><span class="co"> # 'one_cycle' | 'log_sweep' | empty for cosine</span></span>
+<span id="cb1-449"><a href="#cb1-449" aria-hidden="true" tabindex="-1"></a><span class="fu">lr_scheduler_kwargs</span><span class="kw">:</span></span>
+<span id="cb1-450"><a href="#cb1-450" aria-hidden="true" tabindex="-1"></a><span class="fu">cosine_min_lr_ratio</span><span class="kw">:</span><span class="co"> # decay lr to some percentage of the peak lr, e.g. cosine_min_lr_ratio=0.1 for 10% of peak lr</span></span>
+<span id="cb1-451"><a href="#cb1-451" aria-hidden="true" tabindex="-1"></a><span class="fu">cosine_constant_lr_ratio</span><span class="kw">:</span><span class="co"> # freeze lr at some percentage of the step, e.g. cosine_constant_lr_ratio=0.8 means start cosine_min_lr at 80% of training step (https://arxiv.org/pdf/2308.04014.pdf)</span></span>
+<span id="cb1-452"><a href="#cb1-452" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb1-453"><a href="#cb1-453" aria-hidden="true" tabindex="-1"></a><span class="co"># For one_cycle optim</span></span>
+<span id="cb1-454"><a href="#cb1-454" aria-hidden="true" tabindex="-1"></a><span class="fu">lr_div_factor</span><span class="kw">:</span><span class="co"> # Learning rate div factor</span></span>
+<span id="cb1-455"><a href="#cb1-455" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb1-456"><a href="#cb1-456" aria-hidden="true" tabindex="-1"></a><span class="co"># Specify optimizer</span></span>
+<span id="cb1-457"><a href="#cb1-457" aria-hidden="true" tabindex="-1"></a><span class="co"># Valid values are driven by the Transformers OptimizerNames class, see:</span></span>
+<span id="cb1-458"><a href="#cb1-458" aria-hidden="true" tabindex="-1"></a><span class="co"># https://github.com/huggingface/transformers/blob/95b374952dc27d8511541d6f5a4e22c9ec11fb24/src/transformers/training_args.py#L134</span></span>
+<span id="cb1-459"><a href="#cb1-459" aria-hidden="true" tabindex="-1"></a><span class="co">#</span></span>
+<span id="cb1-460"><a href="#cb1-460" aria-hidden="true" tabindex="-1"></a><span class="co"># Note that not all optimizers may be available in your environment, ex: 'adamw_anyprecision' is part of</span></span>
+<span id="cb1-461"><a href="#cb1-461" aria-hidden="true" tabindex="-1"></a><span class="co"># torchdistx, 'adamw_bnb_8bit' is part of bnb.optim.Adam8bit, etc. When in doubt, it is recommended to start with the optimizer used</span></span>
+<span id="cb1-462"><a href="#cb1-462" aria-hidden="true" tabindex="-1"></a><span class="co"># in the examples/ for your model and fine-tuning use case.</span></span>
+<span id="cb1-463"><a href="#cb1-463" aria-hidden="true" tabindex="-1"></a><span class="co">#</span></span>
+<span id="cb1-464"><a href="#cb1-464" aria-hidden="true" tabindex="-1"></a><span class="co"># Valid values for 'optimizer' include:</span></span>
+<span id="cb1-465"><a href="#cb1-465" aria-hidden="true" tabindex="-1"></a><span class="co"># - adamw_hf</span></span>
+<span id="cb1-466"><a href="#cb1-466" aria-hidden="true" tabindex="-1"></a><span class="co"># - adamw_torch</span></span>
+<span id="cb1-467"><a href="#cb1-467" aria-hidden="true" tabindex="-1"></a><span class="co"># - adamw_torch_fused</span></span>
+<span id="cb1-468"><a href="#cb1-468" aria-hidden="true" tabindex="-1"></a><span class="co"># - adamw_torch_xla</span></span>
+<span id="cb1-469"><a href="#cb1-469" aria-hidden="true" tabindex="-1"></a><span class="co"># - adamw_apex_fused</span></span>
+<span id="cb1-470"><a href="#cb1-470" aria-hidden="true" tabindex="-1"></a><span class="co"># - adopt_adamw (an EXPERIMENTAL optimizer, only for torch version &gt;= 2.5.1)</span></span>
+<span id="cb1-471"><a href="#cb1-471" aria-hidden="true" tabindex="-1"></a><span class="co"># - adafactor</span></span>
+<span id="cb1-472"><a href="#cb1-472" aria-hidden="true" tabindex="-1"></a><span class="co"># - adamw_anyprecision</span></span>
+<span id="cb1-473"><a href="#cb1-473" aria-hidden="true" tabindex="-1"></a><span class="co"># - sgd</span></span>
+<span id="cb1-474"><a href="#cb1-474" aria-hidden="true" tabindex="-1"></a><span class="co"># - adagrad</span></span>
+<span id="cb1-475"><a href="#cb1-475" aria-hidden="true" tabindex="-1"></a><span class="co"># - adamw_bnb_8bit</span></span>
+<span id="cb1-476"><a href="#cb1-476" aria-hidden="true" tabindex="-1"></a><span class="co"># - lion_8bit</span></span>
+<span id="cb1-477"><a href="#cb1-477" aria-hidden="true" tabindex="-1"></a><span class="co"># - lion_32bit</span></span>
+<span id="cb1-478"><a href="#cb1-478" aria-hidden="true" tabindex="-1"></a><span class="co"># - paged_adamw_32bit</span></span>
+<span id="cb1-479"><a href="#cb1-479" aria-hidden="true" tabindex="-1"></a><span class="co"># - paged_adamw_8bit</span></span>
+<span id="cb1-480"><a href="#cb1-480" aria-hidden="true" tabindex="-1"></a><span class="co"># - paged_lion_32bit</span></span>
+<span id="cb1-481"><a href="#cb1-481" aria-hidden="true" tabindex="-1"></a><span class="co"># - paged_lion_8bit</span></span>
+<span id="cb1-482"><a href="#cb1-482" aria-hidden="true" tabindex="-1"></a><span class="co"># - galore_adamw</span></span>
+<span id="cb1-483"><a href="#cb1-483" aria-hidden="true" tabindex="-1"></a><span class="co"># - galore_adamw_8bit</span></span>
+<span id="cb1-484"><a href="#cb1-484" aria-hidden="true" tabindex="-1"></a><span class="co"># - galore_adafactor</span></span>
+<span id="cb1-485"><a href="#cb1-485" aria-hidden="true" tabindex="-1"></a><span class="co"># - galore_adamw_layerwise</span></span>
+<span id="cb1-486"><a href="#cb1-486" aria-hidden="true" tabindex="-1"></a><span class="co"># - galore_adamw_8bit_layerwise</span></span>
+<span id="cb1-487"><a href="#cb1-487" aria-hidden="true" tabindex="-1"></a><span class="co"># - galore_adafactor_layerwise</span></span>
+<span id="cb1-488"><a href="#cb1-488" aria-hidden="true" tabindex="-1"></a><span class="fu">optimizer</span><span class="kw">:</span></span>
+<span id="cb1-489"><a href="#cb1-489" aria-hidden="true" tabindex="-1"></a><span class="co"># Dictionary of arguments to pass to the optimizer</span></span>
+<span id="cb1-490"><a href="#cb1-490" aria-hidden="true" tabindex="-1"></a><span class="fu">optim_args</span><span class="kw">:</span></span>
+<span id="cb1-491"><a href="#cb1-491" aria-hidden="true" tabindex="-1"></a><span class="co"># For Galore Optimizers the following optim_args are available</span></span>
+<span id="cb1-492"><a href="#cb1-492" aria-hidden="true" tabindex="-1"></a><span class="co"># rank:  # type: int</span></span>
+<span id="cb1-493"><a href="#cb1-493" aria-hidden="true" tabindex="-1"></a><span class="co"># update_proj_gap  # type: int</span></span>
+<span id="cb1-494"><a href="#cb1-494" aria-hidden="true" tabindex="-1"></a><span class="co"># scale  # type: float</span></span>
+<span id="cb1-495"><a href="#cb1-495" aria-hidden="true" tabindex="-1"></a><span class="co"># proj_type:  # type: str, default = std</span></span>
+<span id="cb1-496"><a href="#cb1-496" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb1-497"><a href="#cb1-497" aria-hidden="true" tabindex="-1"></a><span class="co"># The target modules to optimize, i.e. the module names that you would like to train, right now this is used only for GaLore algorithm</span></span>
+<span id="cb1-498"><a href="#cb1-498" aria-hidden="true" tabindex="-1"></a><span class="fu">optim_target_modules</span><span class="kw">:</span></span>
+<span id="cb1-499"><a href="#cb1-499" aria-hidden="true" tabindex="-1"></a><span class="co"># - self_attn  # for llama</span></span>
+<span id="cb1-500"><a href="#cb1-500" aria-hidden="true" tabindex="-1"></a><span class="co"># - mlp</span></span>
+<span id="cb1-501"><a href="#cb1-501" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb1-502"><a href="#cb1-502" aria-hidden="true" tabindex="-1"></a><span class="co"># Specify weight decay</span></span>
+<span id="cb1-503"><a href="#cb1-503" aria-hidden="true" tabindex="-1"></a><span class="fu">weight_decay</span><span class="kw">:</span></span>
+<span id="cb1-504"><a href="#cb1-504" aria-hidden="true" tabindex="-1"></a><span class="co"># adamw hyperparams</span></span>
+<span id="cb1-505"><a href="#cb1-505" aria-hidden="true" tabindex="-1"></a><span class="fu">adam_beta1</span><span class="kw">:</span></span>
+<span id="cb1-506"><a href="#cb1-506" aria-hidden="true" tabindex="-1"></a><span class="fu">adam_beta2</span><span class="kw">:</span></span>
+<span id="cb1-507"><a href="#cb1-507" aria-hidden="true" tabindex="-1"></a><span class="fu">adam_epsilon</span><span class="kw">:</span></span>
+<span id="cb1-508"><a href="#cb1-508" aria-hidden="true" tabindex="-1"></a><span class="co"># Gradient clipping max norm</span></span>
+<span id="cb1-509"><a href="#cb1-509" aria-hidden="true" tabindex="-1"></a><span class="fu">max_grad_norm</span><span class="kw">:</span></span>
+<span id="cb1-510"><a href="#cb1-510" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb1-511"><a href="#cb1-511" aria-hidden="true" tabindex="-1"></a><span class="co"># Augmentation techniques</span></span>
+<span id="cb1-512"><a href="#cb1-512" aria-hidden="true" tabindex="-1"></a><span class="co"># NEFT https://arxiv.org/abs/2310.05914, set this to a number (paper default is 5) to add noise to embeddings</span></span>
+<span id="cb1-513"><a href="#cb1-513" aria-hidden="true" tabindex="-1"></a><span class="co"># currently only supported on Llama and Mistral</span></span>
+<span id="cb1-514"><a href="#cb1-514" aria-hidden="true" tabindex="-1"></a><span class="fu">neftune_noise_alpha</span><span class="kw">:</span></span>
+<span id="cb1-515"><a href="#cb1-515" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb1-516"><a href="#cb1-516" aria-hidden="true" tabindex="-1"></a><span class="co"># Whether to bettertransformers</span></span>
+<span id="cb1-517"><a href="#cb1-517" aria-hidden="true" tabindex="-1"></a><span class="fu">flash_optimum</span><span class="kw">:</span></span>
+<span id="cb1-518"><a href="#cb1-518" aria-hidden="true" tabindex="-1"></a><span class="co"># Whether to use xformers attention patch https://github.com/facebookresearch/xformers:</span></span>
+<span id="cb1-519"><a href="#cb1-519" aria-hidden="true" tabindex="-1"></a><span class="fu">xformers_attention</span><span class="kw">:</span></span>
+<span id="cb1-520"><a href="#cb1-520" aria-hidden="true" tabindex="-1"></a><span class="co"># Whether to use flash attention patch https://github.com/Dao-AILab/flash-attention:</span></span>
+<span id="cb1-521"><a href="#cb1-521" aria-hidden="true" tabindex="-1"></a><span class="fu">flash_attention</span><span class="kw">:</span></span>
+<span id="cb1-522"><a href="#cb1-522" aria-hidden="true" tabindex="-1"></a><span class="fu">flash_attn_cross_entropy</span><span class="kw">:</span><span class="co">  # Whether to use flash-attention cross entropy implementation - advanced use only</span></span>
+<span id="cb1-523"><a href="#cb1-523" aria-hidden="true" tabindex="-1"></a><span class="fu">flash_attn_rms_norm</span><span class="kw">:</span><span class="co">  # Whether to use flash-attention rms norm implementation - advanced use only</span></span>
+<span id="cb1-524"><a href="#cb1-524" aria-hidden="true" tabindex="-1"></a><span class="fu">flash_attn_fuse_qkv</span><span class="kw">:</span><span class="co"> # Whether to fuse QKV into a single operation</span></span>
+<span id="cb1-525"><a href="#cb1-525" aria-hidden="true" tabindex="-1"></a><span class="fu">flash_attn_fuse_mlp</span><span class="kw">:</span><span class="co"> # Whether to fuse part of the MLP into a single operation</span></span>
+<span id="cb1-526"><a href="#cb1-526" aria-hidden="true" tabindex="-1"></a><span class="co"># Whether to use scaled-dot-product attention</span></span>
+<span id="cb1-527"><a href="#cb1-527" aria-hidden="true" tabindex="-1"></a><span class="co"># https://pytorch.org/docs/stable/generated/torch.nn.functional.scaled_dot_product_attention.html</span></span>
+<span id="cb1-528"><a href="#cb1-528" aria-hidden="true" tabindex="-1"></a><span class="fu">sdp_attention</span><span class="kw">:</span></span>
+<span id="cb1-529"><a href="#cb1-529" aria-hidden="true" tabindex="-1"></a><span class="co"># Shifted-sparse attention (only llama) - https://arxiv.org/pdf/2309.12307.pdf</span></span>
+<span id="cb1-530"><a href="#cb1-530" aria-hidden="true" tabindex="-1"></a><span class="fu">s2_attention</span><span class="kw">:</span></span>
+<span id="cb1-531"><a href="#cb1-531" aria-hidden="true" tabindex="-1"></a><span class="co"># Optional[bool]. Whether to use low_cpu_mem_usage</span></span>
+<span id="cb1-532"><a href="#cb1-532" aria-hidden="true" tabindex="-1"></a><span class="fu">low_cpu_mem_usage</span><span class="kw">:</span></span>
+<span id="cb1-533"><a href="#cb1-533" aria-hidden="true" tabindex="-1"></a><span class="co"># Resume from a specific checkpoint dir</span></span>
+<span id="cb1-534"><a href="#cb1-534" aria-hidden="true" tabindex="-1"></a><span class="fu">resume_from_checkpoint</span><span class="kw">:</span></span>
+<span id="cb1-535"><a href="#cb1-535" aria-hidden="true" tabindex="-1"></a><span class="co"># If resume_from_checkpoint isn't set and you simply want it to start where it left off.</span></span>
+<span id="cb1-536"><a href="#cb1-536" aria-hidden="true" tabindex="-1"></a><span class="co"># Be careful with this being turned on between different models.</span></span>
+<span id="cb1-537"><a href="#cb1-537" aria-hidden="true" tabindex="-1"></a><span class="fu">auto_resume_from_checkpoints</span><span class="kw">:</span><span class="at"> </span><span class="ch">false</span></span>
+<span id="cb1-538"><a href="#cb1-538" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb1-539"><a href="#cb1-539" aria-hidden="true" tabindex="-1"></a><span class="co"># Don't mess with this, it's here for accelerate and torchrun</span></span>
+<span id="cb1-540"><a href="#cb1-540" aria-hidden="true" tabindex="-1"></a><span class="fu">local_rank</span><span class="kw">:</span></span>
 <span id="cb1-541"><a href="#cb1-541" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb1-542"><a href="#cb1-542" aria-hidden="true" tabindex="-1"></a><span class="co"># Add extra tokens.</span></span>
-<span id="cb1-543"><a href="#cb1-543" aria-hidden="true" tabindex="-1"></a><span class="fu">tokens</span><span class="kw">:</span></span>
-<span id="cb1-544"><a href="#cb1-544" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb1-545"><a href="#cb1-545" aria-hidden="true" tabindex="-1"></a><span class="co"># FSDP</span></span>
-<span id="cb1-546"><a href="#cb1-546" aria-hidden="true" tabindex="-1"></a><span class="fu">fsdp</span><span class="kw">:</span></span>
-<span id="cb1-547"><a href="#cb1-547" aria-hidden="true" tabindex="-1"></a><span class="fu">fsdp_config</span><span class="kw">:</span></span>
-<span id="cb1-548"><a href="#cb1-548" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb1-549"><a href="#cb1-549" aria-hidden="true" tabindex="-1"></a><span class="co"># Deepspeed config path. e.g., deepspeed_configs/zero3.json</span></span>
-<span id="cb1-550"><a href="#cb1-550" aria-hidden="true" tabindex="-1"></a><span class="fu">deepspeed</span><span class="kw">:</span></span>
-<span id="cb1-551"><a href="#cb1-551" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb1-552"><a href="#cb1-552" aria-hidden="true" tabindex="-1"></a><span class="co"># Advanced DDP Arguments</span></span>
-<span id="cb1-553"><a href="#cb1-553" aria-hidden="true" tabindex="-1"></a><span class="fu">ddp_timeout</span><span class="kw">:</span></span>
-<span id="cb1-554"><a href="#cb1-554" aria-hidden="true" tabindex="-1"></a><span class="fu">ddp_bucket_cap_mb</span><span class="kw">:</span></span>
-<span id="cb1-555"><a href="#cb1-555" aria-hidden="true" tabindex="-1"></a><span class="fu">ddp_broadcast_buffers</span><span class="kw">:</span></span>
+<span id="cb1-542"><a href="#cb1-542" aria-hidden="true" tabindex="-1"></a><span class="co"># Add or change special tokens.</span></span>
+<span id="cb1-543"><a href="#cb1-543" aria-hidden="true" tabindex="-1"></a><span class="co"># If you add tokens here, you don't need to add them to the `tokens` list.</span></span>
+<span id="cb1-544"><a href="#cb1-544" aria-hidden="true" tabindex="-1"></a><span class="fu">special_tokens</span><span class="kw">:</span></span>
+<span id="cb1-545"><a href="#cb1-545" aria-hidden="true" tabindex="-1"></a><span class="co">  # bos_token: "&lt;s&gt;"</span></span>
+<span id="cb1-546"><a href="#cb1-546" aria-hidden="true" tabindex="-1"></a><span class="co">  # eos_token: "&lt;/s&gt;"</span></span>
+<span id="cb1-547"><a href="#cb1-547" aria-hidden="true" tabindex="-1"></a><span class="co">  # unk_token: "&lt;unk&gt;"</span></span>
+<span id="cb1-548"><a href="#cb1-548" aria-hidden="true" tabindex="-1"></a><span class="co">  # pad_token: "[PAD]"</span></span>
+<span id="cb1-549"><a href="#cb1-549" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb1-550"><a href="#cb1-550" aria-hidden="true" tabindex="-1"></a><span class="co"># Add extra tokens.</span></span>
+<span id="cb1-551"><a href="#cb1-551" aria-hidden="true" tabindex="-1"></a><span class="fu">tokens</span><span class="kw">:</span></span>
+<span id="cb1-552"><a href="#cb1-552" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb1-553"><a href="#cb1-553" aria-hidden="true" tabindex="-1"></a><span class="co"># FSDP</span></span>
+<span id="cb1-554"><a href="#cb1-554" aria-hidden="true" tabindex="-1"></a><span class="fu">fsdp</span><span class="kw">:</span></span>
+<span id="cb1-555"><a href="#cb1-555" aria-hidden="true" tabindex="-1"></a><span class="fu">fsdp_config</span><span class="kw">:</span></span>
 <span id="cb1-556"><a href="#cb1-556" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb1-557"><a href="#cb1-557" aria-hidden="true" tabindex="-1"></a><span class="co"># Path to torch distx for optim 'adamw_anyprecision'</span></span>
-<span id="cb1-558"><a href="#cb1-558" aria-hidden="true" tabindex="-1"></a><span class="fu">torchdistx_path</span><span class="kw">:</span></span>
+<span id="cb1-557"><a href="#cb1-557" aria-hidden="true" tabindex="-1"></a><span class="co"># Deepspeed config path. e.g., deepspeed_configs/zero3.json</span></span>
+<span id="cb1-558"><a href="#cb1-558" aria-hidden="true" tabindex="-1"></a><span class="fu">deepspeed</span><span class="kw">:</span></span>
 <span id="cb1-559"><a href="#cb1-559" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb1-560"><a href="#cb1-560" aria-hidden="true" tabindex="-1"></a><span class="co"># Set to HF dataset for type: 'completion' for streaming instead of pre-tokenize</span></span>
-<span id="cb1-561"><a href="#cb1-561" aria-hidden="true" tabindex="-1"></a><span class="fu">pretraining_dataset</span><span class="kw">:</span></span>
-<span id="cb1-562"><a href="#cb1-562" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb1-563"><a href="#cb1-563" aria-hidden="true" tabindex="-1"></a><span class="co"># Debug mode</span></span>
-<span id="cb1-564"><a href="#cb1-564" aria-hidden="true" tabindex="-1"></a><span class="fu">debug</span><span class="kw">:</span></span>
-<span id="cb1-565"><a href="#cb1-565" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb1-566"><a href="#cb1-566" aria-hidden="true" tabindex="-1"></a><span class="co"># Seed</span></span>
-<span id="cb1-567"><a href="#cb1-567" aria-hidden="true" tabindex="-1"></a><span class="fu">seed</span><span class="kw">:</span></span>
-<span id="cb1-568"><a href="#cb1-568" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb1-569"><a href="#cb1-569" aria-hidden="true" tabindex="-1"></a><span class="co"># Allow overwrite yml config using from cli</span></span>
-<span id="cb1-570"><a href="#cb1-570" aria-hidden="true" tabindex="-1"></a><span class="fu">strict</span><span class="kw">:</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<span id="cb1-560"><a href="#cb1-560" aria-hidden="true" tabindex="-1"></a><span class="co"># Advanced DDP Arguments</span></span>
+<span id="cb1-561"><a href="#cb1-561" aria-hidden="true" tabindex="-1"></a><span class="fu">ddp_timeout</span><span class="kw">:</span></span>
+<span id="cb1-562"><a href="#cb1-562" aria-hidden="true" tabindex="-1"></a><span class="fu">ddp_bucket_cap_mb</span><span class="kw">:</span></span>
+<span id="cb1-563"><a href="#cb1-563" aria-hidden="true" tabindex="-1"></a><span class="fu">ddp_broadcast_buffers</span><span class="kw">:</span></span>
+<span id="cb1-564"><a href="#cb1-564" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb1-565"><a href="#cb1-565" aria-hidden="true" tabindex="-1"></a><span class="co"># Path to torch distx for optim 'adamw_anyprecision'</span></span>
+<span id="cb1-566"><a href="#cb1-566" aria-hidden="true" tabindex="-1"></a><span class="fu">torchdistx_path</span><span class="kw">:</span></span>
+<span id="cb1-567"><a href="#cb1-567" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb1-568"><a href="#cb1-568" aria-hidden="true" tabindex="-1"></a><span class="co"># Set to HF dataset for type: 'completion' for streaming instead of pre-tokenize</span></span>
+<span id="cb1-569"><a href="#cb1-569" aria-hidden="true" tabindex="-1"></a><span class="fu">pretraining_dataset</span><span class="kw">:</span></span>
+<span id="cb1-570"><a href="#cb1-570" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb1-571"><a href="#cb1-571" aria-hidden="true" tabindex="-1"></a><span class="co"># Debug mode</span></span>
+<span id="cb1-572"><a href="#cb1-572" aria-hidden="true" tabindex="-1"></a><span class="fu">debug</span><span class="kw">:</span></span>
+<span id="cb1-573"><a href="#cb1-573" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb1-574"><a href="#cb1-574" aria-hidden="true" tabindex="-1"></a><span class="co"># Seed</span></span>
+<span id="cb1-575"><a href="#cb1-575" aria-hidden="true" tabindex="-1"></a><span class="fu">seed</span><span class="kw">:</span></span>
+<span id="cb1-576"><a href="#cb1-576" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb1-577"><a href="#cb1-577" aria-hidden="true" tabindex="-1"></a><span class="co"># Allow overwrite yml config using from cli</span></span>
+<span id="cb1-578"><a href="#cb1-578" aria-hidden="true" tabindex="-1"></a><span class="fu">strict</span><span class="kw">:</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>



--- a/docs/custom_integrations.html
+++ b/docs/custom_integrations.html
@@ -240,6 +240,12 @@ pre > code.sourceCode > span > a:first-child::before { text-decoration: underlin
      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
          <li class="sidebar-item">
  <div class="sidebar-item-container"> 
+  <a href="../docs/docker.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Docker</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
  <a href="../docs/multi-gpu.html" class="sidebar-item-text sidebar-link">
 <span class="menu-text">Multi-GPU</span></a>
  </div>
@@ -433,6 +439,8 @@ pre > code.sourceCode > span > a:first-child::before { text-decoration: underlin
  <ul>
  <li><a href="#cut-cross-entropy" id="toc-cut-cross-entropy" class="nav-link active" data-scroll-target="#cut-cross-entropy">Cut Cross Entropy</a>
  <ul class="collapse">
+  <li><a href="#requirements" id="toc-requirements" class="nav-link" data-scroll-target="#requirements">Requirements</a></li>
+  <li><a href="#installation" id="toc-installation" class="nav-link" data-scroll-target="#installation">Installation</a></li>
  <li><a href="#usage" id="toc-usage" class="nav-link" data-scroll-target="#usage">Usage</a></li>
  <li><a href="#citation" id="toc-citation" class="nav-link" data-scroll-target="#citation">Citation</a></li>
  </ul></li>
@@ -492,26 +500,39 @@ pre > code.sourceCode > span > a:first-child::before { text-decoration: underlin
 <h2 class="anchored" data-anchor-id="cut-cross-entropy">Cut Cross Entropy</h2>
 <p>Cut Cross Entropy reduces VRAM usage through optimization on the cross-entropy operation during loss calculation.</p>
 <p>See https://github.com/apple/ml-cross-entropy</p>
+<section id="requirements" class="level3">
+<h3 class="anchored" data-anchor-id="requirements">Requirements</h3>
+<ul>
+<li>PyTorch 2.4.0 or higher</li>
+</ul>
+</section>
+<section id="installation" class="level3">
+<h3 class="anchored" data-anchor-id="installation">Installation</h3>
+<p>Run the following command to install <code>cut_cross_entropy[transformers]</code> if you don’t have it already.</p>
+<div class="sourceCode" id="cb1"><pre class="sourceCode bash code-with-copy"><code class="sourceCode bash"><span id="cb1-1"><a href="#cb1-1" aria-hidden="true" tabindex="-1"></a><span class="ex">python</span> scripts/cutcrossentropy_install.py <span class="kw">|</span> <span class="fu">sh</span></span>
+<span id="cb1-2"><a href="#cb1-2" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb1-3"><a href="#cb1-3" aria-hidden="true" tabindex="-1"></a><span class="ex">pip3</span> uninstall <span class="at">-y</span> cut-cross-entropy <span class="kw">&amp;&amp;</span> <span class="ex">pip3</span> install <span class="st">"cut-cross-entropy @ git+https://github.com/apple/ml-cross-entropy.git@9c297c905f55b73594b5d650722d1e78183b77bd"'</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+</section>
 <section id="usage" class="level3">
 <h3 class="anchored" data-anchor-id="usage">Usage</h3>
-<div class="sourceCode" id="cb1"><pre class="sourceCode yaml code-with-copy"><code class="sourceCode yaml"><span id="cb1-1"><a href="#cb1-1" aria-hidden="true" tabindex="-1"></a><span class="fu">plugins</span><span class="kw">:</span></span>
-<span id="cb1-2"><a href="#cb1-2" aria-hidden="true" tabindex="-1"></a><span class="at">  </span><span class="kw">-</span><span class="at"> axolotl.integrations.cut_cross_entropy.CutCrossEntropyPlugin</span></span>
-<span id="cb1-3"><a href="#cb1-3" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb1-4"><a href="#cb1-4" aria-hidden="true" tabindex="-1"></a><span class="fu">cut_cross_entropy</span><span class="kw">:</span><span class="at"> </span><span class="ch">true</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<div class="sourceCode" id="cb2"><pre class="sourceCode yaml code-with-copy"><code class="sourceCode yaml"><span id="cb2-1"><a href="#cb2-1" aria-hidden="true" tabindex="-1"></a><span class="fu">plugins</span><span class="kw">:</span></span>
+<span id="cb2-2"><a href="#cb2-2" aria-hidden="true" tabindex="-1"></a><span class="at">  </span><span class="kw">-</span><span class="at"> axolotl.integrations.cut_cross_entropy.CutCrossEntropyPlugin</span></span>
+<span id="cb2-3"><a href="#cb2-3" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb2-4"><a href="#cb2-4" aria-hidden="true" tabindex="-1"></a><span class="fu">cut_cross_entropy</span><span class="kw">:</span><span class="at"> </span><span class="ch">true</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
 </section>
 <section id="citation" class="level3">
 <h3 class="anchored" data-anchor-id="citation">Citation</h3>
-<div class="sourceCode" id="cb2"><pre class="sourceCode bib code-with-copy"><code class="sourceCode bibtex"><span id="cb2-1"><a href="#cb2-1" aria-hidden="true" tabindex="-1"></a><span class="va">@article</span>{<span class="ot">wijmans2024cut</span>,</span>
-<span id="cb2-2"><a href="#cb2-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">author</span>       = {Erik Wijmans and</span>
-<span id="cb2-3"><a href="#cb2-3" aria-hidden="true" tabindex="-1"></a>                  Brody Huval and</span>
-<span id="cb2-4"><a href="#cb2-4" aria-hidden="true" tabindex="-1"></a>                  Alexander Hertzberg and</span>
-<span id="cb2-5"><a href="#cb2-5" aria-hidden="true" tabindex="-1"></a>                  Vladlen Koltun and</span>
-<span id="cb2-6"><a href="#cb2-6" aria-hidden="true" tabindex="-1"></a>                  Philipp Kr<span class="ch">\"</span>ahenb<span class="ch">\"</span>uhl},</span>
-<span id="cb2-7"><a href="#cb2-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">title</span>        = {Cut Your Losses in Large-Vocabulary Language Models},</span>
-<span id="cb2-8"><a href="#cb2-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">journal</span>      = {arXiv},</span>
-<span id="cb2-9"><a href="#cb2-9" aria-hidden="true" tabindex="-1"></a>  <span class="dt">year</span>         = {2024},</span>
-<span id="cb2-10"><a href="#cb2-10" aria-hidden="true" tabindex="-1"></a>  <span class="dt">url</span>          = {https://arxiv.org/abs/2411.09009},</span>
-<span id="cb2-11"><a href="#cb2-11" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<div class="sourceCode" id="cb3"><pre class="sourceCode bib code-with-copy"><code class="sourceCode bibtex"><span id="cb3-1"><a href="#cb3-1" aria-hidden="true" tabindex="-1"></a><span class="va">@article</span>{<span class="ot">wijmans2024cut</span>,</span>
+<span id="cb3-2"><a href="#cb3-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">author</span>       = {Erik Wijmans and</span>
+<span id="cb3-3"><a href="#cb3-3" aria-hidden="true" tabindex="-1"></a>                  Brody Huval and</span>
+<span id="cb3-4"><a href="#cb3-4" aria-hidden="true" tabindex="-1"></a>                  Alexander Hertzberg and</span>
+<span id="cb3-5"><a href="#cb3-5" aria-hidden="true" tabindex="-1"></a>                  Vladlen Koltun and</span>
+<span id="cb3-6"><a href="#cb3-6" aria-hidden="true" tabindex="-1"></a>                  Philipp Kr<span class="ch">\"</span>ahenb<span class="ch">\"</span>uhl},</span>
+<span id="cb3-7"><a href="#cb3-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">title</span>        = {Cut Your Losses in Large-Vocabulary Language Models},</span>
+<span id="cb3-8"><a href="#cb3-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">journal</span>      = {arXiv},</span>
+<span id="cb3-9"><a href="#cb3-9" aria-hidden="true" tabindex="-1"></a>  <span class="dt">year</span>         = {2024},</span>
+<span id="cb3-10"><a href="#cb3-10" aria-hidden="true" tabindex="-1"></a>  <span class="dt">url</span>          = {https://arxiv.org/abs/2411.09009},</span>
+<span id="cb3-11"><a href="#cb3-11" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
 <p>Please see reference <a href="https://github.com/axolotl-ai-cloud/axolotl/tree/main/src/axolotl/integrations/cut_cross_entropy">here</a></p>
 </section>
 </section>
@@ -520,20 +541,20 @@ pre > code.sourceCode > span > a:first-child::before { text-decoration: underlin
 <p>See https://github.com/ironjr/grokfast</p>
 <section id="usage-1" class="level3">
 <h3 class="anchored" data-anchor-id="usage-1">Usage</h3>
-<div class="sourceCode" id="cb3"><pre class="sourceCode yaml code-with-copy"><code class="sourceCode yaml"><span id="cb3-1"><a href="#cb3-1" aria-hidden="true" tabindex="-1"></a><span class="fu">plugins</span><span class="kw">:</span></span>
-<span id="cb3-2"><a href="#cb3-2" aria-hidden="true" tabindex="-1"></a><span class="at">  </span><span class="kw">-</span><span class="at"> axolotl.integrations.grokfast.GrokfastPlugin</span></span>
-<span id="cb3-3"><a href="#cb3-3" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb3-4"><a href="#cb3-4" aria-hidden="true" tabindex="-1"></a><span class="fu">grokfast_alpha</span><span class="kw">:</span><span class="at"> </span><span class="fl">2.0</span></span>
-<span id="cb3-5"><a href="#cb3-5" aria-hidden="true" tabindex="-1"></a><span class="fu">grokfast_lamb</span><span class="kw">:</span><span class="at"> </span><span class="fl">0.98</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<div class="sourceCode" id="cb4"><pre class="sourceCode yaml code-with-copy"><code class="sourceCode yaml"><span id="cb4-1"><a href="#cb4-1" aria-hidden="true" tabindex="-1"></a><span class="fu">plugins</span><span class="kw">:</span></span>
+<span id="cb4-2"><a href="#cb4-2" aria-hidden="true" tabindex="-1"></a><span class="at">  </span><span class="kw">-</span><span class="at"> axolotl.integrations.grokfast.GrokfastPlugin</span></span>
+<span id="cb4-3"><a href="#cb4-3" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb4-4"><a href="#cb4-4" aria-hidden="true" tabindex="-1"></a><span class="fu">grokfast_alpha</span><span class="kw">:</span><span class="at"> </span><span class="fl">2.0</span></span>
+<span id="cb4-5"><a href="#cb4-5" aria-hidden="true" tabindex="-1"></a><span class="fu">grokfast_lamb</span><span class="kw">:</span><span class="at"> </span><span class="fl">0.98</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
 </section>
 <section id="citation-1" class="level3">
 <h3 class="anchored" data-anchor-id="citation-1">Citation</h3>
-<div class="sourceCode" id="cb4"><pre class="sourceCode bib code-with-copy"><code class="sourceCode bibtex"><span id="cb4-1"><a href="#cb4-1" aria-hidden="true" tabindex="-1"></a><span class="va">@article</span>{<span class="ot">lee2024grokfast</span>,</span>
-<span id="cb4-2"><a href="#cb4-2" aria-hidden="true" tabindex="-1"></a>    <span class="dt">title</span>={{Grokfast}: Accelerated Grokking by Amplifying Slow Gradients},</span>
-<span id="cb4-3"><a href="#cb4-3" aria-hidden="true" tabindex="-1"></a>    <span class="dt">author</span>={Lee, Jaerin and Kang, Bong Gyun and Kim, Kihoon and Lee, Kyoung Mu},</span>
-<span id="cb4-4"><a href="#cb4-4" aria-hidden="true" tabindex="-1"></a>    <span class="dt">journal</span>={arXiv preprint arXiv:2405.20233},</span>
-<span id="cb4-5"><a href="#cb4-5" aria-hidden="true" tabindex="-1"></a>    <span class="dt">year</span>={2024}</span>
-<span id="cb4-6"><a href="#cb4-6" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<div class="sourceCode" id="cb5"><pre class="sourceCode bib code-with-copy"><code class="sourceCode bibtex"><span id="cb5-1"><a href="#cb5-1" aria-hidden="true" tabindex="-1"></a><span class="va">@article</span>{<span class="ot">lee2024grokfast</span>,</span>
+<span id="cb5-2"><a href="#cb5-2" aria-hidden="true" tabindex="-1"></a>    <span class="dt">title</span>={{Grokfast}: Accelerated Grokking by Amplifying Slow Gradients},</span>
+<span id="cb5-3"><a href="#cb5-3" aria-hidden="true" tabindex="-1"></a>    <span class="dt">author</span>={Lee, Jaerin and Kang, Bong Gyun and Kim, Kihoon and Lee, Kyoung Mu},</span>
+<span id="cb5-4"><a href="#cb5-4" aria-hidden="true" tabindex="-1"></a>    <span class="dt">journal</span>={arXiv preprint arXiv:2405.20233},</span>
+<span id="cb5-5"><a href="#cb5-5" aria-hidden="true" tabindex="-1"></a>    <span class="dt">year</span>={2024}</span>
+<span id="cb5-6"><a href="#cb5-6" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
 <p>Please see reference <a href="https://github.com/axolotl-ai-cloud/axolotl/tree/main/src/axolotl/integrations/grokfast">here</a></p>
 </section>
 </section>
@@ -541,21 +562,21 @@ pre > code.sourceCode > span > a:first-child::before { text-decoration: underlin
 <h2 class="anchored" data-anchor-id="knowledge-distillation-kd">Knowledge Distillation (KD)</h2>
 <section id="usage-2" class="level3">
 <h3 class="anchored" data-anchor-id="usage-2">Usage</h3>
-<div class="sourceCode" id="cb5"><pre class="sourceCode yaml code-with-copy"><code class="sourceCode yaml"><span id="cb5-1"><a href="#cb5-1" aria-hidden="true" tabindex="-1"></a><span class="fu">plugins</span><span class="kw">:</span></span>
-<span id="cb5-2"><a href="#cb5-2" aria-hidden="true" tabindex="-1"></a><span class="at">  </span><span class="kw">-</span><span class="at"> </span><span class="st">"axolotl.integrations.kd.KDPlugin"</span></span>
-<span id="cb5-3"><a href="#cb5-3" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb5-4"><a href="#cb5-4" aria-hidden="true" tabindex="-1"></a><span class="fu">kd_trainer</span><span class="kw">:</span><span class="at"> </span><span class="ch">True</span></span>
-<span id="cb5-5"><a href="#cb5-5" aria-hidden="true" tabindex="-1"></a><span class="fu">kd_ce_alpha</span><span class="kw">:</span><span class="at"> </span><span class="fl">0.1</span></span>
-<span id="cb5-6"><a href="#cb5-6" aria-hidden="true" tabindex="-1"></a><span class="fu">kd_alpha</span><span class="kw">:</span><span class="at"> </span><span class="fl">0.9</span></span>
-<span id="cb5-7"><a href="#cb5-7" aria-hidden="true" tabindex="-1"></a><span class="fu">kd_temperature</span><span class="kw">:</span><span class="at"> </span><span class="fl">1.0</span></span>
-<span id="cb5-8"><a href="#cb5-8" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb5-9"><a href="#cb5-9" aria-hidden="true" tabindex="-1"></a><span class="fu">torch_compile</span><span class="kw">:</span><span class="at"> </span><span class="ch">True</span><span class="co">  # torch&gt;=2.5.1, recommended to reduce vram</span></span>
-<span id="cb5-10"><a href="#cb5-10" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb5-11"><a href="#cb5-11" aria-hidden="true" tabindex="-1"></a><span class="fu">datasets</span><span class="kw">:</span></span>
-<span id="cb5-12"><a href="#cb5-12" aria-hidden="true" tabindex="-1"></a><span class="at">  </span><span class="kw">-</span><span class="at"> </span><span class="fu">path</span><span class="kw">:</span><span class="at"> ...</span></span>
-<span id="cb5-13"><a href="#cb5-13" aria-hidden="true" tabindex="-1"></a><span class="at">    </span><span class="fu">type</span><span class="kw">:</span><span class="at"> </span><span class="st">"axolotl.integrations.kd.chat_template"</span></span>
-<span id="cb5-14"><a href="#cb5-14" aria-hidden="true" tabindex="-1"></a><span class="at">    </span><span class="fu">field_messages</span><span class="kw">:</span><span class="at"> </span><span class="st">"messages_combined"</span></span>
-<span id="cb5-15"><a href="#cb5-15" aria-hidden="true" tabindex="-1"></a><span class="at">    </span><span class="fu">logprobs_field</span><span class="kw">:</span><span class="at"> </span><span class="st">"llm_text_generation_vllm_logprobs"</span><span class="co">  # for kd only, field of logprobs</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<div class="sourceCode" id="cb6"><pre class="sourceCode yaml code-with-copy"><code class="sourceCode yaml"><span id="cb6-1"><a href="#cb6-1" aria-hidden="true" tabindex="-1"></a><span class="fu">plugins</span><span class="kw">:</span></span>
+<span id="cb6-2"><a href="#cb6-2" aria-hidden="true" tabindex="-1"></a><span class="at">  </span><span class="kw">-</span><span class="at"> </span><span class="st">"axolotl.integrations.kd.KDPlugin"</span></span>
+<span id="cb6-3"><a href="#cb6-3" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb6-4"><a href="#cb6-4" aria-hidden="true" tabindex="-1"></a><span class="fu">kd_trainer</span><span class="kw">:</span><span class="at"> </span><span class="ch">True</span></span>
+<span id="cb6-5"><a href="#cb6-5" aria-hidden="true" tabindex="-1"></a><span class="fu">kd_ce_alpha</span><span class="kw">:</span><span class="at"> </span><span class="fl">0.1</span></span>
+<span id="cb6-6"><a href="#cb6-6" aria-hidden="true" tabindex="-1"></a><span class="fu">kd_alpha</span><span class="kw">:</span><span class="at"> </span><span class="fl">0.9</span></span>
+<span id="cb6-7"><a href="#cb6-7" aria-hidden="true" tabindex="-1"></a><span class="fu">kd_temperature</span><span class="kw">:</span><span class="at"> </span><span class="fl">1.0</span></span>
+<span id="cb6-8"><a href="#cb6-8" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb6-9"><a href="#cb6-9" aria-hidden="true" tabindex="-1"></a><span class="fu">torch_compile</span><span class="kw">:</span><span class="at"> </span><span class="ch">True</span><span class="co">  # torch&gt;=2.5.1, recommended to reduce vram</span></span>
+<span id="cb6-10"><a href="#cb6-10" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb6-11"><a href="#cb6-11" aria-hidden="true" tabindex="-1"></a><span class="fu">datasets</span><span class="kw">:</span></span>
+<span id="cb6-12"><a href="#cb6-12" aria-hidden="true" tabindex="-1"></a><span class="at">  </span><span class="kw">-</span><span class="at"> </span><span class="fu">path</span><span class="kw">:</span><span class="at"> ...</span></span>
+<span id="cb6-13"><a href="#cb6-13" aria-hidden="true" tabindex="-1"></a><span class="at">    </span><span class="fu">type</span><span class="kw">:</span><span class="at"> </span><span class="st">"axolotl.integrations.kd.chat_template"</span></span>
+<span id="cb6-14"><a href="#cb6-14" aria-hidden="true" tabindex="-1"></a><span class="at">    </span><span class="fu">field_messages</span><span class="kw">:</span><span class="at"> </span><span class="st">"messages_combined"</span></span>
+<span id="cb6-15"><a href="#cb6-15" aria-hidden="true" tabindex="-1"></a><span class="at">    </span><span class="fu">logprobs_field</span><span class="kw">:</span><span class="at"> </span><span class="st">"llm_text_generation_vllm_logprobs"</span><span class="co">  # for kd only, field of logprobs</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
 <p>An example dataset can be found at <a href="https://huggingface.co/datasets/axolotl-ai-co/evolkit-logprobs-pipeline-75k-v2-sample"><code>axolotl-ai-co/evolkit-logprobs-pipeline-75k-v2-sample</code></a></p>
 <p>Please see reference <a href="https://github.com/axolotl-ai-cloud/axolotl/tree/main/src/axolotl/integrations/kd">here</a></p>
 </section>
@@ -571,26 +592,26 @@ pre > code.sourceCode > span > a:first-child::before { text-decoration: underlin
 <p>See https://github.com/linkedin/Liger-Kernel</p>
 <section id="usage-3" class="level3">
 <h3 class="anchored" data-anchor-id="usage-3">Usage</h3>
-<div class="sourceCode" id="cb6"><pre class="sourceCode yaml code-with-copy"><code class="sourceCode yaml"><span id="cb6-1"><a href="#cb6-1" aria-hidden="true" tabindex="-1"></a><span class="fu">plugins</span><span class="kw">:</span></span>
-<span id="cb6-2"><a href="#cb6-2" aria-hidden="true" tabindex="-1"></a><span class="at">  </span><span class="kw">-</span><span class="at"> axolotl.integrations.liger.LigerPlugin</span></span>
-<span id="cb6-3"><a href="#cb6-3" aria-hidden="true" tabindex="-1"></a><span class="fu">liger_rope</span><span class="kw">:</span><span class="at"> </span><span class="ch">true</span></span>
-<span id="cb6-4"><a href="#cb6-4" aria-hidden="true" tabindex="-1"></a><span class="fu">liger_rms_norm</span><span class="kw">:</span><span class="at"> </span><span class="ch">true</span></span>
-<span id="cb6-5"><a href="#cb6-5" aria-hidden="true" tabindex="-1"></a><span class="fu">liger_glu_activation</span><span class="kw">:</span><span class="at"> </span><span class="ch">true</span></span>
-<span id="cb6-6"><a href="#cb6-6" aria-hidden="true" tabindex="-1"></a><span class="fu">liger_layer_norm</span><span class="kw">:</span><span class="at"> </span><span class="ch">true</span></span>
-<span id="cb6-7"><a href="#cb6-7" aria-hidden="true" tabindex="-1"></a><span class="fu">liger_fused_linear_cross_entropy</span><span class="kw">:</span><span class="at"> </span><span class="ch">true</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<div class="sourceCode" id="cb7"><pre class="sourceCode yaml code-with-copy"><code class="sourceCode yaml"><span id="cb7-1"><a href="#cb7-1" aria-hidden="true" tabindex="-1"></a><span class="fu">plugins</span><span class="kw">:</span></span>
+<span id="cb7-2"><a href="#cb7-2" aria-hidden="true" tabindex="-1"></a><span class="at">  </span><span class="kw">-</span><span class="at"> axolotl.integrations.liger.LigerPlugin</span></span>
+<span id="cb7-3"><a href="#cb7-3" aria-hidden="true" tabindex="-1"></a><span class="fu">liger_rope</span><span class="kw">:</span><span class="at"> </span><span class="ch">true</span></span>
+<span id="cb7-4"><a href="#cb7-4" aria-hidden="true" tabindex="-1"></a><span class="fu">liger_rms_norm</span><span class="kw">:</span><span class="at"> </span><span class="ch">true</span></span>
+<span id="cb7-5"><a href="#cb7-5" aria-hidden="true" tabindex="-1"></a><span class="fu">liger_glu_activation</span><span class="kw">:</span><span class="at"> </span><span class="ch">true</span></span>
+<span id="cb7-6"><a href="#cb7-6" aria-hidden="true" tabindex="-1"></a><span class="fu">liger_layer_norm</span><span class="kw">:</span><span class="at"> </span><span class="ch">true</span></span>
+<span id="cb7-7"><a href="#cb7-7" aria-hidden="true" tabindex="-1"></a><span class="fu">liger_fused_linear_cross_entropy</span><span class="kw">:</span><span class="at"> </span><span class="ch">true</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
 </section>
 <section id="citation-2" class="level3">
 <h3 class="anchored" data-anchor-id="citation-2">Citation</h3>
-<div class="sourceCode" id="cb7"><pre class="sourceCode bib code-with-copy"><code class="sourceCode bibtex"><span id="cb7-1"><a href="#cb7-1" aria-hidden="true" tabindex="-1"></a><span class="va">@article</span>{<span class="ot">hsu2024ligerkernelefficienttriton</span>,</span>
-<span id="cb7-2"><a href="#cb7-2" aria-hidden="true" tabindex="-1"></a>      <span class="dt">title</span>={Liger Kernel: Efficient Triton Kernels for LLM Training},</span>
-<span id="cb7-3"><a href="#cb7-3" aria-hidden="true" tabindex="-1"></a>      <span class="dt">author</span>={Pin-Lun Hsu and Yun Dai and Vignesh Kothapalli and Qingquan Song and Shao Tang and Siyu Zhu and Steven Shimizu and Shivam Sahni and Haowen Ning and Yanning Chen},</span>
-<span id="cb7-4"><a href="#cb7-4" aria-hidden="true" tabindex="-1"></a>      <span class="dt">year</span>={2024},</span>
-<span id="cb7-5"><a href="#cb7-5" aria-hidden="true" tabindex="-1"></a>      <span class="dt">eprint</span>={2410.10989},</span>
-<span id="cb7-6"><a href="#cb7-6" aria-hidden="true" tabindex="-1"></a>      <span class="dt">archivePrefix</span>={arXiv},</span>
-<span id="cb7-7"><a href="#cb7-7" aria-hidden="true" tabindex="-1"></a>      <span class="dt">primaryClass</span>={cs.LG},</span>
-<span id="cb7-8"><a href="#cb7-8" aria-hidden="true" tabindex="-1"></a>      <span class="dt">url</span>={https://arxiv.org/abs/2410.10989},</span>
-<span id="cb7-9"><a href="#cb7-9" aria-hidden="true" tabindex="-1"></a>      <span class="dt">journal</span>={arXiv preprint arXiv:2410.10989},</span>
-<span id="cb7-10"><a href="#cb7-10" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<div class="sourceCode" id="cb8"><pre class="sourceCode bib code-with-copy"><code class="sourceCode bibtex"><span id="cb8-1"><a href="#cb8-1" aria-hidden="true" tabindex="-1"></a><span class="va">@article</span>{<span class="ot">hsu2024ligerkernelefficienttriton</span>,</span>
+<span id="cb8-2"><a href="#cb8-2" aria-hidden="true" tabindex="-1"></a>      <span class="dt">title</span>={Liger Kernel: Efficient Triton Kernels for LLM Training},</span>
+<span id="cb8-3"><a href="#cb8-3" aria-hidden="true" tabindex="-1"></a>      <span class="dt">author</span>={Pin-Lun Hsu and Yun Dai and Vignesh Kothapalli and Qingquan Song and Shao Tang and Siyu Zhu and Steven Shimizu and Shivam Sahni and Haowen Ning and Yanning Chen},</span>
+<span id="cb8-4"><a href="#cb8-4" aria-hidden="true" tabindex="-1"></a>      <span class="dt">year</span>={2024},</span>
+<span id="cb8-5"><a href="#cb8-5" aria-hidden="true" tabindex="-1"></a>      <span class="dt">eprint</span>={2410.10989},</span>
+<span id="cb8-6"><a href="#cb8-6" aria-hidden="true" tabindex="-1"></a>      <span class="dt">archivePrefix</span>={arXiv},</span>
+<span id="cb8-7"><a href="#cb8-7" aria-hidden="true" tabindex="-1"></a>      <span class="dt">primaryClass</span>={cs.LG},</span>
+<span id="cb8-8"><a href="#cb8-8" aria-hidden="true" tabindex="-1"></a>      <span class="dt">url</span>={https://arxiv.org/abs/2410.10989},</span>
+<span id="cb8-9"><a href="#cb8-9" aria-hidden="true" tabindex="-1"></a>      <span class="dt">journal</span>={arXiv preprint arXiv:2410.10989},</span>
+<span id="cb8-10"><a href="#cb8-10" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
 <p>Please see reference <a href="https://github.com/axolotl-ai-cloud/axolotl/tree/main/src/axolotl/integrations/liger">here</a></p>
 </section>
 </section>
@@ -600,29 +621,29 @@ pre > code.sourceCode > span > a:first-child::before { text-decoration: underlin
 <p>See https://github.com/EleutherAI/lm-evaluation-harness</p>
 <section id="usage-4" class="level3">
 <h3 class="anchored" data-anchor-id="usage-4">Usage</h3>
-<div class="sourceCode" id="cb8"><pre class="sourceCode yaml code-with-copy"><code class="sourceCode yaml"><span id="cb8-1"><a href="#cb8-1" aria-hidden="true" tabindex="-1"></a><span class="fu">plugins</span><span class="kw">:</span></span>
-<span id="cb8-2"><a href="#cb8-2" aria-hidden="true" tabindex="-1"></a><span class="at">  </span><span class="kw">-</span><span class="at"> axolotl.integrations.lm_eval.LMEvalPlugin</span></span>
-<span id="cb8-3"><a href="#cb8-3" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb8-4"><a href="#cb8-4" aria-hidden="true" tabindex="-1"></a><span class="fu">lm_eval_tasks</span><span class="kw">:</span></span>
-<span id="cb8-5"><a href="#cb8-5" aria-hidden="true" tabindex="-1"></a><span class="at">  </span><span class="kw">-</span><span class="at"> gsm8k</span></span>
-<span id="cb8-6"><a href="#cb8-6" aria-hidden="true" tabindex="-1"></a><span class="at">  </span><span class="kw">-</span><span class="at"> hellaswag</span></span>
-<span id="cb8-7"><a href="#cb8-7" aria-hidden="true" tabindex="-1"></a><span class="at">  </span><span class="kw">-</span><span class="at"> arc_easy</span></span>
-<span id="cb8-8"><a href="#cb8-8" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb8-9"><a href="#cb8-9" aria-hidden="true" tabindex="-1"></a><span class="fu">lm_eval_batch_size</span><span class="kw">:</span><span class="co"> # Batch size for evaluation</span></span>
-<span id="cb8-10"><a href="#cb8-10" aria-hidden="true" tabindex="-1"></a><span class="fu">output_dir</span><span class="kw">:</span><span class="co"> # Directory to save evaluation results</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<div class="sourceCode" id="cb9"><pre class="sourceCode yaml code-with-copy"><code class="sourceCode yaml"><span id="cb9-1"><a href="#cb9-1" aria-hidden="true" tabindex="-1"></a><span class="fu">plugins</span><span class="kw">:</span></span>
+<span id="cb9-2"><a href="#cb9-2" aria-hidden="true" tabindex="-1"></a><span class="at">  </span><span class="kw">-</span><span class="at"> axolotl.integrations.lm_eval.LMEvalPlugin</span></span>
+<span id="cb9-3"><a href="#cb9-3" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb9-4"><a href="#cb9-4" aria-hidden="true" tabindex="-1"></a><span class="fu">lm_eval_tasks</span><span class="kw">:</span></span>
+<span id="cb9-5"><a href="#cb9-5" aria-hidden="true" tabindex="-1"></a><span class="at">  </span><span class="kw">-</span><span class="at"> gsm8k</span></span>
+<span id="cb9-6"><a href="#cb9-6" aria-hidden="true" tabindex="-1"></a><span class="at">  </span><span class="kw">-</span><span class="at"> hellaswag</span></span>
+<span id="cb9-7"><a href="#cb9-7" aria-hidden="true" tabindex="-1"></a><span class="at">  </span><span class="kw">-</span><span class="at"> arc_easy</span></span>
+<span id="cb9-8"><a href="#cb9-8" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb9-9"><a href="#cb9-9" aria-hidden="true" tabindex="-1"></a><span class="fu">lm_eval_batch_size</span><span class="kw">:</span><span class="co"> # Batch size for evaluation</span></span>
+<span id="cb9-10"><a href="#cb9-10" aria-hidden="true" tabindex="-1"></a><span class="fu">output_dir</span><span class="kw">:</span><span class="co"> # Directory to save evaluation results</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
 </section>
 <section id="citation-3" class="level3">
 <h3 class="anchored" data-anchor-id="citation-3">Citation</h3>
-<div class="sourceCode" id="cb9"><pre class="sourceCode bib code-with-copy"><code class="sourceCode bibtex"><span id="cb9-1"><a href="#cb9-1" aria-hidden="true" tabindex="-1"></a><span class="va">@misc</span>{<span class="ot">eval</span>-<span class="ot">harness</span>,</span>
-<span id="cb9-2"><a href="#cb9-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">author</span>       = {Gao, Leo and Tow, Jonathan and Abbasi, Baber and Biderman, Stella and Black, Sid and DiPofi, Anthony and Foster, Charles and Golding, Laurence and Hsu, Jeffrey and Le Noac'h, Alain and Li, Haonan and McDonell, Kyle and Muennighoff, Niklas and Ociepa, Chris and Phang, Jason and Reynolds, Laria and Schoelkopf, Hailey and Skowron, Aviya and Sutawika, Lintang and Tang, Eric and Thite, Anish and Wang, Ben and Wang, Kevin and Zou, Andy},</span>
-<span id="cb9-3"><a href="#cb9-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">title</span>        = {A framework for few-shot language model evaluation},</span>
-<span id="cb9-4"><a href="#cb9-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">month</span>        = 07,</span>
-<span id="cb9-5"><a href="#cb9-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">year</span>         = 2024,</span>
-<span id="cb9-6"><a href="#cb9-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">publisher</span>    = {Zenodo},</span>
-<span id="cb9-7"><a href="#cb9-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">version</span>      = {v0.4.3},</span>
-<span id="cb9-8"><a href="#cb9-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">doi</span>          = {10.5281/zenodo.12608602},</span>
-<span id="cb9-9"><a href="#cb9-9" aria-hidden="true" tabindex="-1"></a>  <span class="dt">url</span>          = {https://zenodo.org/records/12608602}</span>
-<span id="cb9-10"><a href="#cb9-10" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<div class="sourceCode" id="cb10"><pre class="sourceCode bib code-with-copy"><code class="sourceCode bibtex"><span id="cb10-1"><a href="#cb10-1" aria-hidden="true" tabindex="-1"></a><span class="va">@misc</span>{<span class="ot">eval</span>-<span class="ot">harness</span>,</span>
+<span id="cb10-2"><a href="#cb10-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">author</span>       = {Gao, Leo and Tow, Jonathan and Abbasi, Baber and Biderman, Stella and Black, Sid and DiPofi, Anthony and Foster, Charles and Golding, Laurence and Hsu, Jeffrey and Le Noac'h, Alain and Li, Haonan and McDonell, Kyle and Muennighoff, Niklas and Ociepa, Chris and Phang, Jason and Reynolds, Laria and Schoelkopf, Hailey and Skowron, Aviya and Sutawika, Lintang and Tang, Eric and Thite, Anish and Wang, Ben and Wang, Kevin and Zou, Andy},</span>
+<span id="cb10-3"><a href="#cb10-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">title</span>        = {A framework for few-shot language model evaluation},</span>
+<span id="cb10-4"><a href="#cb10-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">month</span>        = 07,</span>
+<span id="cb10-5"><a href="#cb10-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">year</span>         = 2024,</span>
+<span id="cb10-6"><a href="#cb10-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">publisher</span>    = {Zenodo},</span>
+<span id="cb10-7"><a href="#cb10-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">version</span>      = {v0.4.3},</span>
+<span id="cb10-8"><a href="#cb10-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">doi</span>          = {10.5281/zenodo.12608602},</span>
+<span id="cb10-9"><a href="#cb10-9" aria-hidden="true" tabindex="-1"></a>  <span class="dt">url</span>          = {https://zenodo.org/records/12608602}</span>
+<span id="cb10-10"><a href="#cb10-10" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
 <p>Please see reference <a href="https://github.com/axolotl-ai-cloud/axolotl/tree/main/src/axolotl/integrations/lm_eval">here</a></p>
 </section>
 </section>
@@ -637,23 +658,23 @@ pre > code.sourceCode > span > a:first-child::before { text-decoration: underlin
 </section>
 <section id="usage-5" class="level3">
 <h3 class="anchored" data-anchor-id="usage-5">Usage</h3>
-<div class="sourceCode" id="cb10"><pre class="sourceCode yaml code-with-copy"><code class="sourceCode yaml"><span id="cb10-1"><a href="#cb10-1" aria-hidden="true" tabindex="-1"></a><span class="fu">plugins</span><span class="kw">:</span></span>
-<span id="cb10-2"><a href="#cb10-2" aria-hidden="true" tabindex="-1"></a><span class="at">  </span><span class="kw">-</span><span class="at"> axolotl.integrations.spectrum.SpectrumPlugin</span></span>
-<span id="cb10-3"><a href="#cb10-3" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb10-4"><a href="#cb10-4" aria-hidden="true" tabindex="-1"></a><span class="fu">spectrum_top_fraction</span><span class="kw">:</span><span class="at"> </span><span class="fl">0.5</span></span>
-<span id="cb10-5"><a href="#cb10-5" aria-hidden="true" tabindex="-1"></a><span class="fu">spectrum_model_name</span><span class="kw">:</span><span class="at"> meta-llama/Meta-Llama-3.1-8B</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<div class="sourceCode" id="cb11"><pre class="sourceCode yaml code-with-copy"><code class="sourceCode yaml"><span id="cb11-1"><a href="#cb11-1" aria-hidden="true" tabindex="-1"></a><span class="fu">plugins</span><span class="kw">:</span></span>
+<span id="cb11-2"><a href="#cb11-2" aria-hidden="true" tabindex="-1"></a><span class="at">  </span><span class="kw">-</span><span class="at"> axolotl.integrations.spectrum.SpectrumPlugin</span></span>
+<span id="cb11-3"><a href="#cb11-3" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb11-4"><a href="#cb11-4" aria-hidden="true" tabindex="-1"></a><span class="fu">spectrum_top_fraction</span><span class="kw">:</span><span class="at"> </span><span class="fl">0.5</span></span>
+<span id="cb11-5"><a href="#cb11-5" aria-hidden="true" tabindex="-1"></a><span class="fu">spectrum_model_name</span><span class="kw">:</span><span class="at"> meta-llama/Meta-Llama-3.1-8B</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
 </section>
 <section id="citation-4" class="level3">
 <h3 class="anchored" data-anchor-id="citation-4">Citation</h3>
-<div class="sourceCode" id="cb11"><pre class="sourceCode bib code-with-copy"><code class="sourceCode bibtex"><span id="cb11-1"><a href="#cb11-1" aria-hidden="true" tabindex="-1"></a><span class="va">@misc</span>{<span class="ot">hartford2024spectrumtargetedtrainingsignal</span>,</span>
-<span id="cb11-2"><a href="#cb11-2" aria-hidden="true" tabindex="-1"></a>      <span class="dt">title</span>={Spectrum: Targeted Training on Signal to Noise Ratio},</span>
-<span id="cb11-3"><a href="#cb11-3" aria-hidden="true" tabindex="-1"></a>      <span class="dt">author</span>={Eric Hartford and Lucas Atkins and Fernando Fernandes Neto and David Golchinfar},</span>
-<span id="cb11-4"><a href="#cb11-4" aria-hidden="true" tabindex="-1"></a>      <span class="dt">year</span>={2024},</span>
-<span id="cb11-5"><a href="#cb11-5" aria-hidden="true" tabindex="-1"></a>      <span class="dt">eprint</span>={2406.06623},</span>
-<span id="cb11-6"><a href="#cb11-6" aria-hidden="true" tabindex="-1"></a>      <span class="dt">archivePrefix</span>={arXiv},</span>
-<span id="cb11-7"><a href="#cb11-7" aria-hidden="true" tabindex="-1"></a>      <span class="dt">primaryClass</span>={cs.LG},</span>
-<span id="cb11-8"><a href="#cb11-8" aria-hidden="true" tabindex="-1"></a>      <span class="dt">url</span>={https://arxiv.org/abs/2406.06623},</span>
-<span id="cb11-9"><a href="#cb11-9" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<div class="sourceCode" id="cb12"><pre class="sourceCode bib code-with-copy"><code class="sourceCode bibtex"><span id="cb12-1"><a href="#cb12-1" aria-hidden="true" tabindex="-1"></a><span class="va">@misc</span>{<span class="ot">hartford2024spectrumtargetedtrainingsignal</span>,</span>
+<span id="cb12-2"><a href="#cb12-2" aria-hidden="true" tabindex="-1"></a>      <span class="dt">title</span>={Spectrum: Targeted Training on Signal to Noise Ratio},</span>
+<span id="cb12-3"><a href="#cb12-3" aria-hidden="true" tabindex="-1"></a>      <span class="dt">author</span>={Eric Hartford and Lucas Atkins and Fernando Fernandes Neto and David Golchinfar},</span>
+<span id="cb12-4"><a href="#cb12-4" aria-hidden="true" tabindex="-1"></a>      <span class="dt">year</span>={2024},</span>
+<span id="cb12-5"><a href="#cb12-5" aria-hidden="true" tabindex="-1"></a>      <span class="dt">eprint</span>={2406.06623},</span>
+<span id="cb12-6"><a href="#cb12-6" aria-hidden="true" tabindex="-1"></a>      <span class="dt">archivePrefix</span>={arXiv},</span>
+<span id="cb12-7"><a href="#cb12-7" aria-hidden="true" tabindex="-1"></a>      <span class="dt">primaryClass</span>={cs.LG},</span>
+<span id="cb12-8"><a href="#cb12-8" aria-hidden="true" tabindex="-1"></a>      <span class="dt">url</span>={https://arxiv.org/abs/2406.06623},</span>
+<span id="cb12-9"><a href="#cb12-9" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
 <p>Please see reference <a href="https://github.com/axolotl-ai-cloud/axolotl/tree/main/src/axolotl/integrations/spectrum">here</a></p>


--- a/docs/dataset-formats/conversation.html
+++ b/docs/dataset-formats/conversation.html
@@ -241,6 +241,12 @@ pre > code.sourceCode > span > a:first-child::before { text-decoration: underlin
      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
          <li class="sidebar-item">
  <div class="sidebar-item-container"> 
+  <a href="../../docs/docker.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Docker</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
  <a href="../../docs/multi-gpu.html" class="sidebar-item-text sidebar-link">
 <span class="menu-text">Multi-GPU</span></a>
  </div>
--- a/docs/dataset-formats/index.html
+++ b/docs/dataset-formats/index.html
@@ -241,6 +241,12 @@ pre > code.sourceCode > span > a:first-child::before { text-decoration: underlin
      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
          <li class="sidebar-item">
  <div class="sidebar-item-container"> 
+  <a href="../../docs/docker.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Docker</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
  <a href="../../docs/multi-gpu.html" class="sidebar-item-text sidebar-link">
 <span class="menu-text">Multi-GPU</span></a>
  </div>
@@ -602,7 +608,12 @@ Tip
 <section id="pre-tokenized-dataset" class="level3">
 <h3 class="anchored" data-anchor-id="pre-tokenized-dataset">Pre-Tokenized Dataset</h3>
 <p>We suggest this approach when you want to bring your own tokenized dataset.</p>
-<p>Axolotl expects the dataset to have three keys: - <code>input_ids</code>: from tokenizing formatted prompt - <code>attention_mask</code>: for masking padding. If you don’t add padding, it would be equal to <code>len(input_ids) * [1]</code> - <code>labels</code>: this is the same as <code>input_ids</code>, however, if you want to mask certain tokens, you would set those indices to <code>-100</code>.</p>
+<p>Axolotl expects the dataset to have three keys:</p>
+<ul>
+<li><code>input_ids</code>: from tokenizing formatted prompt</li>
+<li><code>attention_mask</code>: for masking padding. If you don’t add padding, it would be equal to <code>len(input_ids) * [1]</code></li>
+<li><code>labels</code>: this is the same as <code>input_ids</code>, however, if you want to mask certain tokens, you would set those indices to <code>-100</code>.</li>
+</ul>
 <div class="callout callout-style-default callout-tip callout-titled">
 <div class="callout-header d-flex align-content-center">
 <div class="callout-icon-container">
--- a/docs/dataset-formats/inst_tune.html
+++ b/docs/dataset-formats/inst_tune.html
@@ -241,6 +241,12 @@ pre > code.sourceCode > span > a:first-child::before { text-decoration: underlin
      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
          <li class="sidebar-item">
  <div class="sidebar-item-container"> 
+  <a href="../../docs/docker.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Docker</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
  <a href="../../docs/multi-gpu.html" class="sidebar-item-text sidebar-link">
 <span class="menu-text">Multi-GPU</span></a>
  </div>
--- a/docs/dataset-formats/pretraining.html
+++ b/docs/dataset-formats/pretraining.html
@@ -241,6 +241,12 @@ pre > code.sourceCode > span > a:first-child::before { text-decoration: underlin
      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
          <li class="sidebar-item">
  <div class="sidebar-item-container"> 
+  <a href="../../docs/docker.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Docker</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
  <a href="../../docs/multi-gpu.html" class="sidebar-item-text sidebar-link">
 <span class="menu-text">Multi-GPU</span></a>
  </div>
--- a/docs/dataset-formats/stepwise_supervised.html
+++ b/docs/dataset-formats/stepwise_supervised.html
@@ -241,6 +241,12 @@ pre > code.sourceCode > span > a:first-child::before { text-decoration: underlin
      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
          <li class="sidebar-item">
  <div class="sidebar-item-container"> 
+  <a href="../../docs/docker.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Docker</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
  <a href="../../docs/multi-gpu.html" class="sidebar-item-text sidebar-link">
 <span class="menu-text">Multi-GPU</span></a>
  </div>
--- a/docs/dataset-formats/template_free.html
+++ b/docs/dataset-formats/template_free.html
@@ -241,6 +241,12 @@ pre > code.sourceCode > span > a:first-child::before { text-decoration: underlin
      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
          <li class="sidebar-item">
  <div class="sidebar-item-container"> 
+  <a href="../../docs/docker.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Docker</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
  <a href="../../docs/multi-gpu.html" class="sidebar-item-text sidebar-link">
 <span class="menu-text">Multi-GPU</span></a>
  </div>
--- a/docs/dataset-formats/tokenized.html
+++ b/docs/dataset-formats/tokenized.html
@@ -241,6 +241,12 @@ pre > code.sourceCode > span > a:first-child::before { text-decoration: underlin
      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
          <li class="sidebar-item">
  <div class="sidebar-item-container"> 
+  <a href="../../docs/docker.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Docker</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
  <a href="../../docs/multi-gpu.html" class="sidebar-item-text sidebar-link">
 <span class="menu-text">Multi-GPU</span></a>
  </div>
--- a/docs/dataset_preprocessing.html
+++ b/docs/dataset_preprocessing.html
@@ -207,6 +207,12 @@ ul.task-list li input[type="checkbox"] {
      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
          <li class="sidebar-item">
  <div class="sidebar-item-container"> 
+  <a href="../docs/docker.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Docker</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
  <a href="../docs/multi-gpu.html" class="sidebar-item-text sidebar-link">
 <span class="menu-text">Multi-GPU</span></a>
  </div>
--- a/docs/debugging.html
+++ b/docs/debugging.html
@@ -241,6 +241,12 @@ pre > code.sourceCode > span > a:first-child::before { text-decoration: underlin
      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
          <li class="sidebar-item">
  <div class="sidebar-item-container"> 
+  <a href="../docs/docker.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Docker</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
  <a href="../docs/multi-gpu.html" class="sidebar-item-text sidebar-link">
 <span class="menu-text">Multi-GPU</span></a>
  </div>
--- a/docs/docker.html
+++ b/docs/docker.html
--- a/docs/faq.html
+++ b/docs/faq.html
@@ -207,6 +207,12 @@ ul.task-list li input[type="checkbox"] {
      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
          <li class="sidebar-item">
  <div class="sidebar-item-container"> 
+  <a href="../docs/docker.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Docker</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
  <a href="../docs/multi-gpu.html" class="sidebar-item-text sidebar-link">
 <span class="menu-text">Multi-GPU</span></a>
  </div>
@@ -445,8 +451,9 @@ ul.task-list li input[type="checkbox"] {
 <p>A: Try upgrading deepspeed w: <code>pip install -U deepspeed</code></p>
 </blockquote>
 <p><strong>Q: AttributeError: ‘DummyOptim’ object has no attribute ‘step’</strong></p>
+<p><strong>Q: ModuleNotFoundError: No module named ‘mpi4py’ using single GPU with deepspeed</strong></p>
 <blockquote class="blockquote">
-<p>A: You may be using deepspeed with single gpu. Please don’t set <code>deepspeed:</code> in yaml or cli.</p>
+<p>A: You may be using deepspeed with single gpu. Please remove the <code>deepspeed:</code> section in the yaml file or <code>--deepspeed</code> CLI flag.</p>
 </blockquote>
 <p><strong>Q: The codes is stuck on saving preprocessed datasets.</strong></p>
 <blockquote class="blockquote">
--- a/docs/fsdp_qlora.html
+++ b/docs/fsdp_qlora.html
@@ -207,6 +207,12 @@ ul.task-list li input[type="checkbox"] {
      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
          <li class="sidebar-item">
  <div class="sidebar-item-container"> 
+  <a href="../docs/docker.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Docker</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
  <a href="../docs/multi-gpu.html" class="sidebar-item-text sidebar-link">
 <span class="menu-text">Multi-GPU</span></a>
  </div>
--- a/docs/getting-started.html
+++ b/docs/getting-started.html
@@ -240,6 +240,12 @@ pre > code.sourceCode > span > a:first-child::before { text-decoration: underlin
      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
          <li class="sidebar-item">
  <div class="sidebar-item-container"> 
+  <a href="../docs/docker.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Docker</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
  <a href="../docs/multi-gpu.html" class="sidebar-item-text sidebar-link">
 <span class="menu-text">Multi-GPU</span></a>
  </div>
--- a/docs/inference.html
+++ b/docs/inference.html
@@ -240,6 +240,12 @@ pre > code.sourceCode > span > a:first-child::before { text-decoration: underlin
      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
          <li class="sidebar-item">
  <div class="sidebar-item-container"> 
+  <a href="../docs/docker.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Docker</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
  <a href="../docs/multi-gpu.html" class="sidebar-item-text sidebar-link">
 <span class="menu-text">Multi-GPU</span></a>
  </div>
--- a/docs/input_output.html
+++ b/docs/input_output.html
@@ -207,6 +207,12 @@ ul.task-list li input[type="checkbox"] {
      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
          <li class="sidebar-item">
  <div class="sidebar-item-container"> 
+  <a href="../docs/docker.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Docker</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
  <a href="../docs/multi-gpu.html" class="sidebar-item-text sidebar-link">
 <span class="menu-text">Multi-GPU</span></a>
  </div>
--- a/docs/installation.html
+++ b/docs/installation.html
@@ -240,6 +240,12 @@ pre > code.sourceCode > span > a:first-child::before { text-decoration: underlin
      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
          <li class="sidebar-item">
  <div class="sidebar-item-container"> 
+  <a href="../docs/docker.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Docker</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
  <a href="../docs/multi-gpu.html" class="sidebar-item-text sidebar-link">
 <span class="menu-text">Multi-GPU</span></a>
  </div>
@@ -525,6 +531,7 @@ Advanced Docker Configuration
 <span id="cb5-6"><a href="#cb5-6" aria-hidden="true" tabindex="-1"></a>  axolotlai/axolotl:main-latest</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
 </div>
 </div>
+<p>Please refer to the <a href="../docs/docker.html">Docker documentation</a> for more information on the different Docker images that are available.</p>
 </section>
 </section>
 <section id="sec-cloud" class="level2" data-number="3">
--- a/docs/lora_optims.html
+++ b/docs/lora_optims.html
@@ -241,6 +241,12 @@ pre > code.sourceCode > span > a:first-child::before { text-decoration: underlin
      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
          <li class="sidebar-item">
  <div class="sidebar-item-container"> 
+  <a href="../docs/docker.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Docker</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
  <a href="../docs/multi-gpu.html" class="sidebar-item-text sidebar-link">
 <span class="menu-text">Multi-GPU</span></a>
  </div>
--- a/docs/lr_groups.html
+++ b/docs/lr_groups.html
@@ -241,6 +241,12 @@ pre > code.sourceCode > span > a:first-child::before { text-decoration: underlin
      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
          <li class="sidebar-item">
  <div class="sidebar-item-container"> 
+  <a href="../docs/docker.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Docker</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
  <a href="../docs/multi-gpu.html" class="sidebar-item-text sidebar-link">
 <span class="menu-text">Multi-GPU</span></a>
  </div>
--- a/docs/mac.html
+++ b/docs/mac.html
@@ -96,7 +96,7 @@ ul.task-list li input[type="checkbox"] {
      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
        <i class="bi bi-layout-text-sidebar-reverse"></i>
      </button>
-        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../docs/multi-gpu.html">Deployments</a></li><li class="breadcrumb-item"><a href="../docs/mac.html">Mac M-series</a></li></ol></nav>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../docs/docker.html">Deployments</a></li><li class="breadcrumb-item"><a href="../docs/mac.html">Mac M-series</a></li></ol></nav>
        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
        </a>
    </div>
@@ -207,6 +207,12 @@ ul.task-list li input[type="checkbox"] {
      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
          <li class="sidebar-item">
  <div class="sidebar-item-container"> 
+  <a href="../docs/docker.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Docker</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
  <a href="../docs/multi-gpu.html" class="sidebar-item-text sidebar-link">
 <span class="menu-text">Multi-GPU</span></a>
  </div>
@@ -399,7 +405,7 @@ ul.task-list li input[type="checkbox"] {
 <!-- main -->
 <main class="content" id="quarto-document-content">

-<header id="title-block-header" class="quarto-title-block default"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../docs/multi-gpu.html">Deployments</a></li><li class="breadcrumb-item"><a href="../docs/mac.html">Mac M-series</a></li></ol></nav>
+<header id="title-block-header" class="quarto-title-block default"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../docs/docker.html">Deployments</a></li><li class="breadcrumb-item"><a href="../docs/mac.html">Mac M-series</a></li></ol></nav>
 <div class="quarto-title">
 <h1 class="title">Mac M-series</h1>
 </div>
--- a/docs/multi-gpu.html
+++ b/docs/multi-gpu.html
@@ -129,7 +129,7 @@ pre > code.sourceCode > span > a:first-child::before { text-decoration: underlin
      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
        <i class="bi bi-layout-text-sidebar-reverse"></i>
      </button>
-        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../docs/multi-gpu.html">Deployments</a></li><li class="breadcrumb-item"><a href="../docs/multi-gpu.html">Multi-GPU</a></li></ol></nav>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../docs/docker.html">Deployments</a></li><li class="breadcrumb-item"><a href="../docs/multi-gpu.html">Multi-GPU</a></li></ol></nav>
        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
        </a>
    </div>
@@ -240,6 +240,12 @@ pre > code.sourceCode > span > a:first-child::before { text-decoration: underlin
      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
          <li class="sidebar-item">
  <div class="sidebar-item-container"> 
+  <a href="../docs/docker.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Docker</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
  <a href="../docs/multi-gpu.html" class="sidebar-item-text sidebar-link active">
 <span class="menu-text">Multi-GPU</span></a>
  </div>
@@ -458,7 +464,7 @@ pre > code.sourceCode > span > a:first-child::before { text-decoration: underlin
 <!-- main -->
 <main class="content" id="quarto-document-content">

-<header id="title-block-header" class="quarto-title-block default"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../docs/multi-gpu.html">Deployments</a></li><li class="breadcrumb-item"><a href="../docs/multi-gpu.html">Multi-GPU</a></li></ol></nav>
+<header id="title-block-header" class="quarto-title-block default"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../docs/docker.html">Deployments</a></li><li class="breadcrumb-item"><a href="../docs/multi-gpu.html">Multi-GPU</a></li></ol></nav>
 <div class="quarto-title">
 <div class="quarto-title-block"><div><h1 class="title">Multi-GPU</h1><button type="button" class="btn code-tools-button" id="quarto-code-tools-source"><i class="bi"></i> Code</button></div></div>
 </div>
--- a/docs/multi-node.html
+++ b/docs/multi-node.html
@@ -130,7 +130,7 @@ pre > code.sourceCode > span > a:first-child::before { text-decoration: underlin
      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
        <i class="bi bi-layout-text-sidebar-reverse"></i>
      </button>
-        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../docs/multi-gpu.html">Deployments</a></li><li class="breadcrumb-item"><a href="../docs/multi-node.html">Multi Node</a></li></ol></nav>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../docs/docker.html">Deployments</a></li><li class="breadcrumb-item"><a href="../docs/multi-node.html">Multi Node</a></li></ol></nav>
        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
        </a>
    </div>
@@ -241,6 +241,12 @@ pre > code.sourceCode > span > a:first-child::before { text-decoration: underlin
      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
          <li class="sidebar-item">
  <div class="sidebar-item-container"> 
+  <a href="../docs/docker.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Docker</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
  <a href="../docs/multi-gpu.html" class="sidebar-item-text sidebar-link">
 <span class="menu-text">Multi-GPU</span></a>
  </div>
@@ -441,7 +447,7 @@ pre > code.sourceCode > span > a:first-child::before { text-decoration: underlin
 <!-- main -->
 <main class="content" id="quarto-document-content">

-<header id="title-block-header" class="quarto-title-block default"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../docs/multi-gpu.html">Deployments</a></li><li class="breadcrumb-item"><a href="../docs/multi-node.html">Multi Node</a></li></ol></nav>
+<header id="title-block-header" class="quarto-title-block default"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../docs/docker.html">Deployments</a></li><li class="breadcrumb-item"><a href="../docs/multi-node.html">Multi Node</a></li></ol></nav>
 <div class="quarto-title">
 <h1 class="title">Multi Node</h1>
 </div>
--- a/docs/multimodal.html
+++ b/docs/multimodal.html
@@ -240,6 +240,12 @@ pre > code.sourceCode > span > a:first-child::before { text-decoration: underlin
      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
          <li class="sidebar-item">
  <div class="sidebar-item-container"> 
+  <a href="../docs/docker.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Docker</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
  <a href="../docs/multi-gpu.html" class="sidebar-item-text sidebar-link">
 <span class="menu-text">Multi-GPU</span></a>
  </div>
--- a/docs/multipack.html
+++ b/docs/multipack.html
@@ -207,6 +207,12 @@ ul.task-list li input[type="checkbox"] {
      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
          <li class="sidebar-item">
  <div class="sidebar-item-container"> 
+  <a href="../docs/docker.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Docker</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
  <a href="../docs/multi-gpu.html" class="sidebar-item-text sidebar-link">
 <span class="menu-text">Multi-GPU</span></a>
  </div>
--- a/docs/nccl.html
+++ b/docs/nccl.html
@@ -241,6 +241,12 @@ pre > code.sourceCode > span > a:first-child::before { text-decoration: underlin
      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
          <li class="sidebar-item">
  <div class="sidebar-item-container"> 
+  <a href="../docs/docker.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Docker</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
  <a href="../docs/multi-gpu.html" class="sidebar-item-text sidebar-link">
 <span class="menu-text">Multi-GPU</span></a>
  </div>
--- a/docs/ray-integration.html
+++ b/docs/ray-integration.html
@@ -130,7 +130,7 @@ pre > code.sourceCode > span > a:first-child::before { text-decoration: underlin
      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
        <i class="bi bi-layout-text-sidebar-reverse"></i>
      </button>
-        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../docs/multi-gpu.html">Deployments</a></li><li class="breadcrumb-item"><a href="../docs/ray-integration.html">Ray Train</a></li></ol></nav>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../docs/docker.html">Deployments</a></li><li class="breadcrumb-item"><a href="../docs/ray-integration.html">Ray Train</a></li></ol></nav>
        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
        </a>
    </div>
@@ -241,6 +241,12 @@ pre > code.sourceCode > span > a:first-child::before { text-decoration: underlin
      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
          <li class="sidebar-item">
  <div class="sidebar-item-container"> 
+  <a href="../docs/docker.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Docker</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
  <a href="../docs/multi-gpu.html" class="sidebar-item-text sidebar-link">
 <span class="menu-text">Multi-GPU</span></a>
  </div>
@@ -442,7 +448,7 @@ pre > code.sourceCode > span > a:first-child::before { text-decoration: underlin
 <!-- main -->
 <main class="content" id="quarto-document-content">

-<header id="title-block-header" class="quarto-title-block default"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../docs/multi-gpu.html">Deployments</a></li><li class="breadcrumb-item"><a href="../docs/ray-integration.html">Ray Train</a></li></ol></nav>
+<header id="title-block-header" class="quarto-title-block default"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../docs/docker.html">Deployments</a></li><li class="breadcrumb-item"><a href="../docs/ray-integration.html">Ray Train</a></li></ol></nav>
 <div class="quarto-title">
 <h1 class="title">Ray Train</h1>
 </div>
--- a/docs/reward_modelling.html
+++ b/docs/reward_modelling.html
@@ -241,6 +241,12 @@ pre > code.sourceCode > span > a:first-child::before { text-decoration: underlin
      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
          <li class="sidebar-item">
  <div class="sidebar-item-container"> 
+  <a href="../docs/docker.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Docker</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
  <a href="../docs/multi-gpu.html" class="sidebar-item-text sidebar-link">
 <span class="menu-text">Multi-GPU</span></a>
  </div>
--- a/docs/rlhf.html
+++ b/docs/rlhf.html
@@ -241,6 +241,12 @@ pre > code.sourceCode > span > a:first-child::before { text-decoration: underlin
      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
          <li class="sidebar-item">
  <div class="sidebar-item-container"> 
+  <a href="../docs/docker.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Docker</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
  <a href="../docs/multi-gpu.html" class="sidebar-item-text sidebar-link">
 <span class="menu-text">Multi-GPU</span></a>
  </div>
--- a/docs/torchao.html
+++ b/docs/torchao.html
@@ -241,6 +241,12 @@ pre > code.sourceCode > span > a:first-child::before { text-decoration: underlin
      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
          <li class="sidebar-item">
  <div class="sidebar-item-container"> 
+  <a href="../docs/docker.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Docker</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
  <a href="../docs/multi-gpu.html" class="sidebar-item-text sidebar-link">
 <span class="menu-text">Multi-GPU</span></a>
  </div>
--- a/docs/unsloth.html
+++ b/docs/unsloth.html
@@ -241,6 +241,12 @@ pre > code.sourceCode > span > a:first-child::before { text-decoration: underlin
      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
          <li class="sidebar-item">
  <div class="sidebar-item-container"> 
+  <a href="../docs/docker.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Docker</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
  <a href="../docs/multi-gpu.html" class="sidebar-item-text sidebar-link">
 <span class="menu-text">Multi-GPU</span></a>
  </div>
--- a/examples/colab-notebooks/colab-axolotl-example.html
+++ b/examples/colab-notebooks/colab-axolotl-example.html
@@ -269,6 +269,12 @@ window.Quarto = {
      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
          <li class="sidebar-item">
  <div class="sidebar-item-container"> 
+  <a href="../../docs/docker.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Docker</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
  <a href="../../docs/multi-gpu.html" class="sidebar-item-text sidebar-link">
 <span class="menu-text">Multi-GPU</span></a>
  </div>
--- a/index.html
+++ b/index.html
@@ -240,6 +240,12 @@ pre > code.sourceCode > span > a:first-child::before { text-decoration: underlin
      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
          <li class="sidebar-item">
  <div class="sidebar-item-container"> 
+  <a href="./docs/docker.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Docker</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
  <a href="./docs/multi-gpu.html" class="sidebar-item-text sidebar-link">
 <span class="menu-text">Multi-GPU</span></a>
  </div>
@@ -456,7 +462,7 @@ pre > code.sourceCode > span > a:first-child::before { text-decoration: underlin
 <picture> <source media="(prefers-color-scheme: dark)" srcset="https://raw.githubusercontent.com/axolotl-ai-cloud/axolotl/887513285d98132142bf5db2a74eb5e0928787f1/image/axolotl_logo_digital_white.svg"> <source media="(prefers-color-scheme: light)" srcset="https://raw.githubusercontent.com/axolotl-ai-cloud/axolotl/887513285d98132142bf5db2a74eb5e0928787f1/image/axolotl_logo_digital_black.svg"> <img alt="Axolotl" src="https://raw.githubusercontent.com/axolotl-ai-cloud/axolotl/887513285d98132142bf5db2a74eb5e0928787f1/image/axolotl_logo_digital_black.svg" width="400" height="104" style="max-width: 100%;"> </picture>
 </p>
 <p align="center">
-<img src="https://img.shields.io/github/license/axolotl-ai-cloud/axolotl.svg?color=blue" alt="GitHub License"> <img src="https://github.com/axolotl-ai-cloud/axolotl/actions/workflows/tests.yml/badge.svg" alt="tests"> <a href="https://github.com/axolotl-ai-cloud/axolotl/releases"><img src="https://img.shields.io/github/release/axolotl-ai-cloud/axolotl.svg" alt="Releases"></a> <br> <a href="https://github.com/axolotl-ai-cloud/axolotl/graphs/contributors"><img src="https://img.shields.io/github/contributors-anon/axolotl-ai-cloud/axolotl?color=yellow&amp;style=flat-square" alt="contributors" style="height: 20px;"></a> <img src="https://img.shields.io/github/stars/axolotl-ai-cloud/axolotl" alt="GitHub Repo stars"> <br> <a href="https://discord.com/invite/HhrNrHJPRb"><img src="https://img.shields.io/badge/discord-7289da.svg?style=flat-square&amp;logo=discord" alt="discord" style="height: 20px;"></a> <a href="https://twitter.com/axolotl_ai"><img src="https://img.shields.io/twitter/follow/axolotl_ai?style=social" alt="twitter" style="height: 20px;"></a> <br> <img src="https://github.com/axolotl-ai-cloud/axolotl/actions/workflows/tests-nightly.yml/badge.svg" alt="tests-nightly"> <img src="https://github.com/axolotl-ai-cloud/axolotl/actions/workflows/multi-gpu-e2e.yml/badge.svg" alt="multigpu-semi-weekly tests"> <a href="https://www.phorm.ai/query?projectId=e315ba4a-4e14-421f-ab05-38a1f9076f25"> <img alt="phorm.ai" src="https://img.shields.io/badge/Phorm-Ask_AI-%23F2777A.svg?&amp;logo=data:image/svg+xml;base64,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"> </a>
+<img src="https://img.shields.io/github/license/axolotl-ai-cloud/axolotl.svg?color=blue" alt="GitHub License"> <img src="https://github.com/axolotl-ai-cloud/axolotl/actions/workflows/tests.yml/badge.svg" alt="tests"> <a href="https://github.com/axolotl-ai-cloud/axolotl/releases"><img src="https://img.shields.io/github/release/axolotl-ai-cloud/axolotl.svg" alt="Releases"></a> <br> <a href="https://github.com/axolotl-ai-cloud/axolotl/graphs/contributors"><img src="https://img.shields.io/github/contributors-anon/axolotl-ai-cloud/axolotl?color=yellow&amp;style=flat-square" alt="contributors" style="height: 20px;"></a> <img src="https://img.shields.io/github/stars/axolotl-ai-cloud/axolotl" alt="GitHub Repo stars"> <br> <a href="https://discord.com/invite/HhrNrHJPRb"><img src="https://img.shields.io/badge/discord-7289da.svg?style=flat-square&amp;logo=discord" alt="discord" style="height: 20px;"></a> <a href="https://twitter.com/axolotl_ai"><img src="https://img.shields.io/twitter/follow/axolotl_ai?style=social" alt="twitter" style="height: 20px;"></a> <br> <img src="https://github.com/axolotl-ai-cloud/axolotl/actions/workflows/tests-nightly.yml/badge.svg" alt="tests-nightly"> <img src="https://github.com/axolotl-ai-cloud/axolotl/actions/workflows/multi-gpu-e2e.yml/badge.svg" alt="multigpu-semi-weekly tests">
 </p>
 <p>Axolotl is a tool designed to streamline post-training for various AI models. Post-training refers to any modifications or additional training performed on pre-trained models - including full model fine-tuning, parameter-efficient tuning (like LoRA and QLoRA), supervised fine-tuning (SFT), instruction tuning, and alignment techniques. With support for multiple model architectures and training configurations, Axolotl makes it easy to get started with these techniques.</p>
 <p>Axolotl is designed to work with YAML config files that contain everything you need to preprocess a dataset, train or fine-tune a model, run model inference or evaluation, and much more.</p>
--- a/search.json
+++ b/search.json
--- a/sitemap.xml
+++ b/sitemap.xml
@@ -2,158 +2,162 @@
 <urlset xmlns="http://www.sitemaps.org/schemas/sitemap/0.9">
  <url>
    <loc>https://axolotl-ai-cloud.github.io/axolotl/examples/colab-notebooks/colab-axolotl-example.html</loc>
-    <lastmod>2025-03-05T13:58:47.518Z</lastmod>
+    <lastmod>2025-03-05T15:02:20.716Z</lastmod>
  </url>
  <url>
    <loc>https://axolotl-ai-cloud.github.io/axolotl/docs/dataset-formats/stepwise_supervised.html</loc>
-    <lastmod>2025-03-05T13:58:47.514Z</lastmod>
+    <lastmod>2025-03-05T15:02:20.712Z</lastmod>
  </url>
  <url>
    <loc>https://axolotl-ai-cloud.github.io/axolotl/docs/dataset-formats/template_free.html</loc>
-    <lastmod>2025-03-05T13:58:47.514Z</lastmod>
+    <lastmod>2025-03-05T15:02:20.712Z</lastmod>
  </url>
  <url>
    <loc>https://axolotl-ai-cloud.github.io/axolotl/docs/dataset-formats/tokenized.html</loc>
-    <lastmod>2025-03-05T13:58:47.514Z</lastmod>
+    <lastmod>2025-03-05T15:02:20.712Z</lastmod>
  </url>
  <url>
    <loc>https://axolotl-ai-cloud.github.io/axolotl/docs/nccl.html</loc>
-    <lastmod>2025-03-05T13:58:47.517Z</lastmod>
+    <lastmod>2025-03-05T15:02:20.715Z</lastmod>
  </url>
  <url>
    <loc>https://axolotl-ai-cloud.github.io/axolotl/docs/amd_hpc.html</loc>
-    <lastmod>2025-03-05T13:58:47.513Z</lastmod>
-  </url>
-  <url>
-    <loc>https://axolotl-ai-cloud.github.io/axolotl/docs/mac.html</loc>
-    <lastmod>2025-03-05T13:58:47.517Z</lastmod>
-  </url>
-  <url>
-    <loc>https://axolotl-ai-cloud.github.io/axolotl/docs/custom_integrations.html</loc>
-    <lastmod>2025-03-05T13:58:47.513Z</lastmod>
-  </url>
-  <url>
-    <loc>https://axolotl-ai-cloud.github.io/axolotl/docs/dataset_preprocessing.html</loc>
-    <lastmod>2025-03-05T13:58:47.514Z</lastmod>
-  </url>
-  <url>
-    <loc>https://axolotl-ai-cloud.github.io/axolotl/docs/fsdp_qlora.html</loc>
-    <lastmod>2025-03-05T13:58:47.514Z</lastmod>
-  </url>
-  <url>
-    <loc>https://axolotl-ai-cloud.github.io/axolotl/docs/unsloth.html</loc>
-    <lastmod>2025-03-05T13:58:47.518Z</lastmod>
-  </url>
-  <url>
-    <loc>https://axolotl-ai-cloud.github.io/axolotl/docs/cli.html</loc>
-    <lastmod>2025-03-05T13:58:47.513Z</lastmod>
-  </url>
-  <url>
-    <loc>https://axolotl-ai-cloud.github.io/axolotl/docs/rlhf.html</loc>
-    <lastmod>2025-03-05T13:58:47.518Z</lastmod>
-  </url>
-  <url>
-    <loc>https://axolotl-ai-cloud.github.io/axolotl/docs/lora_optims.html</loc>
-    <lastmod>2025-03-05T13:58:47.517Z</lastmod>
-  </url>
-  <url>
-    <loc>https://axolotl-ai-cloud.github.io/axolotl/docs/batch_vs_grad.html</loc>
-    <lastmod>2025-03-05T13:58:47.513Z</lastmod>
-  </url>
-  <url>
-    <loc>https://axolotl-ai-cloud.github.io/axolotl/docs/faq.html</loc>
-    <lastmod>2025-03-05T13:58:47.514Z</lastmod>
-  </url>
-  <url>
-    <loc>https://axolotl-ai-cloud.github.io/axolotl/docs/multi-node.html</loc>
-    <lastmod>2025-03-05T13:58:47.517Z</lastmod>
-  </url>
-  <url>
-    <loc>https://axolotl-ai-cloud.github.io/axolotl/FAQS.html</loc>
-    <lastmod>2025-03-05T13:58:47.512Z</lastmod>
-  </url>
-  <url>
-    <loc>https://axolotl-ai-cloud.github.io/axolotl/src/axolotl/integrations/cut_cross_entropy/ACKNOWLEDGEMENTS.html</loc>
-    <lastmod>2025-03-05T13:58:47.534Z</lastmod>
-  </url>
-  <url>
-    <loc>https://axolotl-ai-cloud.github.io/axolotl/index.html</loc>
-    <lastmod>2025-03-05T13:58:47.530Z</lastmod>
-  </url>
-  <url>
-    <loc>https://axolotl-ai-cloud.github.io/axolotl/src/axolotl/integrations/LICENSE.html</loc>
-    <lastmod>2025-03-05T13:58:47.534Z</lastmod>
-  </url>
-  <url>
-    <loc>https://axolotl-ai-cloud.github.io/axolotl/TODO.html</loc>
-    <lastmod>2025-03-05T13:58:47.512Z</lastmod>
-  </url>
-  <url>
-    <loc>https://axolotl-ai-cloud.github.io/axolotl/docs/lr_groups.html</loc>
-    <lastmod>2025-03-05T13:58:47.517Z</lastmod>
-  </url>
-  <url>
-    <loc>https://axolotl-ai-cloud.github.io/axolotl/docs/debugging.html</loc>
-    <lastmod>2025-03-05T13:58:47.514Z</lastmod>
-  </url>
-  <url>
-    <loc>https://axolotl-ai-cloud.github.io/axolotl/docs/multipack.html</loc>
-    <lastmod>2025-03-05T13:58:47.517Z</lastmod>
-  </url>
-  <url>
-    <loc>https://axolotl-ai-cloud.github.io/axolotl/docs/inference.html</loc>
-    <lastmod>2025-03-05T13:58:47.517Z</lastmod>
-  </url>
-  <url>
-    <loc>https://axolotl-ai-cloud.github.io/axolotl/docs/getting-started.html</loc>
-    <lastmod>2025-03-05T13:58:47.514Z</lastmod>
-  </url>
-  <url>
-    <loc>https://axolotl-ai-cloud.github.io/axolotl/docs/multimodal.html</loc>
-    <lastmod>2025-03-05T13:58:47.517Z</lastmod>
-  </url>
-  <url>
-    <loc>https://axolotl-ai-cloud.github.io/axolotl/docs/input_output.html</loc>
-    <lastmod>2025-03-05T13:58:47.517Z</lastmod>
-  </url>
-  <url>
-    <loc>https://axolotl-ai-cloud.github.io/axolotl/docs/reward_modelling.html</loc>
-    <lastmod>2025-03-05T13:58:47.518Z</lastmod>
-  </url>
-  <url>
-    <loc>https://axolotl-ai-cloud.github.io/axolotl/docs/torchao.html</loc>
-    <lastmod>2025-03-05T13:58:47.518Z</lastmod>
-  </url>
-  <url>
-    <loc>https://axolotl-ai-cloud.github.io/axolotl/docs/installation.html</loc>
-    <lastmod>2025-03-05T13:58:47.517Z</lastmod>
-  </url>
-  <url>
-    <loc>https://axolotl-ai-cloud.github.io/axolotl/docs/multi-gpu.html</loc>
-    <lastmod>2025-03-05T13:58:47.517Z</lastmod>
+    <lastmod>2025-03-05T15:02:20.711Z</lastmod>
  </url>
  <url>
    <loc>https://axolotl-ai-cloud.github.io/axolotl/docs/config.html</loc>
-    <lastmod>2025-03-05T13:58:47.513Z</lastmod>
+    <lastmod>2025-03-05T15:02:20.711Z</lastmod>
+  </url>
+  <url>
+    <loc>https://axolotl-ai-cloud.github.io/axolotl/docs/multi-gpu.html</loc>
+    <lastmod>2025-03-05T15:02:20.715Z</lastmod>
+  </url>
+  <url>
+    <loc>https://axolotl-ai-cloud.github.io/axolotl/docs/installation.html</loc>
+    <lastmod>2025-03-05T15:02:20.715Z</lastmod>
+  </url>
+  <url>
+    <loc>https://axolotl-ai-cloud.github.io/axolotl/docs/torchao.html</loc>
+    <lastmod>2025-03-05T15:02:20.715Z</lastmod>
+  </url>
+  <url>
+    <loc>https://axolotl-ai-cloud.github.io/axolotl/docs/reward_modelling.html</loc>
+    <lastmod>2025-03-05T15:02:20.715Z</lastmod>
+  </url>
+  <url>
+    <loc>https://axolotl-ai-cloud.github.io/axolotl/docs/input_output.html</loc>
+    <lastmod>2025-03-05T15:02:20.715Z</lastmod>
+  </url>
+  <url>
+    <loc>https://axolotl-ai-cloud.github.io/axolotl/docs/multimodal.html</loc>
+    <lastmod>2025-03-05T15:02:20.715Z</lastmod>
+  </url>
+  <url>
+    <loc>https://axolotl-ai-cloud.github.io/axolotl/docs/getting-started.html</loc>
+    <lastmod>2025-03-05T15:02:20.712Z</lastmod>
+  </url>
+  <url>
+    <loc>https://axolotl-ai-cloud.github.io/axolotl/docs/inference.html</loc>
+    <lastmod>2025-03-05T15:02:20.715Z</lastmod>
+  </url>
+  <url>
+    <loc>https://axolotl-ai-cloud.github.io/axolotl/docs/multipack.html</loc>
+    <lastmod>2025-03-05T15:02:20.715Z</lastmod>
+  </url>
+  <url>
+    <loc>https://axolotl-ai-cloud.github.io/axolotl/docs/debugging.html</loc>
+    <lastmod>2025-03-05T15:02:20.712Z</lastmod>
+  </url>
+  <url>
+    <loc>https://axolotl-ai-cloud.github.io/axolotl/docs/lr_groups.html</loc>
+    <lastmod>2025-03-05T15:02:20.715Z</lastmod>
+  </url>
+  <url>
+    <loc>https://axolotl-ai-cloud.github.io/axolotl/TODO.html</loc>
+    <lastmod>2025-03-05T15:02:20.710Z</lastmod>
+  </url>
+  <url>
+    <loc>https://axolotl-ai-cloud.github.io/axolotl/src/axolotl/integrations/LICENSE.html</loc>
+    <lastmod>2025-03-05T15:02:20.730Z</lastmod>
+  </url>
+  <url>
+    <loc>https://axolotl-ai-cloud.github.io/axolotl/index.html</loc>
+    <lastmod>2025-03-05T15:02:20.727Z</lastmod>
+  </url>
+  <url>
+    <loc>https://axolotl-ai-cloud.github.io/axolotl/src/axolotl/integrations/cut_cross_entropy/ACKNOWLEDGEMENTS.html</loc>
+    <lastmod>2025-03-05T15:02:20.730Z</lastmod>
+  </url>
+  <url>
+    <loc>https://axolotl-ai-cloud.github.io/axolotl/FAQS.html</loc>
+    <lastmod>2025-03-05T15:02:20.710Z</lastmod>
+  </url>
+  <url>
+    <loc>https://axolotl-ai-cloud.github.io/axolotl/docs/multi-node.html</loc>
+    <lastmod>2025-03-05T15:02:20.715Z</lastmod>
+  </url>
+  <url>
+    <loc>https://axolotl-ai-cloud.github.io/axolotl/docs/faq.html</loc>
+    <lastmod>2025-03-05T15:02:20.712Z</lastmod>
+  </url>
+  <url>
+    <loc>https://axolotl-ai-cloud.github.io/axolotl/docs/batch_vs_grad.html</loc>
+    <lastmod>2025-03-05T15:02:20.711Z</lastmod>
+  </url>
+  <url>
+    <loc>https://axolotl-ai-cloud.github.io/axolotl/docs/lora_optims.html</loc>
+    <lastmod>2025-03-05T15:02:20.715Z</lastmod>
+  </url>
+  <url>
+    <loc>https://axolotl-ai-cloud.github.io/axolotl/docs/rlhf.html</loc>
+    <lastmod>2025-03-05T15:02:20.715Z</lastmod>
+  </url>
+  <url>
+    <loc>https://axolotl-ai-cloud.github.io/axolotl/docs/cli.html</loc>
+    <lastmod>2025-03-05T15:02:20.711Z</lastmod>
+  </url>
+  <url>
+    <loc>https://axolotl-ai-cloud.github.io/axolotl/docs/unsloth.html</loc>
+    <lastmod>2025-03-05T15:02:20.715Z</lastmod>
+  </url>
+  <url>
+    <loc>https://axolotl-ai-cloud.github.io/axolotl/docs/fsdp_qlora.html</loc>
+    <lastmod>2025-03-05T15:02:20.712Z</lastmod>
+  </url>
+  <url>
+    <loc>https://axolotl-ai-cloud.github.io/axolotl/docs/dataset_preprocessing.html</loc>
+    <lastmod>2025-03-05T15:02:20.712Z</lastmod>
+  </url>
+  <url>
+    <loc>https://axolotl-ai-cloud.github.io/axolotl/docs/custom_integrations.html</loc>
+    <lastmod>2025-03-05T15:02:20.711Z</lastmod>
+  </url>
+  <url>
+    <loc>https://axolotl-ai-cloud.github.io/axolotl/docs/mac.html</loc>
+    <lastmod>2025-03-05T15:02:20.715Z</lastmod>
+  </url>
+  <url>
+    <loc>https://axolotl-ai-cloud.github.io/axolotl/docs/docker.html</loc>
+    <lastmod>2025-03-05T15:02:20.712Z</lastmod>
  </url>
  <url>
    <loc>https://axolotl-ai-cloud.github.io/axolotl/docs/ray-integration.html</loc>
-    <lastmod>2025-03-05T13:58:47.517Z</lastmod>
+    <lastmod>2025-03-05T15:02:20.715Z</lastmod>
  </url>
  <url>
    <loc>https://axolotl-ai-cloud.github.io/axolotl/docs/dataset-formats/index.html</loc>
-    <lastmod>2025-03-05T13:58:47.514Z</lastmod>
+    <lastmod>2025-03-05T15:02:20.712Z</lastmod>
  </url>
  <url>
    <loc>https://axolotl-ai-cloud.github.io/axolotl/docs/dataset-formats/conversation.html</loc>
-    <lastmod>2025-03-05T13:58:47.514Z</lastmod>
+    <lastmod>2025-03-05T15:02:20.711Z</lastmod>
  </url>
  <url>
    <loc>https://axolotl-ai-cloud.github.io/axolotl/docs/dataset-formats/pretraining.html</loc>
-    <lastmod>2025-03-05T13:58:47.514Z</lastmod>
+    <lastmod>2025-03-05T15:02:20.712Z</lastmod>
  </url>
  <url>
    <loc>https://axolotl-ai-cloud.github.io/axolotl/docs/dataset-formats/inst_tune.html</loc>
-    <lastmod>2025-03-05T13:58:47.514Z</lastmod>
+    <lastmod>2025-03-05T15:02:20.712Z</lastmod>
  </url>
 </urlset>
--- a/src/axolotl/integrations/LICENSE.html
+++ b/src/axolotl/integrations/LICENSE.html
@@ -206,6 +206,12 @@ ul.task-list li input[type="checkbox"] {
      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
          <li class="sidebar-item">
  <div class="sidebar-item-container"> 
+  <a href="../../../docs/docker.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Docker</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
  <a href="../../../docs/multi-gpu.html" class="sidebar-item-text sidebar-link">
 <span class="menu-text">Multi-GPU</span></a>
  </div>
--- a/src/axolotl/integrations/cut_cross_entropy/ACKNOWLEDGEMENTS.html
+++ b/src/axolotl/integrations/cut_cross_entropy/ACKNOWLEDGEMENTS.html
@@ -206,6 +206,12 @@ ul.task-list li input[type="checkbox"] {
      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
          <li class="sidebar-item">
  <div class="sidebar-item-container"> 
+  <a href="../../../../docs/docker.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Docker</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
  <a href="../../../../docs/multi-gpu.html" class="sidebar-item-text sidebar-link">
 <span class="menu-text">Multi-GPU</span></a>
  </div>
@@ -1 +1 @@
 c1dbc37
 d357