Built site for gh-pages
This commit is contained in:
Quarto GHA Workflow Runner
4
.github/workflows/main.yml
vendored
4
.github/workflows/main.yml
vendored
@@ -25,12 +25,12 @@ jobs:
|
||||
python_version: "3.11"
|
||||
pytorch: 2.5.1
|
||||
axolotl_extras: vllm
|
||||
is_latest: true
|
||||
- cuda: 124
|
||||
cuda_version: 12.4.1
|
||||
python_version: "3.11"
|
||||
pytorch: 2.6.0
|
||||
axolotl_extras:
|
||||
is_latest: true
|
||||
runs-on: axolotl-gpu-runner
|
||||
steps:
|
||||
- name: Checkout
|
||||
@@ -87,12 +87,12 @@ jobs:
|
||||
python_version: "3.11"
|
||||
pytorch: 2.5.1
|
||||
axolotl_extras:
|
||||
is_latest: true
|
||||
- cuda: 124
|
||||
cuda_version: 12.4.1
|
||||
python_version: "3.11"
|
||||
pytorch: 2.6.0
|
||||
axolotl_extras:
|
||||
is_latest: true
|
||||
runs-on: axolotl-gpu-runner
|
||||
steps:
|
||||
- name: Checkout
|
||||
|
||||
@@ -602,7 +602,7 @@ the CLI commands, their usage, and common examples.</p>
|
||||
</section>
|
||||
<section id="evaluate" class="level3">
|
||||
<h3 class="anchored" data-anchor-id="evaluate">evaluate</h3>
|
||||
<p>Evaluates a model’s performance using metrics specified in the config.</p>
|
||||
<p>Evaluates a model’s performance (loss etc) on the train and eval datasets.</p>
|
||||
<div class="sourceCode" id="cb12"><pre class="sourceCode bash code-with-copy"><code class="sourceCode bash"><span id="cb12-1"><a href="#cb12-1" aria-hidden="true" tabindex="-1"></a><span class="co"># Basic evaluation</span></span>
|
||||
<span id="cb12-2"><a href="#cb12-2" aria-hidden="true" tabindex="-1"></a><span class="ex">axolotl</span> evaluate config.yml</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
|
||||
</section>
|
||||
@@ -618,6 +618,7 @@ the CLI commands, their usage, and common examples.</p>
|
||||
<span id="cb14-4"><a href="#cb14-4" aria-hidden="true" tabindex="-1"></a><span class="at"> </span><span class="kw">-</span><span class="at"> hellaswag</span></span>
|
||||
<span id="cb14-5"><a href="#cb14-5" 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="cb14-6"><a href="#cb14-6" 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>
|
||||
<p>See <a href="https://github.com/EleutherAI/lm-evaluation-harness">LM Eval Harness</a> for more details.</p>
|
||||
</section>
|
||||
</section>
|
||||
<section id="legacy-cli-usage" class="level2">
|
||||
@@ -660,7 +661,7 @@ cloud YAML file alongside your regular Axolotl config.</p>
|
||||
<p>Create a cloud config YAML with your Modal settings:</p>
|
||||
<div class="sourceCode" id="cb16"><pre class="sourceCode yaml code-with-copy"><code class="sourceCode yaml"><span id="cb16-1"><a href="#cb16-1" aria-hidden="true" tabindex="-1"></a><span class="co"># cloud_config.yml</span></span>
|
||||
<span id="cb16-2"><a href="#cb16-2" aria-hidden="true" tabindex="-1"></a><span class="fu">provider</span><span class="kw">:</span><span class="at"> modal</span></span>
|
||||
<span id="cb16-3"><a href="#cb16-3" aria-hidden="true" tabindex="-1"></a><span class="fu">gpu</span><span class="kw">:</span><span class="at"> a100</span><span class="co"> # Supported: l40s, a100-40gb, a100-80gb, a10g, h100, t4, l4</span></span>
|
||||
<span id="cb16-3"><a href="#cb16-3" aria-hidden="true" tabindex="-1"></a><span class="fu">gpu</span><span class="kw">:</span><span class="at"> a100</span><span class="co"> # Supported: l40s, a100-40gb, a100-80gb, a10g, h100, t4, l4</span></span>
|
||||
<span id="cb16-4"><a href="#cb16-4" aria-hidden="true" tabindex="-1"></a><span class="fu">gpu_count</span><span class="kw">:</span><span class="at"> </span><span class="dv">1</span><span class="co"> # Number of GPUs to use</span></span>
|
||||
<span id="cb16-5"><a href="#cb16-5" aria-hidden="true" tabindex="-1"></a><span class="fu">timeout</span><span class="kw">:</span><span class="at"> </span><span class="dv">86400</span><span class="co"> # Maximum runtime in seconds (24 hours)</span></span>
|
||||
<span id="cb16-6"><a href="#cb16-6" aria-hidden="true" tabindex="-1"></a><span class="fu">branch</span><span class="kw">:</span><span class="at"> main</span><span class="co"> # Git branch to use (optional)</span></span>
|
||||
@@ -673,7 +674,7 @@ cloud YAML file alongside your regular Axolotl config.</p>
|
||||
<span id="cb16-13"><a href="#cb16-13" aria-hidden="true" tabindex="-1"></a><span class="at"> </span><span class="kw">-</span><span class="at"> </span><span class="fu">name</span><span class="kw">:</span><span class="at"> axolotl-artifacts</span></span>
|
||||
<span id="cb16-14"><a href="#cb16-14" aria-hidden="true" tabindex="-1"></a><span class="at"> </span><span class="fu">mount</span><span class="kw">:</span><span class="at"> /workspace/artifacts</span></span>
|
||||
<span id="cb16-15"><a href="#cb16-15" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb16-16"><a href="#cb16-16" aria-hidden="true" tabindex="-1"></a><span class="fu">env</span><span class="kw">:</span><span class="co"> # Environment variables</span></span>
|
||||
<span id="cb16-16"><a href="#cb16-16" aria-hidden="true" tabindex="-1"></a><span class="fu">secrets</span><span class="kw">:</span><span class="co"> # Secrets to inject</span></span>
|
||||
<span id="cb16-17"><a href="#cb16-17" aria-hidden="true" tabindex="-1"></a><span class="at"> </span><span class="kw">-</span><span class="at"> WANDB_API_KEY</span></span>
|
||||
<span id="cb16-18"><a href="#cb16-18" aria-hidden="true" tabindex="-1"></a><span class="at"> </span><span class="kw">-</span><span class="at"> HF_TOKEN</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
|
||||
</section>
|
||||
@@ -694,17 +695,29 @@ cloud YAML file alongside your regular Axolotl config.</p>
|
||||
</section>
|
||||
<section id="cloud-configuration-options" class="level3">
|
||||
<h3 class="anchored" data-anchor-id="cloud-configuration-options">Cloud Configuration Options</h3>
|
||||
<div class="sourceCode" id="cb18"><pre class="sourceCode yaml code-with-copy"><code class="sourceCode yaml"><span id="cb18-1"><a href="#cb18-1" aria-hidden="true" tabindex="-1"></a><span class="fu">provider</span><span class="kw">:</span><span class="co"> # compute provider, currently only `modal` is supported</span></span>
|
||||
<span id="cb18-2"><a href="#cb18-2" aria-hidden="true" tabindex="-1"></a><span class="fu">gpu</span><span class="kw">:</span><span class="co"> # GPU type to use</span></span>
|
||||
<span id="cb18-3"><a href="#cb18-3" aria-hidden="true" tabindex="-1"></a><span class="fu">gpu_count</span><span class="kw">:</span><span class="co"> # Number of GPUs (default: 1)</span></span>
|
||||
<span id="cb18-4"><a href="#cb18-4" aria-hidden="true" tabindex="-1"></a><span class="fu">memory</span><span class="kw">:</span><span class="co"> # RAM in GB (default: 128)</span></span>
|
||||
<span id="cb18-5"><a href="#cb18-5" aria-hidden="true" tabindex="-1"></a><span class="fu">timeout</span><span class="kw">:</span><span class="co"> # Maximum runtime in seconds</span></span>
|
||||
<div class="sourceCode" id="cb18"><pre class="sourceCode yaml code-with-copy"><code class="sourceCode yaml"><span id="cb18-1"><a href="#cb18-1" aria-hidden="true" tabindex="-1"></a><span class="fu">provider</span><span class="kw">:</span><span class="co"> # compute provider, currently only `modal` is supported</span></span>
|
||||
<span id="cb18-2"><a href="#cb18-2" aria-hidden="true" tabindex="-1"></a><span class="fu">gpu</span><span class="kw">:</span><span class="co"> # GPU type to use</span></span>
|
||||
<span id="cb18-3"><a href="#cb18-3" aria-hidden="true" tabindex="-1"></a><span class="fu">gpu_count</span><span class="kw">:</span><span class="co"> # Number of GPUs (default: 1)</span></span>
|
||||
<span id="cb18-4"><a href="#cb18-4" aria-hidden="true" tabindex="-1"></a><span class="fu">memory</span><span class="kw">:</span><span class="co"> # RAM in GB (default: 128)</span></span>
|
||||
<span id="cb18-5"><a href="#cb18-5" aria-hidden="true" tabindex="-1"></a><span class="fu">timeout</span><span class="kw">:</span><span class="co"> # Maximum runtime in seconds</span></span>
|
||||
<span id="cb18-6"><a href="#cb18-6" aria-hidden="true" tabindex="-1"></a><span class="fu">timeout_preprocess</span><span class="kw">:</span><span class="co"> # Preprocessing timeout</span></span>
|
||||
<span id="cb18-7"><a href="#cb18-7" aria-hidden="true" tabindex="-1"></a><span class="fu">branch</span><span class="kw">:</span><span class="co"> # Git branch to use</span></span>
|
||||
<span id="cb18-8"><a href="#cb18-8" aria-hidden="true" tabindex="-1"></a><span class="fu">docker_tag</span><span class="kw">:</span><span class="co"> # Custom Docker image tag</span></span>
|
||||
<span id="cb18-9"><a href="#cb18-9" aria-hidden="true" tabindex="-1"></a><span class="fu">volumes</span><span class="kw">:</span><span class="co"> # List of persistent storage volumes</span></span>
|
||||
<span id="cb18-10"><a href="#cb18-10" aria-hidden="true" tabindex="-1"></a><span class="fu">env</span><span class="kw">:</span><span class="co"> # Environment variables to pass</span></span>
|
||||
<span id="cb18-11"><a href="#cb18-11" aria-hidden="true" tabindex="-1"></a><span class="fu">secrets</span><span class="kw">:</span><span class="co"> # Secrets to inject</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
|
||||
<span id="cb18-7"><a href="#cb18-7" aria-hidden="true" tabindex="-1"></a><span class="fu">branch</span><span class="kw">:</span><span class="co"> # Git branch to use</span></span>
|
||||
<span id="cb18-8"><a href="#cb18-8" aria-hidden="true" tabindex="-1"></a><span class="fu">docker_tag</span><span class="kw">:</span><span class="co"> # Custom Docker image tag</span></span>
|
||||
<span id="cb18-9"><a href="#cb18-9" aria-hidden="true" tabindex="-1"></a><span class="fu">volumes</span><span class="kw">:</span><span class="co"> # List of persistent storage volumes</span></span>
|
||||
<span id="cb18-10"><a href="#cb18-10" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb18-11"><a href="#cb18-11" aria-hidden="true" tabindex="-1"></a><span class="co"># Environment variables to pass. Can be specified in two ways:</span></span>
|
||||
<span id="cb18-12"><a href="#cb18-12" aria-hidden="true" tabindex="-1"></a><span class="co"># 1. As a string: Will load the value from the host computer's environment variables</span></span>
|
||||
<span id="cb18-13"><a href="#cb18-13" aria-hidden="true" tabindex="-1"></a><span class="co"># 2. As a key-value pair: Will use the specified value directly</span></span>
|
||||
<span id="cb18-14"><a href="#cb18-14" aria-hidden="true" tabindex="-1"></a><span class="co"># Example:</span></span>
|
||||
<span id="cb18-15"><a href="#cb18-15" aria-hidden="true" tabindex="-1"></a><span class="co"># env:</span></span>
|
||||
<span id="cb18-16"><a href="#cb18-16" aria-hidden="true" tabindex="-1"></a><span class="co"># - CUSTOM_VAR # Loads from host's $CUSTOM_VAR</span></span>
|
||||
<span id="cb18-17"><a href="#cb18-17" aria-hidden="true" tabindex="-1"></a><span class="co"># - {CUSTOM_VAR: "value"} # Uses "value" directly</span></span>
|
||||
<span id="cb18-18"><a href="#cb18-18" aria-hidden="true" tabindex="-1"></a><span class="fu">env</span><span class="kw">:</span></span>
|
||||
<span id="cb18-19"><a href="#cb18-19" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb18-20"><a href="#cb18-20" aria-hidden="true" tabindex="-1"></a><span class="co"># Secrets to inject. Same input format as `env` but for sensitive data.</span></span>
|
||||
<span id="cb18-21"><a href="#cb18-21" aria-hidden="true" tabindex="-1"></a><span class="fu">secrets</span><span class="kw">:</span></span>
|
||||
<span id="cb18-22"><a href="#cb18-22" aria-hidden="true" tabindex="-1"></a><span class="co"> # - HF_TOKEN</span></span>
|
||||
<span id="cb18-23"><a href="#cb18-23" aria-hidden="true" tabindex="-1"></a><span class="co"> # - WANDB_API_KEY</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
|
||||
|
||||
|
||||
</section>
|
||||
|
||||
738
docs/config.html
738
docs/config.html
@@ -780,367 +780,391 @@ pre > code.sourceCode > span > a:first-child::before { text-decoration: underlin
|
||||
<span id="cb1-314"><a href="#cb1-314" 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="cb1-315"><a href="#cb1-315" 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="cb1-316"><a href="#cb1-316" 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="cb1-317"><a href="#cb1-317" aria-hidden="true" tabindex="-1"></a><span class="co"># whether to concatenate samples during pretraining</span></span>
|
||||
<span id="cb1-318"><a href="#cb1-318" aria-hidden="true" tabindex="-1"></a><span class="fu">pretraining_sample_concatenation</span><span class="kw">:</span></span>
|
||||
<span id="cb1-319"><a href="#cb1-319" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-320"><a href="#cb1-320" aria-hidden="true" tabindex="-1"></a><span class="co"># Use batch flattening for speedups when not using sample_packing</span></span>
|
||||
<span id="cb1-321"><a href="#cb1-321" aria-hidden="true" tabindex="-1"></a><span class="fu">batch_flattening</span><span class="kw">:</span></span>
|
||||
<span id="cb1-322"><a href="#cb1-322" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-323"><a href="#cb1-323" 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="cb1-324"><a href="#cb1-324" aria-hidden="true" tabindex="-1"></a><span class="co"># Use `sequential` when training w/ model parallelism to limit memory</span></span>
|
||||
<span id="cb1-325"><a href="#cb1-325" aria-hidden="true" tabindex="-1"></a><span class="fu">device_map</span><span class="kw">:</span></span>
|
||||
<span id="cb1-326"><a href="#cb1-326" 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="cb1-327"><a href="#cb1-327" aria-hidden="true" tabindex="-1"></a><span class="fu">max_memory</span><span class="kw">:</span></span>
|
||||
<span id="cb1-328"><a href="#cb1-328" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-329"><a href="#cb1-329" 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="cb1-330"><a href="#cb1-330" aria-hidden="true" tabindex="-1"></a><span class="fu">adapter</span><span class="kw">:</span><span class="at"> lora</span></span>
|
||||
<span id="cb1-331"><a href="#cb1-331" 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="cb1-332"><a href="#cb1-332" 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="cb1-333"><a href="#cb1-333" 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="cb1-334"><a href="#cb1-334" aria-hidden="true" tabindex="-1"></a><span class="fu">lora_model_dir</span><span class="kw">:</span></span>
|
||||
<span id="cb1-335"><a href="#cb1-335" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-336"><a href="#cb1-336" aria-hidden="true" tabindex="-1"></a><span class="co"># LoRA hyperparameters</span></span>
|
||||
<span id="cb1-337"><a href="#cb1-337" aria-hidden="true" tabindex="-1"></a><span class="co"># For more details about the following options, see:</span></span>
|
||||
<span id="cb1-338"><a href="#cb1-338" 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="cb1-339"><a href="#cb1-339" 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="cb1-340"><a href="#cb1-340" 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="cb1-341"><a href="#cb1-341" 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="cb1-342"><a href="#cb1-342" aria-hidden="true" tabindex="-1"></a><span class="fu">lora_target_modules</span><span class="kw">:</span></span>
|
||||
<span id="cb1-343"><a href="#cb1-343" aria-hidden="true" tabindex="-1"></a><span class="at"> </span><span class="kw">-</span><span class="at"> q_proj</span></span>
|
||||
<span id="cb1-344"><a href="#cb1-344" aria-hidden="true" tabindex="-1"></a><span class="at"> </span><span class="kw">-</span><span class="at"> v_proj</span></span>
|
||||
<span id="cb1-345"><a href="#cb1-345" aria-hidden="true" tabindex="-1"></a><span class="co"># - k_proj</span></span>
|
||||
<span id="cb1-346"><a href="#cb1-346" aria-hidden="true" tabindex="-1"></a><span class="co"># - o_proj</span></span>
|
||||
<span id="cb1-347"><a href="#cb1-347" aria-hidden="true" tabindex="-1"></a><span class="co"># - gate_proj</span></span>
|
||||
<span id="cb1-348"><a href="#cb1-348" aria-hidden="true" tabindex="-1"></a><span class="co"># - down_proj</span></span>
|
||||
<span id="cb1-349"><a href="#cb1-349" aria-hidden="true" tabindex="-1"></a><span class="co"># - up_proj</span></span>
|
||||
<span id="cb1-350"><a href="#cb1-350" 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="cb1-351"><a href="#cb1-351" 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="cb1-352"><a href="#cb1-352" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-353"><a href="#cb1-353" 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="cb1-354"><a href="#cb1-354" 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="cb1-355"><a href="#cb1-355" 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="cb1-356"><a href="#cb1-356" aria-hidden="true" tabindex="-1"></a><span class="co"># https://github.com/huggingface/peft/issues/334#issuecomment-1561727994</span></span>
|
||||
<span id="cb1-357"><a href="#cb1-357" aria-hidden="true" tabindex="-1"></a><span class="fu">lora_modules_to_save</span><span class="kw">:</span></span>
|
||||
<span id="cb1-358"><a href="#cb1-358" aria-hidden="true" tabindex="-1"></a><span class="co"># - embed_tokens</span></span>
|
||||
<span id="cb1-359"><a href="#cb1-359" aria-hidden="true" tabindex="-1"></a><span class="co"># - lm_head</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="fu">lora_fan_in_fan_out</span><span class="kw">:</span><span class="at"> </span><span class="ch">false</span></span>
|
||||
<span id="cb1-362"><a href="#cb1-362" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-363"><a href="#cb1-363" aria-hidden="true" tabindex="-1"></a><span class="co"># Apply custom LoRA autograd functions and activation function Triton kernels for</span></span>
|
||||
<span id="cb1-364"><a href="#cb1-364" aria-hidden="true" tabindex="-1"></a><span class="co"># speed and memory savings</span></span>
|
||||
<span id="cb1-365"><a href="#cb1-365" 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="cb1-366"><a href="#cb1-366" 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="cb1-367"><a href="#cb1-367" 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="cb1-368"><a href="#cb1-368" 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="cb1-369"><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"># LoRA+ hyperparameters</span></span>
|
||||
<span id="cb1-371"><a href="#cb1-371" aria-hidden="true" tabindex="-1"></a><span class="co"># For more details about the following options, see:</span></span>
|
||||
<span id="cb1-372"><a href="#cb1-372" 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="cb1-373"><a href="#cb1-373" 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="cb1-374"><a href="#cb1-374" 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="cb1-375"><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="fu">peft</span><span class="kw">:</span></span>
|
||||
<span id="cb1-377"><a href="#cb1-377" aria-hidden="true" tabindex="-1"></a><span class="co"> # Configuration options for loftq initialization for LoRA</span></span>
|
||||
<span id="cb1-378"><a href="#cb1-378" aria-hidden="true" tabindex="-1"></a><span class="co"> # https://huggingface.co/docs/peft/developer_guides/quantization#loftq-initialization</span></span>
|
||||
<span id="cb1-379"><a href="#cb1-379" aria-hidden="true" tabindex="-1"></a><span class="at"> </span><span class="fu">loftq_config</span><span class="kw">:</span></span>
|
||||
<span id="cb1-380"><a href="#cb1-380" 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="cb1-381"><a href="#cb1-381" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-382"><a href="#cb1-382" aria-hidden="true" tabindex="-1"></a><span class="co"># ReLoRA configuration</span></span>
|
||||
<span id="cb1-383"><a href="#cb1-383" 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="cb1-384"><a href="#cb1-384" 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="cb1-385"><a href="#cb1-385" 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="cb1-386"><a href="#cb1-386" 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="cb1-387"><a href="#cb1-387" 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="cb1-388"><a href="#cb1-388" 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="cb1-389"><a href="#cb1-389" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-390"><a href="#cb1-390" aria-hidden="true" tabindex="-1"></a><span class="co"># wandb configuration if you're using it</span></span>
|
||||
<span id="cb1-391"><a href="#cb1-391" 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="cb1-392"><a href="#cb1-392" 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="cb1-393"><a href="#cb1-393" 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="cb1-394"><a href="#cb1-394" 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="cb1-395"><a href="#cb1-395" aria-hidden="true" tabindex="-1"></a><span class="fu">wandb_watch</span><span class="kw">:</span></span>
|
||||
<span id="cb1-396"><a href="#cb1-396" 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="cb1-397"><a href="#cb1-397" 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="cb1-398"><a href="#cb1-398" 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="cb1-317"><a href="#cb1-317" aria-hidden="true" tabindex="-1"></a><span class="fu">sample_pack_sequentially</span><span class="kw">:</span><span class="co"> # Optional[bool]. Whether to pack samples sequentially.</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"># whether to concatenate samples during pretraining</span></span>
|
||||
<span id="cb1-320"><a href="#cb1-320" aria-hidden="true" tabindex="-1"></a><span class="fu">pretraining_sample_concatenation</span><span class="kw">:</span></span>
|
||||
<span id="cb1-321"><a href="#cb1-321" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-322"><a href="#cb1-322" aria-hidden="true" tabindex="-1"></a><span class="fu">curriculum_sampling</span><span class="kw">:</span><span class="co"> # Optional[bool]. Whether to use sequential sampling for curriculum learning</span></span>
|
||||
<span id="cb1-323"><a href="#cb1-323" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-324"><a href="#cb1-324" aria-hidden="true" tabindex="-1"></a><span class="co"># Use batch flattening for speedups when not using sample_packing</span></span>
|
||||
<span id="cb1-325"><a href="#cb1-325" aria-hidden="true" tabindex="-1"></a><span class="fu">batch_flattening</span><span class="kw">:</span></span>
|
||||
<span id="cb1-326"><a href="#cb1-326" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-327"><a href="#cb1-327" 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="cb1-328"><a href="#cb1-328" aria-hidden="true" tabindex="-1"></a><span class="co"># Use `sequential` when training w/ model parallelism to limit memory</span></span>
|
||||
<span id="cb1-329"><a href="#cb1-329" aria-hidden="true" tabindex="-1"></a><span class="fu">device_map</span><span class="kw">:</span></span>
|
||||
<span id="cb1-330"><a href="#cb1-330" 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="cb1-331"><a href="#cb1-331" aria-hidden="true" tabindex="-1"></a><span class="fu">max_memory</span><span class="kw">:</span></span>
|
||||
<span id="cb1-332"><a href="#cb1-332" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-333"><a href="#cb1-333" 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="cb1-334"><a href="#cb1-334" aria-hidden="true" tabindex="-1"></a><span class="fu">adapter</span><span class="kw">:</span><span class="at"> lora</span></span>
|
||||
<span id="cb1-335"><a href="#cb1-335" 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="cb1-336"><a href="#cb1-336" 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="cb1-337"><a href="#cb1-337" 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="cb1-338"><a href="#cb1-338" aria-hidden="true" tabindex="-1"></a><span class="fu">lora_model_dir</span><span class="kw">:</span></span>
|
||||
<span id="cb1-339"><a href="#cb1-339" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-340"><a href="#cb1-340" aria-hidden="true" tabindex="-1"></a><span class="co"># LoRA hyperparameters</span></span>
|
||||
<span id="cb1-341"><a href="#cb1-341" aria-hidden="true" tabindex="-1"></a><span class="co"># For more details about the following options, see:</span></span>
|
||||
<span id="cb1-342"><a href="#cb1-342" 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="cb1-343"><a href="#cb1-343" 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="cb1-344"><a href="#cb1-344" 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="cb1-345"><a href="#cb1-345" 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="cb1-346"><a href="#cb1-346" aria-hidden="true" tabindex="-1"></a><span class="fu">lora_target_modules</span><span class="kw">:</span></span>
|
||||
<span id="cb1-347"><a href="#cb1-347" aria-hidden="true" tabindex="-1"></a><span class="at"> </span><span class="kw">-</span><span class="at"> q_proj</span></span>
|
||||
<span id="cb1-348"><a href="#cb1-348" aria-hidden="true" tabindex="-1"></a><span class="at"> </span><span class="kw">-</span><span class="at"> v_proj</span></span>
|
||||
<span id="cb1-349"><a href="#cb1-349" aria-hidden="true" tabindex="-1"></a><span class="co"># - k_proj</span></span>
|
||||
<span id="cb1-350"><a href="#cb1-350" aria-hidden="true" tabindex="-1"></a><span class="co"># - o_proj</span></span>
|
||||
<span id="cb1-351"><a href="#cb1-351" aria-hidden="true" tabindex="-1"></a><span class="co"># - gate_proj</span></span>
|
||||
<span id="cb1-352"><a href="#cb1-352" aria-hidden="true" tabindex="-1"></a><span class="co"># - down_proj</span></span>
|
||||
<span id="cb1-353"><a href="#cb1-353" aria-hidden="true" tabindex="-1"></a><span class="co"># - up_proj</span></span>
|
||||
<span id="cb1-354"><a href="#cb1-354" 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="cb1-355"><a href="#cb1-355" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-356"><a href="#cb1-356" aria-hidden="true" tabindex="-1"></a><span class="co"># List[int] | int. # The layer indices to transform, otherwise, apply to all layers</span></span>
|
||||
<span id="cb1-357"><a href="#cb1-357" aria-hidden="true" tabindex="-1"></a><span class="co"># https://huggingface.co/docs/peft/v0.15.0/en/package_reference/lora#peft.LoraConfig.layers_to_transform</span></span>
|
||||
<span id="cb1-358"><a href="#cb1-358" aria-hidden="true" tabindex="-1"></a><span class="fu">peft_layers_to_transform</span><span class="kw">:</span></span>
|
||||
<span id="cb1-359"><a href="#cb1-359" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-360"><a href="#cb1-360" aria-hidden="true" tabindex="-1"></a><span class="co"># Optional[bool]. Whether to use DoRA.</span></span>
|
||||
<span id="cb1-361"><a href="#cb1-361" aria-hidden="true" tabindex="-1"></a><span class="co"># https://huggingface.co/docs/peft/v0.15.0/en/developer_guides/lora#weight-decomposed-low-rank-adaptation-dora</span></span>
|
||||
<span id="cb1-362"><a href="#cb1-362" aria-hidden="true" tabindex="-1"></a><span class="fu">peft_use_dora</span><span class="kw">:</span></span>
|
||||
<span id="cb1-363"><a href="#cb1-363" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-364"><a href="#cb1-364" aria-hidden="true" tabindex="-1"></a><span class="co"># Optional[bool]. Whether to use RSLoRA.</span></span>
|
||||
<span id="cb1-365"><a href="#cb1-365" aria-hidden="true" tabindex="-1"></a><span class="co"># https://huggingface.co/docs/peft/v0.15.0/en/developer_guides/lora#rank-stabilized-lora</span></span>
|
||||
<span id="cb1-366"><a href="#cb1-366" aria-hidden="true" tabindex="-1"></a><span class="fu">peft_use_rslora</span><span class="kw">:</span></span>
|
||||
<span id="cb1-367"><a href="#cb1-367" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-368"><a href="#cb1-368" aria-hidden="true" tabindex="-1"></a><span class="co"># Optional[list[tuple[int, int]]]. List of layer indices to replicate.</span></span>
|
||||
<span id="cb1-369"><a href="#cb1-369" aria-hidden="true" tabindex="-1"></a><span class="co"># https://huggingface.co/docs/peft/v0.15.0/en/developer_guides/lora#memory-efficient-layer-replication-with-lora</span></span>
|
||||
<span id="cb1-370"><a href="#cb1-370" aria-hidden="true" tabindex="-1"></a><span class="fu">peft_layer_replication</span><span class="kw">:</span></span>
|
||||
<span id="cb1-371"><a href="#cb1-371" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-372"><a href="#cb1-372" aria-hidden="true" tabindex="-1"></a><span class="co"># bool | Literal["gaussian", "eva", "olora", "pissa", "pissa_niter_[number of iters]", "corda", "loftq"]</span></span>
|
||||
<span id="cb1-373"><a href="#cb1-373" aria-hidden="true" tabindex="-1"></a><span class="co"># How to initialize LoRA weights. Default to True which is MS original implementation.</span></span>
|
||||
<span id="cb1-374"><a href="#cb1-374" aria-hidden="true" tabindex="-1"></a><span class="co"># https://huggingface.co/docs/peft/v0.15.0/en/developer_guides/lora#initialization</span></span>
|
||||
<span id="cb1-375"><a href="#cb1-375" aria-hidden="true" tabindex="-1"></a><span class="fu">peft_init_lora_weights</span><span class="kw">:</span></span>
|
||||
<span id="cb1-376"><a href="#cb1-376" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-377"><a href="#cb1-377" 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="cb1-378"><a href="#cb1-378" 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="cb1-379"><a href="#cb1-379" 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="cb1-380"><a href="#cb1-380" aria-hidden="true" tabindex="-1"></a><span class="co"># https://github.com/huggingface/peft/issues/334#issuecomment-1561727994</span></span>
|
||||
<span id="cb1-381"><a href="#cb1-381" aria-hidden="true" tabindex="-1"></a><span class="fu">lora_modules_to_save</span><span class="kw">:</span></span>
|
||||
<span id="cb1-382"><a href="#cb1-382" aria-hidden="true" tabindex="-1"></a><span class="co"># - embed_tokens</span></span>
|
||||
<span id="cb1-383"><a href="#cb1-383" aria-hidden="true" tabindex="-1"></a><span class="co"># - lm_head</span></span>
|
||||
<span id="cb1-384"><a href="#cb1-384" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-385"><a href="#cb1-385" 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="cb1-386"><a href="#cb1-386" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-387"><a href="#cb1-387" aria-hidden="true" tabindex="-1"></a><span class="co"># Apply custom LoRA autograd functions and activation function Triton kernels for</span></span>
|
||||
<span id="cb1-388"><a href="#cb1-388" aria-hidden="true" tabindex="-1"></a><span class="co"># speed and memory savings</span></span>
|
||||
<span id="cb1-389"><a href="#cb1-389" 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="cb1-390"><a href="#cb1-390" 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="cb1-391"><a href="#cb1-391" 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="cb1-392"><a href="#cb1-392" 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="cb1-393"><a href="#cb1-393" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-394"><a href="#cb1-394" aria-hidden="true" tabindex="-1"></a><span class="co"># LoRA+ hyperparameters</span></span>
|
||||
<span id="cb1-395"><a href="#cb1-395" aria-hidden="true" tabindex="-1"></a><span class="co"># For more details about the following options, see:</span></span>
|
||||
<span id="cb1-396"><a href="#cb1-396" 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="cb1-397"><a href="#cb1-397" 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="cb1-398"><a href="#cb1-398" 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="cb1-399"><a href="#cb1-399" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-400"><a href="#cb1-400" aria-hidden="true" tabindex="-1"></a><span class="co"># mlflow configuration if you're using it</span></span>
|
||||
<span id="cb1-401"><a href="#cb1-401" 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="cb1-402"><a href="#cb1-402" 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="cb1-403"><a href="#cb1-403" 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="cb1-404"><a href="#cb1-404" 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="cb1-400"><a href="#cb1-400" aria-hidden="true" tabindex="-1"></a><span class="fu">peft</span><span class="kw">:</span></span>
|
||||
<span id="cb1-401"><a href="#cb1-401" aria-hidden="true" tabindex="-1"></a><span class="co"> # Configuration options for loftq initialization for LoRA</span></span>
|
||||
<span id="cb1-402"><a href="#cb1-402" aria-hidden="true" tabindex="-1"></a><span class="co"> # https://huggingface.co/docs/peft/developer_guides/quantization#loftq-initialization</span></span>
|
||||
<span id="cb1-403"><a href="#cb1-403" aria-hidden="true" tabindex="-1"></a><span class="at"> </span><span class="fu">loftq_config</span><span class="kw">:</span></span>
|
||||
<span id="cb1-404"><a href="#cb1-404" 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="cb1-405"><a href="#cb1-405" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-406"><a href="#cb1-406" aria-hidden="true" tabindex="-1"></a><span class="co"># Comet configuration if you're using it</span></span>
|
||||
<span id="cb1-407"><a href="#cb1-407" 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="cb1-408"><a href="#cb1-408" 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="cb1-409"><a href="#cb1-409" 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="cb1-410"><a href="#cb1-410" 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="cb1-411"><a href="#cb1-411" 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="cb1-412"><a href="#cb1-412" 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="cb1-413"><a href="#cb1-413" 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="cb1-414"><a href="#cb1-414" 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="cb1-415"><a href="#cb1-415" 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="cb1-416"><a href="#cb1-416" 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="cb1-417"><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="co"># Tensorboard</span></span>
|
||||
<span id="cb1-419"><a href="#cb1-419" 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="cb1-420"><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"># Where to save the full-finetuned model to</span></span>
|
||||
<span id="cb1-422"><a href="#cb1-422" 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="cb1-406"><a href="#cb1-406" aria-hidden="true" tabindex="-1"></a><span class="co"># ReLoRA configuration</span></span>
|
||||
<span id="cb1-407"><a href="#cb1-407" 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="cb1-408"><a href="#cb1-408" 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="cb1-409"><a href="#cb1-409" 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="cb1-410"><a href="#cb1-410" 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="cb1-411"><a href="#cb1-411" 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="cb1-412"><a href="#cb1-412" 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="cb1-413"><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="co"># wandb configuration if you're using it</span></span>
|
||||
<span id="cb1-415"><a href="#cb1-415" 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="cb1-416"><a href="#cb1-416" 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="cb1-417"><a href="#cb1-417" 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="cb1-418"><a href="#cb1-418" 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="cb1-419"><a href="#cb1-419" aria-hidden="true" tabindex="-1"></a><span class="fu">wandb_watch</span><span class="kw">:</span></span>
|
||||
<span id="cb1-420"><a href="#cb1-420" 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="cb1-421"><a href="#cb1-421" 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="cb1-422"><a href="#cb1-422" 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="cb1-423"><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 use torch.compile and which backend to use</span></span>
|
||||
<span id="cb1-425"><a href="#cb1-425" aria-hidden="true" tabindex="-1"></a><span class="co"># setting to `auto` will enable torch compile when torch>=2.5.1</span></span>
|
||||
<span id="cb1-426"><a href="#cb1-426" 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="cb1-427"><a href="#cb1-427" 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="cb1-428"><a href="#cb1-428" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-429"><a href="#cb1-429" aria-hidden="true" tabindex="-1"></a><span class="co"># Training hyperparameters</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"># If greater than 1, backpropagation will be skipped and the gradients will be accumulated for the given number of steps.</span></span>
|
||||
<span id="cb1-432"><a href="#cb1-432" 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="cb1-433"><a href="#cb1-433" 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="cb1-434"><a href="#cb1-434" aria-hidden="true" tabindex="-1"></a><span class="co"># Batch size per gpu = micro_batch_size * gradient_accumulation_steps</span></span>
|
||||
<span id="cb1-435"><a href="#cb1-435" 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="cb1-436"><a href="#cb1-436" aria-hidden="true" tabindex="-1"></a><span class="fu">eval_batch_size</span><span class="kw">:</span></span>
|
||||
<span id="cb1-437"><a href="#cb1-437" 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="cb1-438"><a href="#cb1-438" 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="cb1-439"><a href="#cb1-439" 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="cb1-440"><a href="#cb1-440" 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="cb1-441"><a href="#cb1-441" aria-hidden="true" tabindex="-1"></a><span class="fu">lr_quadratic_warmup</span><span class="kw">:</span></span>
|
||||
<span id="cb1-442"><a href="#cb1-442" aria-hidden="true" tabindex="-1"></a><span class="fu">logging_steps</span><span class="kw">:</span></span>
|
||||
<span id="cb1-443"><a href="#cb1-443" 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="cb1-444"><a href="#cb1-444" 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="cb1-445"><a href="#cb1-445" 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="cb1-446"><a href="#cb1-446" 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="cb1-447"><a href="#cb1-447" 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="cb1-448"><a href="#cb1-448" 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="cb1-449"><a href="#cb1-449" 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="cb1-450"><a href="#cb1-450" 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="cb1-451"><a href="#cb1-451" aria-hidden="true" tabindex="-1"></a><span class="co"># if both are set, num_epochs will not be guaranteed.</span></span>
|
||||
<span id="cb1-452"><a href="#cb1-452" aria-hidden="true" tabindex="-1"></a><span class="co"># e.g., when 1 epoch is 1000 steps => `num_epochs: 2` and `max_steps: 100` will train for 100 steps</span></span>
|
||||
<span id="cb1-453"><a href="#cb1-453" aria-hidden="true" tabindex="-1"></a><span class="fu">max_steps</span><span class="kw">:</span></span>
|
||||
<span id="cb1-424"><a href="#cb1-424" aria-hidden="true" tabindex="-1"></a><span class="co"># mlflow configuration if you're using it</span></span>
|
||||
<span id="cb1-425"><a href="#cb1-425" 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="cb1-426"><a href="#cb1-426" 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="cb1-427"><a href="#cb1-427" 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="cb1-428"><a href="#cb1-428" 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="cb1-429"><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"># Comet configuration if you're using it</span></span>
|
||||
<span id="cb1-431"><a href="#cb1-431" 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="cb1-432"><a href="#cb1-432" 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="cb1-433"><a href="#cb1-433" 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="cb1-434"><a href="#cb1-434" 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="cb1-435"><a href="#cb1-435" 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="cb1-436"><a href="#cb1-436" 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="cb1-437"><a href="#cb1-437" 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="cb1-438"><a href="#cb1-438" 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="cb1-439"><a href="#cb1-439" 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="cb1-440"><a href="#cb1-440" 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="cb1-441"><a href="#cb1-441" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-442"><a href="#cb1-442" aria-hidden="true" tabindex="-1"></a><span class="co"># Tensorboard</span></span>
|
||||
<span id="cb1-443"><a href="#cb1-443" 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="cb1-444"><a href="#cb1-444" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-445"><a href="#cb1-445" aria-hidden="true" tabindex="-1"></a><span class="co"># Where to save the full-finetuned model to</span></span>
|
||||
<span id="cb1-446"><a href="#cb1-446" 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="cb1-447"><a href="#cb1-447" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-448"><a href="#cb1-448" aria-hidden="true" tabindex="-1"></a><span class="co"># Whether to use torch.compile and which backend to use</span></span>
|
||||
<span id="cb1-449"><a href="#cb1-449" aria-hidden="true" tabindex="-1"></a><span class="co"># setting to `auto` will enable torch compile when torch>=2.5.1</span></span>
|
||||
<span id="cb1-450"><a href="#cb1-450" 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="cb1-451"><a href="#cb1-451" 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="cb1-452"><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"># Training hyperparameters</span></span>
|
||||
<span id="cb1-454"><a href="#cb1-454" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-455"><a href="#cb1-455" 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="cb1-456"><a href="#cb1-456" 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="cb1-457"><a href="#cb1-457" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-458"><a href="#cb1-458" 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="cb1-459"><a href="#cb1-459" 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="cb1-460"><a href="#cb1-460" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-461"><a href="#cb1-461" 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="cb1-462"><a href="#cb1-462" 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="cb1-463"><a href="#cb1-463" aria-hidden="true" tabindex="-1"></a><span class="fu">do_causal_lm_eval</span><span class="kw">:</span><span class="co"> # Whether to run causal language model evaluation for metrics in `eval_causal_lm_metrics`.</span></span>
|
||||
<span id="cb1-464"><a href="#cb1-464" 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="cb1-465"><a href="#cb1-465" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-466"><a href="#cb1-466" 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="cb1-467"><a href="#cb1-467" 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="cb1-468"><a href="#cb1-468" aria-hidden="true" tabindex="-1"></a><span class="co"> # snapshots can be visualized @ https://pytorch.org/memory_viz</span></span>
|
||||
<span id="cb1-469"><a href="#cb1-469" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-470"><a href="#cb1-470" 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="cb1-471"><a href="#cb1-471" 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="cb1-472"><a href="#cb1-472" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-473"><a href="#cb1-473" aria-hidden="true" tabindex="-1"></a><span class="co"># Save model as safetensors (require safetensors package)</span></span>
|
||||
<span id="cb1-474"><a href="#cb1-474" aria-hidden="true" tabindex="-1"></a><span class="fu">save_safetensors</span><span class="kw">:</span></span>
|
||||
<span id="cb1-475"><a href="#cb1-475" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-476"><a href="#cb1-476" 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="cb1-477"><a href="#cb1-477" 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="cb1-478"><a href="#cb1-478" aria-hidden="true" tabindex="-1"></a><span class="co"># Group similarly sized data to minimize padding.</span></span>
|
||||
<span id="cb1-479"><a href="#cb1-479" 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="cb1-480"><a href="#cb1-480" 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="cb1-481"><a href="#cb1-481" 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="cb1-482"><a href="#cb1-482" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-483"><a href="#cb1-483" 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="cb1-484"><a href="#cb1-484" 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="cb1-485"><a href="#cb1-485" aria-hidden="true" tabindex="-1"></a><span class="co"># additional kwargs to pass to the trainer for gradient checkpointing</span></span>
|
||||
<span id="cb1-486"><a href="#cb1-486" aria-hidden="true" tabindex="-1"></a><span class="co"># gradient_checkpointing_kwargs:</span></span>
|
||||
<span id="cb1-487"><a href="#cb1-487" aria-hidden="true" tabindex="-1"></a><span class="co"># use_reentrant: true</span></span>
|
||||
<span id="cb1-488"><a href="#cb1-488" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-489"><a href="#cb1-489" 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="cb1-490"><a href="#cb1-490" 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="cb1-491"><a href="#cb1-491" 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="cb1-492"><a href="#cb1-492" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-493"><a href="#cb1-493" aria-hidden="true" tabindex="-1"></a><span class="co"># Specify a scheduler and kwargs to use with the optimizer</span></span>
|
||||
<span id="cb1-494"><a href="#cb1-494" aria-hidden="true" tabindex="-1"></a><span class="fu">lr_scheduler</span><span class="kw">:</span><span class="co"> # 'one_cycle' | 'rex' | 'log_sweep' | empty for cosine</span></span>
|
||||
<span id="cb1-495"><a href="#cb1-495" aria-hidden="true" tabindex="-1"></a><span class="fu">lr_scheduler_kwargs</span><span class="kw">:</span></span>
|
||||
<span id="cb1-496"><a href="#cb1-496" 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="cb1-497"><a href="#cb1-497" 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="cb1-498"><a href="#cb1-498" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-499"><a href="#cb1-499" aria-hidden="true" tabindex="-1"></a><span class="co"># For one_cycle optim</span></span>
|
||||
<span id="cb1-500"><a href="#cb1-500" 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="cb1-501"><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 optimizer</span></span>
|
||||
<span id="cb1-503"><a href="#cb1-503" aria-hidden="true" tabindex="-1"></a><span class="co"># Valid values are driven by the Transformers OptimizerNames class, see:</span></span>
|
||||
<span id="cb1-504"><a href="#cb1-504" aria-hidden="true" tabindex="-1"></a><span class="co"># https://github.com/huggingface/transformers/blob/cbf924b76c03828101a34069a96d209314114fd5/src/transformers/training_args.py#L144-L189</span></span>
|
||||
<span id="cb1-505"><a href="#cb1-505" aria-hidden="true" tabindex="-1"></a><span class="co">#</span></span>
|
||||
<span id="cb1-506"><a href="#cb1-506" 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="cb1-507"><a href="#cb1-507" 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="cb1-508"><a href="#cb1-508" aria-hidden="true" tabindex="-1"></a><span class="co"># in the examples/ for your model and fine-tuning use case.</span></span>
|
||||
<span id="cb1-509"><a href="#cb1-509" aria-hidden="true" tabindex="-1"></a><span class="co">#</span></span>
|
||||
<span id="cb1-510"><a href="#cb1-510" aria-hidden="true" tabindex="-1"></a><span class="co"># Valid values for 'optimizer' include:</span></span>
|
||||
<span id="cb1-511"><a href="#cb1-511" aria-hidden="true" tabindex="-1"></a><span class="co"># - adamw_torch</span></span>
|
||||
<span id="cb1-512"><a href="#cb1-512" aria-hidden="true" tabindex="-1"></a><span class="co"># - adamw_torch_fused</span></span>
|
||||
<span id="cb1-513"><a href="#cb1-513" aria-hidden="true" tabindex="-1"></a><span class="co"># - adamw_torch_xla</span></span>
|
||||
<span id="cb1-514"><a href="#cb1-514" aria-hidden="true" tabindex="-1"></a><span class="co"># - adamw_torch_npu_fused</span></span>
|
||||
<span id="cb1-515"><a href="#cb1-515" aria-hidden="true" tabindex="-1"></a><span class="co"># - adamw_apex_fused</span></span>
|
||||
<span id="cb1-516"><a href="#cb1-516" aria-hidden="true" tabindex="-1"></a><span class="co"># - adopt_adamw (an EXPERIMENTAL optimizer, only for torch version >= 2.5.1)</span></span>
|
||||
<span id="cb1-517"><a href="#cb1-517" aria-hidden="true" tabindex="-1"></a><span class="co"># - adafactor</span></span>
|
||||
<span id="cb1-518"><a href="#cb1-518" aria-hidden="true" tabindex="-1"></a><span class="co"># - adamw_anyprecision</span></span>
|
||||
<span id="cb1-519"><a href="#cb1-519" aria-hidden="true" tabindex="-1"></a><span class="co"># - adamw_torch_4bit</span></span>
|
||||
<span id="cb1-520"><a href="#cb1-520" aria-hidden="true" tabindex="-1"></a><span class="co"># - ademamix</span></span>
|
||||
<span id="cb1-521"><a href="#cb1-521" aria-hidden="true" tabindex="-1"></a><span class="co"># - sgd</span></span>
|
||||
<span id="cb1-522"><a href="#cb1-522" aria-hidden="true" tabindex="-1"></a><span class="co"># - adagrad</span></span>
|
||||
<span id="cb1-523"><a href="#cb1-523" aria-hidden="true" tabindex="-1"></a><span class="co"># - adamw_bnb_8bit</span></span>
|
||||
<span id="cb1-524"><a href="#cb1-524" aria-hidden="true" tabindex="-1"></a><span class="co"># - adamw_8bit # alias for adamw_bnb_8bit</span></span>
|
||||
<span id="cb1-525"><a href="#cb1-525" aria-hidden="true" tabindex="-1"></a><span class="co"># - ademamix_8bit</span></span>
|
||||
<span id="cb1-526"><a href="#cb1-526" aria-hidden="true" tabindex="-1"></a><span class="co"># - lion_8bit</span></span>
|
||||
<span id="cb1-527"><a href="#cb1-527" aria-hidden="true" tabindex="-1"></a><span class="co"># - lion_32bit</span></span>
|
||||
<span id="cb1-528"><a href="#cb1-528" aria-hidden="true" tabindex="-1"></a><span class="co"># - paged_adamw_32bit</span></span>
|
||||
<span id="cb1-529"><a href="#cb1-529" aria-hidden="true" tabindex="-1"></a><span class="co"># - paged_adamw_8bit</span></span>
|
||||
<span id="cb1-530"><a href="#cb1-530" aria-hidden="true" tabindex="-1"></a><span class="co"># - paged_ademamix_32bit</span></span>
|
||||
<span id="cb1-531"><a href="#cb1-531" aria-hidden="true" tabindex="-1"></a><span class="co"># - paged_ademamix_8bit</span></span>
|
||||
<span id="cb1-532"><a href="#cb1-532" aria-hidden="true" tabindex="-1"></a><span class="co"># - paged_lion_32bit</span></span>
|
||||
<span id="cb1-533"><a href="#cb1-533" aria-hidden="true" tabindex="-1"></a><span class="co"># - paged_lion_8bit</span></span>
|
||||
<span id="cb1-534"><a href="#cb1-534" aria-hidden="true" tabindex="-1"></a><span class="co"># - rmsprop</span></span>
|
||||
<span id="cb1-535"><a href="#cb1-535" aria-hidden="true" tabindex="-1"></a><span class="co"># - rmsprop_bnb</span></span>
|
||||
<span id="cb1-536"><a href="#cb1-536" aria-hidden="true" tabindex="-1"></a><span class="co"># - rmsprop_bnb_8bit</span></span>
|
||||
<span id="cb1-537"><a href="#cb1-537" aria-hidden="true" tabindex="-1"></a><span class="co"># - rmsprop_bnb_32bit</span></span>
|
||||
<span id="cb1-538"><a href="#cb1-538" aria-hidden="true" tabindex="-1"></a><span class="co"># - galore_adamw</span></span>
|
||||
<span id="cb1-539"><a href="#cb1-539" aria-hidden="true" tabindex="-1"></a><span class="co"># - galore_adamw_8bit</span></span>
|
||||
<span id="cb1-540"><a href="#cb1-540" aria-hidden="true" tabindex="-1"></a><span class="co"># - galore_adafactor</span></span>
|
||||
<span id="cb1-541"><a href="#cb1-541" aria-hidden="true" tabindex="-1"></a><span class="co"># - galore_adamw_layerwise</span></span>
|
||||
<span id="cb1-542"><a href="#cb1-542" aria-hidden="true" tabindex="-1"></a><span class="co"># - galore_adamw_8bit_layerwise</span></span>
|
||||
<span id="cb1-543"><a href="#cb1-543" aria-hidden="true" tabindex="-1"></a><span class="co"># - galore_adafactor_layerwise</span></span>
|
||||
<span id="cb1-544"><a href="#cb1-544" aria-hidden="true" tabindex="-1"></a><span class="co"># - lomo</span></span>
|
||||
<span id="cb1-545"><a href="#cb1-545" aria-hidden="true" tabindex="-1"></a><span class="co"># - adalomo</span></span>
|
||||
<span id="cb1-546"><a href="#cb1-546" aria-hidden="true" tabindex="-1"></a><span class="co"># - grokadamw</span></span>
|
||||
<span id="cb1-547"><a href="#cb1-547" aria-hidden="true" tabindex="-1"></a><span class="co"># - schedule_free_adamw</span></span>
|
||||
<span id="cb1-548"><a href="#cb1-548" aria-hidden="true" tabindex="-1"></a><span class="co"># - schedule_free_sgd</span></span>
|
||||
<span id="cb1-549"><a href="#cb1-549" aria-hidden="true" tabindex="-1"></a><span class="co"># - apollo_adamw</span></span>
|
||||
<span id="cb1-550"><a href="#cb1-550" aria-hidden="true" tabindex="-1"></a><span class="co"># - apollo_adamw_layerwise</span></span>
|
||||
<span id="cb1-551"><a href="#cb1-551" aria-hidden="true" tabindex="-1"></a><span class="co">#</span></span>
|
||||
<span id="cb1-552"><a href="#cb1-552" aria-hidden="true" tabindex="-1"></a><span class="co"># Additional custom optimizers include:</span></span>
|
||||
<span id="cb1-553"><a href="#cb1-553" aria-hidden="true" tabindex="-1"></a><span class="co"># - optimi_adamw</span></span>
|
||||
<span id="cb1-554"><a href="#cb1-554" aria-hidden="true" tabindex="-1"></a><span class="co"># - ao_adamw_8bit</span></span>
|
||||
<span id="cb1-555"><a href="#cb1-555" aria-hidden="true" tabindex="-1"></a><span class="co"># - ao_adamw_fp8</span></span>
|
||||
<span id="cb1-556"><a href="#cb1-556" aria-hidden="true" tabindex="-1"></a><span class="fu">optimizer</span><span class="kw">:</span></span>
|
||||
<span id="cb1-557"><a href="#cb1-557" aria-hidden="true" tabindex="-1"></a><span class="co"># Dictionary of arguments to pass to the optimizer</span></span>
|
||||
<span id="cb1-558"><a href="#cb1-558" aria-hidden="true" tabindex="-1"></a><span class="fu">optim_args</span><span class="kw">:</span></span>
|
||||
<span id="cb1-559"><a href="#cb1-559" aria-hidden="true" tabindex="-1"></a><span class="co"># For Galore Optimizers the following optim_args are available</span></span>
|
||||
<span id="cb1-560"><a href="#cb1-560" aria-hidden="true" tabindex="-1"></a><span class="co"># rank: # type: int</span></span>
|
||||
<span id="cb1-561"><a href="#cb1-561" aria-hidden="true" tabindex="-1"></a><span class="co"># update_proj_gap # type: int</span></span>
|
||||
<span id="cb1-562"><a href="#cb1-562" aria-hidden="true" tabindex="-1"></a><span class="co"># scale # type: float</span></span>
|
||||
<span id="cb1-563"><a href="#cb1-563" aria-hidden="true" tabindex="-1"></a><span class="co"># proj_type: # type: str, default = std</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"># 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="cb1-566"><a href="#cb1-566" aria-hidden="true" tabindex="-1"></a><span class="fu">optim_target_modules</span><span class="kw">:</span></span>
|
||||
<span id="cb1-567"><a href="#cb1-567" aria-hidden="true" tabindex="-1"></a><span class="co"># - self_attn # for llama</span></span>
|
||||
<span id="cb1-568"><a href="#cb1-568" aria-hidden="true" tabindex="-1"></a><span class="co"># - mlp</span></span>
|
||||
<span id="cb1-569"><a href="#cb1-569" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-570"><a href="#cb1-570" aria-hidden="true" tabindex="-1"></a><span class="co"># Specify weight decay</span></span>
|
||||
<span id="cb1-571"><a href="#cb1-571" aria-hidden="true" tabindex="-1"></a><span class="fu">weight_decay</span><span class="kw">:</span></span>
|
||||
<span id="cb1-572"><a href="#cb1-572" aria-hidden="true" tabindex="-1"></a><span class="co"># adamw hyperparams</span></span>
|
||||
<span id="cb1-573"><a href="#cb1-573" aria-hidden="true" tabindex="-1"></a><span class="fu">adam_beta1</span><span class="kw">:</span></span>
|
||||
<span id="cb1-574"><a href="#cb1-574" aria-hidden="true" tabindex="-1"></a><span class="fu">adam_beta2</span><span class="kw">:</span></span>
|
||||
<span id="cb1-575"><a href="#cb1-575" aria-hidden="true" tabindex="-1"></a><span class="fu">adam_epsilon</span><span class="kw">:</span></span>
|
||||
<span id="cb1-576"><a href="#cb1-576" aria-hidden="true" tabindex="-1"></a><span class="co"># Gradient clipping max norm</span></span>
|
||||
<span id="cb1-577"><a href="#cb1-577" aria-hidden="true" tabindex="-1"></a><span class="fu">max_grad_norm</span><span class="kw">:</span></span>
|
||||
<span id="cb1-578"><a href="#cb1-578" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-579"><a href="#cb1-579" aria-hidden="true" tabindex="-1"></a><span class="co"># Augmentation techniques</span></span>
|
||||
<span id="cb1-580"><a href="#cb1-580" 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="cb1-581"><a href="#cb1-581" aria-hidden="true" tabindex="-1"></a><span class="co"># currently only supported on Llama and Mistral</span></span>
|
||||
<span id="cb1-582"><a href="#cb1-582" aria-hidden="true" tabindex="-1"></a><span class="fu">neftune_noise_alpha</span><span class="kw">:</span></span>
|
||||
<span id="cb1-583"><a href="#cb1-583" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-584"><a href="#cb1-584" aria-hidden="true" tabindex="-1"></a><span class="co"># Optional[bool]. Whether to bettertransformers</span></span>
|
||||
<span id="cb1-585"><a href="#cb1-585" aria-hidden="true" tabindex="-1"></a><span class="fu">flash_optimum</span><span class="kw">:</span></span>
|
||||
<span id="cb1-586"><a href="#cb1-586" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-587"><a href="#cb1-587" aria-hidden="true" tabindex="-1"></a><span class="co"># Note: Only one of the following attention patches can be used at a time.</span></span>
|
||||
<span id="cb1-588"><a href="#cb1-588" aria-hidden="true" tabindex="-1"></a><span class="co"># For example, if you set `xformers_attention` to `true`, do not set `flash_attention` to `true`.</span></span>
|
||||
<span id="cb1-589"><a href="#cb1-589" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-590"><a href="#cb1-590" aria-hidden="true" tabindex="-1"></a><span class="co"># Optional[bool]. Whether to use xformers attention patch https://github.com/facebookresearch/xformers:</span></span>
|
||||
<span id="cb1-591"><a href="#cb1-591" aria-hidden="true" tabindex="-1"></a><span class="fu">xformers_attention</span><span class="kw">:</span></span>
|
||||
<span id="cb1-592"><a href="#cb1-592" aria-hidden="true" tabindex="-1"></a><span class="co"># Optional[bool]. Whether to use flash attention patch https://github.com/Dao-AILab/flash-attention:</span></span>
|
||||
<span id="cb1-593"><a href="#cb1-593" aria-hidden="true" tabindex="-1"></a><span class="fu">flash_attention</span><span class="kw">:</span></span>
|
||||
<span id="cb1-594"><a href="#cb1-594" aria-hidden="true" tabindex="-1"></a><span class="fu">flash_attn_cross_entropy</span><span class="kw">:</span><span class="co"> # Optional[bool]. Whether to use flash-attention cross entropy implementation - advanced use only</span></span>
|
||||
<span id="cb1-595"><a href="#cb1-595" aria-hidden="true" tabindex="-1"></a><span class="fu">flash_attn_rms_norm</span><span class="kw">:</span><span class="co"> # Optional[bool]. Whether to use flash-attention rms norm implementation - advanced use only</span></span>
|
||||
<span id="cb1-596"><a href="#cb1-596" aria-hidden="true" tabindex="-1"></a><span class="fu">flash_attn_fuse_qkv</span><span class="kw">:</span><span class="co"> # Optional[bool]. Whether to fuse QKV into a single operation</span></span>
|
||||
<span id="cb1-597"><a href="#cb1-597" aria-hidden="true" tabindex="-1"></a><span class="fu">flash_attn_fuse_mlp</span><span class="kw">:</span><span class="co"> # Optional[bool]. Whether to fuse part of the MLP into a single operation</span></span>
|
||||
<span id="cb1-598"><a href="#cb1-598" aria-hidden="true" tabindex="-1"></a><span class="co"># Optional[bool]. Whether to use scaled-dot-product attention</span></span>
|
||||
<span id="cb1-599"><a href="#cb1-599" 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="cb1-600"><a href="#cb1-600" aria-hidden="true" tabindex="-1"></a><span class="fu">sdp_attention</span><span class="kw">:</span></span>
|
||||
<span id="cb1-601"><a href="#cb1-601" aria-hidden="true" tabindex="-1"></a><span class="co"># Optional[bool]. Shifted-sparse attention (only llama) - https://arxiv.org/pdf/2309.12307.pdf</span></span>
|
||||
<span id="cb1-602"><a href="#cb1-602" aria-hidden="true" tabindex="-1"></a><span class="fu">s2_attention</span><span class="kw">:</span></span>
|
||||
<span id="cb1-603"><a href="#cb1-603" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-604"><a href="#cb1-604" aria-hidden="true" tabindex="-1"></a><span class="co"># Optional[bool]. Whether to use low_cpu_mem_usage</span></span>
|
||||
<span id="cb1-605"><a href="#cb1-605" aria-hidden="true" tabindex="-1"></a><span class="fu">low_cpu_mem_usage</span><span class="kw">:</span></span>
|
||||
<span id="cb1-606"><a href="#cb1-606" aria-hidden="true" tabindex="-1"></a><span class="co"># Optional[str]. Resume from a specific checkpoint dir</span></span>
|
||||
<span id="cb1-607"><a href="#cb1-607" aria-hidden="true" tabindex="-1"></a><span class="fu">resume_from_checkpoint</span><span class="kw">:</span></span>
|
||||
<span id="cb1-608"><a href="#cb1-608" aria-hidden="true" tabindex="-1"></a><span class="co"># Optional[bool]. If resume_from_checkpoint isn't set and you simply want it to start where it left off.</span></span>
|
||||
<span id="cb1-609"><a href="#cb1-609" aria-hidden="true" tabindex="-1"></a><span class="co"># Be careful with this being turned on between different models.</span></span>
|
||||
<span id="cb1-610"><a href="#cb1-610" 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="cb1-611"><a href="#cb1-611" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-612"><a href="#cb1-612" aria-hidden="true" tabindex="-1"></a><span class="co">## Multimodal section</span></span>
|
||||
<span id="cb1-613"><a href="#cb1-613" aria-hidden="true" tabindex="-1"></a><span class="co"># int | tuple[int, int] | None . Size to resize images to, width x height.</span></span>
|
||||
<span id="cb1-614"><a href="#cb1-614" aria-hidden="true" tabindex="-1"></a><span class="co"># Will read from model/processor config if not set.</span></span>
|
||||
<span id="cb1-615"><a href="#cb1-615" aria-hidden="true" tabindex="-1"></a><span class="fu">image_size</span><span class="kw">:</span></span>
|
||||
<span id="cb1-616"><a href="#cb1-616" aria-hidden="true" tabindex="-1"></a><span class="co"># str. Algorithm to use for image resizing. "bilinear", "bicubic", "lanczos". Default is "bilinear".</span></span>
|
||||
<span id="cb1-617"><a href="#cb1-617" aria-hidden="true" tabindex="-1"></a><span class="fu">image_resize_algorithm</span><span class="kw">:</span><span class="at"> </span><span class="st">'bilinear'</span></span>
|
||||
<span id="cb1-618"><a href="#cb1-618" aria-hidden="true" tabindex="-1"></a><span class="co">## End of multimodal section</span></span>
|
||||
<span id="cb1-619"><a href="#cb1-619" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-620"><a href="#cb1-620" 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="cb1-621"><a href="#cb1-621" aria-hidden="true" tabindex="-1"></a><span class="fu">local_rank</span><span class="kw">:</span></span>
|
||||
<span id="cb1-622"><a href="#cb1-622" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-623"><a href="#cb1-623" aria-hidden="true" tabindex="-1"></a><span class="co"># Add or change special tokens.</span></span>
|
||||
<span id="cb1-624"><a href="#cb1-624" 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="cb1-625"><a href="#cb1-625" aria-hidden="true" tabindex="-1"></a><span class="fu">special_tokens</span><span class="kw">:</span></span>
|
||||
<span id="cb1-626"><a href="#cb1-626" aria-hidden="true" tabindex="-1"></a><span class="co"> # bos_token: "<s>"</span></span>
|
||||
<span id="cb1-627"><a href="#cb1-627" aria-hidden="true" tabindex="-1"></a><span class="co"> # eos_token: "</s>"</span></span>
|
||||
<span id="cb1-628"><a href="#cb1-628" aria-hidden="true" tabindex="-1"></a><span class="co"> # unk_token: "<unk>"</span></span>
|
||||
<span id="cb1-629"><a href="#cb1-629" aria-hidden="true" tabindex="-1"></a><span class="co"> # pad_token: "[PAD]"</span></span>
|
||||
<span id="cb1-630"><a href="#cb1-630" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-631"><a href="#cb1-631" aria-hidden="true" tabindex="-1"></a><span class="co"># Add extra tokens.</span></span>
|
||||
<span id="cb1-632"><a href="#cb1-632" aria-hidden="true" tabindex="-1"></a><span class="fu">tokens</span><span class="kw">:</span></span>
|
||||
<span id="cb1-633"><a href="#cb1-633" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-634"><a href="#cb1-634" aria-hidden="true" tabindex="-1"></a><span class="co"># Mapping token_id to new_token_string to override reserved added_tokens in the tokenizer.</span></span>
|
||||
<span id="cb1-635"><a href="#cb1-635" aria-hidden="true" tabindex="-1"></a><span class="co"># Only works for tokens that are not part of the base vocab (aka are added_tokens).</span></span>
|
||||
<span id="cb1-636"><a href="#cb1-636" aria-hidden="true" tabindex="-1"></a><span class="co"># Can be checked if they exist in tokenizer.json added_tokens.</span></span>
|
||||
<span id="cb1-637"><a href="#cb1-637" aria-hidden="true" tabindex="-1"></a><span class="fu">added_tokens_overrides</span><span class="kw">:</span><span class="co"> # Dict[int, str]</span></span>
|
||||
<span id="cb1-638"><a href="#cb1-638" aria-hidden="true" tabindex="-1"></a><span class="co"># 128041: "<|im_start|>"</span></span>
|
||||
<span id="cb1-639"><a href="#cb1-639" aria-hidden="true" tabindex="-1"></a><span class="co"># 128042: "<|im_end|>"</span></span>
|
||||
<span id="cb1-640"><a href="#cb1-640" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-641"><a href="#cb1-641" aria-hidden="true" tabindex="-1"></a><span class="co"># FSDP</span></span>
|
||||
<span id="cb1-642"><a href="#cb1-642" aria-hidden="true" tabindex="-1"></a><span class="fu">fsdp</span><span class="kw">:</span></span>
|
||||
<span id="cb1-643"><a href="#cb1-643" aria-hidden="true" tabindex="-1"></a><span class="fu">fsdp_config</span><span class="kw">:</span></span>
|
||||
<span id="cb1-644"><a href="#cb1-644" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-645"><a href="#cb1-645" aria-hidden="true" tabindex="-1"></a><span class="co"># Deepspeed config path. e.g., deepspeed_configs/zero3.json</span></span>
|
||||
<span id="cb1-646"><a href="#cb1-646" aria-hidden="true" tabindex="-1"></a><span class="fu">deepspeed</span><span class="kw">:</span></span>
|
||||
<span id="cb1-647"><a href="#cb1-647" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-648"><a href="#cb1-648" aria-hidden="true" tabindex="-1"></a><span class="co"># Advanced DDP Arguments</span></span>
|
||||
<span id="cb1-649"><a href="#cb1-649" aria-hidden="true" tabindex="-1"></a><span class="fu">ddp_timeout</span><span class="kw">:</span></span>
|
||||
<span id="cb1-650"><a href="#cb1-650" aria-hidden="true" tabindex="-1"></a><span class="fu">ddp_bucket_cap_mb</span><span class="kw">:</span></span>
|
||||
<span id="cb1-651"><a href="#cb1-651" aria-hidden="true" tabindex="-1"></a><span class="fu">ddp_broadcast_buffers</span><span class="kw">:</span></span>
|
||||
<span id="cb1-652"><a href="#cb1-652" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-653"><a href="#cb1-653" aria-hidden="true" tabindex="-1"></a><span class="co"># Sequence parallelism</span></span>
|
||||
<span id="cb1-654"><a href="#cb1-654" aria-hidden="true" tabindex="-1"></a><span class="co"># Set to a divisor of the number of GPUs available to split sequences into chunks of equal size.</span></span>
|
||||
<span id="cb1-655"><a href="#cb1-655" aria-hidden="true" tabindex="-1"></a><span class="co"># Use in long context training to prevent OOM when sequences cannot fit into a single GPU's VRAM.</span></span>
|
||||
<span id="cb1-656"><a href="#cb1-656" aria-hidden="true" tabindex="-1"></a><span class="co"># E.g., if 4 GPUs are available, set this value to 2 to split each sequence into two equal-sized</span></span>
|
||||
<span id="cb1-657"><a href="#cb1-657" aria-hidden="true" tabindex="-1"></a><span class="co"># subsequences, or set to 4 to split into four equal-sized subsequences.</span></span>
|
||||
<span id="cb1-658"><a href="#cb1-658" aria-hidden="true" tabindex="-1"></a><span class="co"># See https://axolotl-ai-cloud.github.io/axolotl/docs/sequence_parallelism.html for more details.</span></span>
|
||||
<span id="cb1-659"><a href="#cb1-659" aria-hidden="true" tabindex="-1"></a><span class="fu">sequence_parallel_degree</span><span class="kw">:</span></span>
|
||||
<span id="cb1-660"><a href="#cb1-660" aria-hidden="true" tabindex="-1"></a><span class="co"># Optional; strides across the key dimension. Larger values use more memory but should make training faster.</span></span>
|
||||
<span id="cb1-661"><a href="#cb1-661" aria-hidden="true" tabindex="-1"></a><span class="co"># Must evenly divide the number of KV heads in your model.</span></span>
|
||||
<span id="cb1-662"><a href="#cb1-662" aria-hidden="true" tabindex="-1"></a><span class="fu">heads_k_stride</span><span class="kw">:</span><span class="at"> </span><span class="dv">1</span></span>
|
||||
<span id="cb1-663"><a href="#cb1-663" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-664"><a href="#cb1-664" aria-hidden="true" tabindex="-1"></a><span class="co"># Path to torch distx for optim 'adamw_anyprecision'</span></span>
|
||||
<span id="cb1-665"><a href="#cb1-665" aria-hidden="true" tabindex="-1"></a><span class="fu">torchdistx_path</span><span class="kw">:</span></span>
|
||||
<span id="cb1-666"><a href="#cb1-666" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-667"><a href="#cb1-667" 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="cb1-668"><a href="#cb1-668" aria-hidden="true" tabindex="-1"></a><span class="fu">pretraining_dataset</span><span class="kw">:</span></span>
|
||||
<span id="cb1-669"><a href="#cb1-669" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-670"><a href="#cb1-670" aria-hidden="true" tabindex="-1"></a><span class="co"># Debug mode</span></span>
|
||||
<span id="cb1-671"><a href="#cb1-671" aria-hidden="true" tabindex="-1"></a><span class="fu">debug</span><span class="kw">:</span></span>
|
||||
<span id="cb1-672"><a href="#cb1-672" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-673"><a href="#cb1-673" aria-hidden="true" tabindex="-1"></a><span class="co"># Seed</span></span>
|
||||
<span id="cb1-674"><a href="#cb1-674" aria-hidden="true" tabindex="-1"></a><span class="fu">seed</span><span class="kw">:</span></span>
|
||||
<span id="cb1-675"><a href="#cb1-675" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-676"><a href="#cb1-676" aria-hidden="true" tabindex="-1"></a><span class="co"># Allow overwrite yml config using from cli</span></span>
|
||||
<span id="cb1-677"><a href="#cb1-677" 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="cb1-455"><a href="#cb1-455" 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="cb1-456"><a href="#cb1-456" 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="cb1-457"><a href="#cb1-457" 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="cb1-458"><a href="#cb1-458" aria-hidden="true" tabindex="-1"></a><span class="co"># Batch size per gpu = micro_batch_size * gradient_accumulation_steps</span></span>
|
||||
<span id="cb1-459"><a href="#cb1-459" 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="cb1-460"><a href="#cb1-460" aria-hidden="true" tabindex="-1"></a><span class="fu">eval_batch_size</span><span class="kw">:</span></span>
|
||||
<span id="cb1-461"><a href="#cb1-461" 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="cb1-462"><a href="#cb1-462" 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="cb1-463"><a href="#cb1-463" 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="cb1-464"><a href="#cb1-464" 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="cb1-465"><a href="#cb1-465" aria-hidden="true" tabindex="-1"></a><span class="fu">lr_quadratic_warmup</span><span class="kw">:</span></span>
|
||||
<span id="cb1-466"><a href="#cb1-466" aria-hidden="true" tabindex="-1"></a><span class="fu">logging_steps</span><span class="kw">:</span></span>
|
||||
<span id="cb1-467"><a href="#cb1-467" 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="cb1-468"><a href="#cb1-468" 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="cb1-469"><a href="#cb1-469" 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="cb1-470"><a href="#cb1-470" 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="cb1-471"><a href="#cb1-471" 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="cb1-472"><a href="#cb1-472" 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="cb1-473"><a href="#cb1-473" 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="cb1-474"><a href="#cb1-474" 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="cb1-475"><a href="#cb1-475" aria-hidden="true" tabindex="-1"></a><span class="co"># if both are set, num_epochs will not be guaranteed.</span></span>
|
||||
<span id="cb1-476"><a href="#cb1-476" aria-hidden="true" tabindex="-1"></a><span class="co"># e.g., when 1 epoch is 1000 steps => `num_epochs: 2` and `max_steps: 100` will train for 100 steps</span></span>
|
||||
<span id="cb1-477"><a href="#cb1-477" aria-hidden="true" tabindex="-1"></a><span class="fu">max_steps</span><span class="kw">:</span></span>
|
||||
<span id="cb1-478"><a href="#cb1-478" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-479"><a href="#cb1-479" 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="cb1-480"><a href="#cb1-480" 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="cb1-481"><a href="#cb1-481" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-482"><a href="#cb1-482" 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="cb1-483"><a href="#cb1-483" 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="cb1-484"><a href="#cb1-484" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-485"><a href="#cb1-485" 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="cb1-486"><a href="#cb1-486" 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="cb1-487"><a href="#cb1-487" aria-hidden="true" tabindex="-1"></a><span class="fu">do_causal_lm_eval</span><span class="kw">:</span><span class="co"> # Whether to run causal language model evaluation for metrics in `eval_causal_lm_metrics`.</span></span>
|
||||
<span id="cb1-488"><a href="#cb1-488" 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="cb1-489"><a href="#cb1-489" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-490"><a href="#cb1-490" 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="cb1-491"><a href="#cb1-491" 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="cb1-492"><a href="#cb1-492" aria-hidden="true" tabindex="-1"></a><span class="co"> # snapshots can be visualized @ https://pytorch.org/memory_viz</span></span>
|
||||
<span id="cb1-493"><a href="#cb1-493" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-494"><a href="#cb1-494" 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="cb1-495"><a href="#cb1-495" 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="cb1-496"><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"># Save model as safetensors (require safetensors package)</span></span>
|
||||
<span id="cb1-498"><a href="#cb1-498" aria-hidden="true" tabindex="-1"></a><span class="fu">save_safetensors</span><span class="kw">:</span></span>
|
||||
<span id="cb1-499"><a href="#cb1-499" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-500"><a href="#cb1-500" 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="cb1-501"><a href="#cb1-501" 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="cb1-502"><a href="#cb1-502" aria-hidden="true" tabindex="-1"></a><span class="co"># Group similarly sized data to minimize padding.</span></span>
|
||||
<span id="cb1-503"><a href="#cb1-503" 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="cb1-504"><a href="#cb1-504" 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="cb1-505"><a href="#cb1-505" 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="cb1-506"><a href="#cb1-506" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-507"><a href="#cb1-507" 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="cb1-508"><a href="#cb1-508" 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="cb1-509"><a href="#cb1-509" aria-hidden="true" tabindex="-1"></a><span class="co"># additional kwargs to pass to the trainer for gradient checkpointing</span></span>
|
||||
<span id="cb1-510"><a href="#cb1-510" aria-hidden="true" tabindex="-1"></a><span class="co"># gradient_checkpointing_kwargs:</span></span>
|
||||
<span id="cb1-511"><a href="#cb1-511" aria-hidden="true" tabindex="-1"></a><span class="co"># use_reentrant: true</span></span>
|
||||
<span id="cb1-512"><a href="#cb1-512" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-513"><a href="#cb1-513" 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="cb1-514"><a href="#cb1-514" 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="cb1-515"><a href="#cb1-515" 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="cb1-516"><a href="#cb1-516" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-517"><a href="#cb1-517" aria-hidden="true" tabindex="-1"></a><span class="co"># Specify a scheduler and kwargs to use with the optimizer</span></span>
|
||||
<span id="cb1-518"><a href="#cb1-518" aria-hidden="true" tabindex="-1"></a><span class="fu">lr_scheduler</span><span class="kw">:</span><span class="co"> # 'one_cycle' | 'rex' | 'log_sweep' | empty for cosine</span></span>
|
||||
<span id="cb1-519"><a href="#cb1-519" aria-hidden="true" tabindex="-1"></a><span class="fu">lr_scheduler_kwargs</span><span class="kw">:</span></span>
|
||||
<span id="cb1-520"><a href="#cb1-520" 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="cb1-521"><a href="#cb1-521" 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="cb1-522"><a href="#cb1-522" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-523"><a href="#cb1-523" aria-hidden="true" tabindex="-1"></a><span class="co"># For one_cycle optim</span></span>
|
||||
<span id="cb1-524"><a href="#cb1-524" 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="cb1-525"><a href="#cb1-525" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-526"><a href="#cb1-526" aria-hidden="true" tabindex="-1"></a><span class="co"># Specify optimizer</span></span>
|
||||
<span id="cb1-527"><a href="#cb1-527" aria-hidden="true" tabindex="-1"></a><span class="co"># Valid values are driven by the Transformers OptimizerNames class, see:</span></span>
|
||||
<span id="cb1-528"><a href="#cb1-528" aria-hidden="true" tabindex="-1"></a><span class="co"># https://github.com/huggingface/transformers/blob/cbf924b76c03828101a34069a96d209314114fd5/src/transformers/training_args.py#L144-L189</span></span>
|
||||
<span id="cb1-529"><a href="#cb1-529" aria-hidden="true" tabindex="-1"></a><span class="co">#</span></span>
|
||||
<span id="cb1-530"><a href="#cb1-530" 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="cb1-531"><a href="#cb1-531" 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="cb1-532"><a href="#cb1-532" aria-hidden="true" tabindex="-1"></a><span class="co"># in the examples/ for your model and fine-tuning use case.</span></span>
|
||||
<span id="cb1-533"><a href="#cb1-533" aria-hidden="true" tabindex="-1"></a><span class="co">#</span></span>
|
||||
<span id="cb1-534"><a href="#cb1-534" aria-hidden="true" tabindex="-1"></a><span class="co"># Valid values for 'optimizer' include:</span></span>
|
||||
<span id="cb1-535"><a href="#cb1-535" aria-hidden="true" tabindex="-1"></a><span class="co"># - adamw_torch</span></span>
|
||||
<span id="cb1-536"><a href="#cb1-536" aria-hidden="true" tabindex="-1"></a><span class="co"># - adamw_torch_fused</span></span>
|
||||
<span id="cb1-537"><a href="#cb1-537" aria-hidden="true" tabindex="-1"></a><span class="co"># - adamw_torch_xla</span></span>
|
||||
<span id="cb1-538"><a href="#cb1-538" aria-hidden="true" tabindex="-1"></a><span class="co"># - adamw_torch_npu_fused</span></span>
|
||||
<span id="cb1-539"><a href="#cb1-539" aria-hidden="true" tabindex="-1"></a><span class="co"># - adamw_apex_fused</span></span>
|
||||
<span id="cb1-540"><a href="#cb1-540" aria-hidden="true" tabindex="-1"></a><span class="co"># - adopt_adamw (an EXPERIMENTAL optimizer, only for torch version >= 2.5.1)</span></span>
|
||||
<span id="cb1-541"><a href="#cb1-541" aria-hidden="true" tabindex="-1"></a><span class="co"># - adafactor</span></span>
|
||||
<span id="cb1-542"><a href="#cb1-542" aria-hidden="true" tabindex="-1"></a><span class="co"># - adamw_anyprecision</span></span>
|
||||
<span id="cb1-543"><a href="#cb1-543" aria-hidden="true" tabindex="-1"></a><span class="co"># - adamw_torch_4bit</span></span>
|
||||
<span id="cb1-544"><a href="#cb1-544" aria-hidden="true" tabindex="-1"></a><span class="co"># - ademamix</span></span>
|
||||
<span id="cb1-545"><a href="#cb1-545" aria-hidden="true" tabindex="-1"></a><span class="co"># - sgd</span></span>
|
||||
<span id="cb1-546"><a href="#cb1-546" aria-hidden="true" tabindex="-1"></a><span class="co"># - adagrad</span></span>
|
||||
<span id="cb1-547"><a href="#cb1-547" aria-hidden="true" tabindex="-1"></a><span class="co"># - adamw_bnb_8bit</span></span>
|
||||
<span id="cb1-548"><a href="#cb1-548" aria-hidden="true" tabindex="-1"></a><span class="co"># - adamw_8bit # alias for adamw_bnb_8bit</span></span>
|
||||
<span id="cb1-549"><a href="#cb1-549" aria-hidden="true" tabindex="-1"></a><span class="co"># - ademamix_8bit</span></span>
|
||||
<span id="cb1-550"><a href="#cb1-550" aria-hidden="true" tabindex="-1"></a><span class="co"># - lion_8bit</span></span>
|
||||
<span id="cb1-551"><a href="#cb1-551" aria-hidden="true" tabindex="-1"></a><span class="co"># - lion_32bit</span></span>
|
||||
<span id="cb1-552"><a href="#cb1-552" aria-hidden="true" tabindex="-1"></a><span class="co"># - paged_adamw_32bit</span></span>
|
||||
<span id="cb1-553"><a href="#cb1-553" aria-hidden="true" tabindex="-1"></a><span class="co"># - paged_adamw_8bit</span></span>
|
||||
<span id="cb1-554"><a href="#cb1-554" aria-hidden="true" tabindex="-1"></a><span class="co"># - paged_ademamix_32bit</span></span>
|
||||
<span id="cb1-555"><a href="#cb1-555" aria-hidden="true" tabindex="-1"></a><span class="co"># - paged_ademamix_8bit</span></span>
|
||||
<span id="cb1-556"><a href="#cb1-556" aria-hidden="true" tabindex="-1"></a><span class="co"># - paged_lion_32bit</span></span>
|
||||
<span id="cb1-557"><a href="#cb1-557" aria-hidden="true" tabindex="-1"></a><span class="co"># - paged_lion_8bit</span></span>
|
||||
<span id="cb1-558"><a href="#cb1-558" aria-hidden="true" tabindex="-1"></a><span class="co"># - rmsprop</span></span>
|
||||
<span id="cb1-559"><a href="#cb1-559" aria-hidden="true" tabindex="-1"></a><span class="co"># - rmsprop_bnb</span></span>
|
||||
<span id="cb1-560"><a href="#cb1-560" aria-hidden="true" tabindex="-1"></a><span class="co"># - rmsprop_bnb_8bit</span></span>
|
||||
<span id="cb1-561"><a href="#cb1-561" aria-hidden="true" tabindex="-1"></a><span class="co"># - rmsprop_bnb_32bit</span></span>
|
||||
<span id="cb1-562"><a href="#cb1-562" aria-hidden="true" tabindex="-1"></a><span class="co"># - galore_adamw</span></span>
|
||||
<span id="cb1-563"><a href="#cb1-563" aria-hidden="true" tabindex="-1"></a><span class="co"># - galore_adamw_8bit</span></span>
|
||||
<span id="cb1-564"><a href="#cb1-564" aria-hidden="true" tabindex="-1"></a><span class="co"># - galore_adafactor</span></span>
|
||||
<span id="cb1-565"><a href="#cb1-565" aria-hidden="true" tabindex="-1"></a><span class="co"># - galore_adamw_layerwise</span></span>
|
||||
<span id="cb1-566"><a href="#cb1-566" aria-hidden="true" tabindex="-1"></a><span class="co"># - galore_adamw_8bit_layerwise</span></span>
|
||||
<span id="cb1-567"><a href="#cb1-567" aria-hidden="true" tabindex="-1"></a><span class="co"># - galore_adafactor_layerwise</span></span>
|
||||
<span id="cb1-568"><a href="#cb1-568" aria-hidden="true" tabindex="-1"></a><span class="co"># - lomo</span></span>
|
||||
<span id="cb1-569"><a href="#cb1-569" aria-hidden="true" tabindex="-1"></a><span class="co"># - adalomo</span></span>
|
||||
<span id="cb1-570"><a href="#cb1-570" aria-hidden="true" tabindex="-1"></a><span class="co"># - grokadamw</span></span>
|
||||
<span id="cb1-571"><a href="#cb1-571" aria-hidden="true" tabindex="-1"></a><span class="co"># - schedule_free_adamw</span></span>
|
||||
<span id="cb1-572"><a href="#cb1-572" aria-hidden="true" tabindex="-1"></a><span class="co"># - schedule_free_sgd</span></span>
|
||||
<span id="cb1-573"><a href="#cb1-573" aria-hidden="true" tabindex="-1"></a><span class="co"># - apollo_adamw</span></span>
|
||||
<span id="cb1-574"><a href="#cb1-574" aria-hidden="true" tabindex="-1"></a><span class="co"># - apollo_adamw_layerwise</span></span>
|
||||
<span id="cb1-575"><a href="#cb1-575" aria-hidden="true" tabindex="-1"></a><span class="co">#</span></span>
|
||||
<span id="cb1-576"><a href="#cb1-576" aria-hidden="true" tabindex="-1"></a><span class="co"># Additional custom optimizers include:</span></span>
|
||||
<span id="cb1-577"><a href="#cb1-577" aria-hidden="true" tabindex="-1"></a><span class="co"># - optimi_adamw</span></span>
|
||||
<span id="cb1-578"><a href="#cb1-578" aria-hidden="true" tabindex="-1"></a><span class="co"># - ao_adamw_8bit</span></span>
|
||||
<span id="cb1-579"><a href="#cb1-579" aria-hidden="true" tabindex="-1"></a><span class="co"># - ao_adamw_fp8</span></span>
|
||||
<span id="cb1-580"><a href="#cb1-580" aria-hidden="true" tabindex="-1"></a><span class="fu">optimizer</span><span class="kw">:</span></span>
|
||||
<span id="cb1-581"><a href="#cb1-581" aria-hidden="true" tabindex="-1"></a><span class="co"># Dictionary of arguments to pass to the optimizer</span></span>
|
||||
<span id="cb1-582"><a href="#cb1-582" aria-hidden="true" tabindex="-1"></a><span class="fu">optim_args</span><span class="kw">:</span></span>
|
||||
<span id="cb1-583"><a href="#cb1-583" aria-hidden="true" tabindex="-1"></a><span class="co"># For Galore Optimizers the following optim_args are available</span></span>
|
||||
<span id="cb1-584"><a href="#cb1-584" aria-hidden="true" tabindex="-1"></a><span class="co"># rank: # type: int</span></span>
|
||||
<span id="cb1-585"><a href="#cb1-585" aria-hidden="true" tabindex="-1"></a><span class="co"># update_proj_gap # type: int</span></span>
|
||||
<span id="cb1-586"><a href="#cb1-586" aria-hidden="true" tabindex="-1"></a><span class="co"># scale # type: float</span></span>
|
||||
<span id="cb1-587"><a href="#cb1-587" aria-hidden="true" tabindex="-1"></a><span class="co"># proj_type: # type: str, default = std</span></span>
|
||||
<span id="cb1-588"><a href="#cb1-588" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-589"><a href="#cb1-589" 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="cb1-590"><a href="#cb1-590" aria-hidden="true" tabindex="-1"></a><span class="fu">optim_target_modules</span><span class="kw">:</span></span>
|
||||
<span id="cb1-591"><a href="#cb1-591" aria-hidden="true" tabindex="-1"></a><span class="co"># - self_attn # for llama</span></span>
|
||||
<span id="cb1-592"><a href="#cb1-592" aria-hidden="true" tabindex="-1"></a><span class="co"># - mlp</span></span>
|
||||
<span id="cb1-593"><a href="#cb1-593" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-594"><a href="#cb1-594" aria-hidden="true" tabindex="-1"></a><span class="co"># Specify weight decay</span></span>
|
||||
<span id="cb1-595"><a href="#cb1-595" aria-hidden="true" tabindex="-1"></a><span class="fu">weight_decay</span><span class="kw">:</span></span>
|
||||
<span id="cb1-596"><a href="#cb1-596" aria-hidden="true" tabindex="-1"></a><span class="co"># adamw hyperparams</span></span>
|
||||
<span id="cb1-597"><a href="#cb1-597" aria-hidden="true" tabindex="-1"></a><span class="fu">adam_beta1</span><span class="kw">:</span></span>
|
||||
<span id="cb1-598"><a href="#cb1-598" aria-hidden="true" tabindex="-1"></a><span class="fu">adam_beta2</span><span class="kw">:</span></span>
|
||||
<span id="cb1-599"><a href="#cb1-599" aria-hidden="true" tabindex="-1"></a><span class="fu">adam_epsilon</span><span class="kw">:</span></span>
|
||||
<span id="cb1-600"><a href="#cb1-600" aria-hidden="true" tabindex="-1"></a><span class="co"># Gradient clipping max norm</span></span>
|
||||
<span id="cb1-601"><a href="#cb1-601" aria-hidden="true" tabindex="-1"></a><span class="fu">max_grad_norm</span><span class="kw">:</span></span>
|
||||
<span id="cb1-602"><a href="#cb1-602" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-603"><a href="#cb1-603" aria-hidden="true" tabindex="-1"></a><span class="co"># Augmentation techniques</span></span>
|
||||
<span id="cb1-604"><a href="#cb1-604" 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="cb1-605"><a href="#cb1-605" aria-hidden="true" tabindex="-1"></a><span class="co"># currently only supported on Llama and Mistral</span></span>
|
||||
<span id="cb1-606"><a href="#cb1-606" aria-hidden="true" tabindex="-1"></a><span class="fu">neftune_noise_alpha</span><span class="kw">:</span></span>
|
||||
<span id="cb1-607"><a href="#cb1-607" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-608"><a href="#cb1-608" aria-hidden="true" tabindex="-1"></a><span class="co"># Optional[bool]. Whether to bettertransformers</span></span>
|
||||
<span id="cb1-609"><a href="#cb1-609" aria-hidden="true" tabindex="-1"></a><span class="fu">flash_optimum</span><span class="kw">:</span></span>
|
||||
<span id="cb1-610"><a href="#cb1-610" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-611"><a href="#cb1-611" aria-hidden="true" tabindex="-1"></a><span class="co"># Note: Only one of the following attention patches can be used at a time.</span></span>
|
||||
<span id="cb1-612"><a href="#cb1-612" aria-hidden="true" tabindex="-1"></a><span class="co"># For example, if you set `xformers_attention` to `true`, do not set `flash_attention` to `true`.</span></span>
|
||||
<span id="cb1-613"><a href="#cb1-613" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-614"><a href="#cb1-614" aria-hidden="true" tabindex="-1"></a><span class="co"># Optional[bool]. Whether to use xformers attention patch https://github.com/facebookresearch/xformers:</span></span>
|
||||
<span id="cb1-615"><a href="#cb1-615" aria-hidden="true" tabindex="-1"></a><span class="fu">xformers_attention</span><span class="kw">:</span></span>
|
||||
<span id="cb1-616"><a href="#cb1-616" aria-hidden="true" tabindex="-1"></a><span class="co"># Optional[bool]. Whether to use flash attention patch https://github.com/Dao-AILab/flash-attention:</span></span>
|
||||
<span id="cb1-617"><a href="#cb1-617" aria-hidden="true" tabindex="-1"></a><span class="fu">flash_attention</span><span class="kw">:</span></span>
|
||||
<span id="cb1-618"><a href="#cb1-618" aria-hidden="true" tabindex="-1"></a><span class="fu">flash_attn_cross_entropy</span><span class="kw">:</span><span class="co"> # Optional[bool]. Whether to use flash-attention cross entropy implementation - advanced use only</span></span>
|
||||
<span id="cb1-619"><a href="#cb1-619" aria-hidden="true" tabindex="-1"></a><span class="fu">flash_attn_rms_norm</span><span class="kw">:</span><span class="co"> # Optional[bool]. Whether to use flash-attention rms norm implementation - advanced use only</span></span>
|
||||
<span id="cb1-620"><a href="#cb1-620" aria-hidden="true" tabindex="-1"></a><span class="fu">flash_attn_fuse_qkv</span><span class="kw">:</span><span class="co"> # Optional[bool]. Whether to fuse QKV into a single operation</span></span>
|
||||
<span id="cb1-621"><a href="#cb1-621" aria-hidden="true" tabindex="-1"></a><span class="fu">flash_attn_fuse_mlp</span><span class="kw">:</span><span class="co"> # Optional[bool]. Whether to fuse part of the MLP into a single operation</span></span>
|
||||
<span id="cb1-622"><a href="#cb1-622" aria-hidden="true" tabindex="-1"></a><span class="co"># Optional[bool]. Whether to use scaled-dot-product attention</span></span>
|
||||
<span id="cb1-623"><a href="#cb1-623" 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="cb1-624"><a href="#cb1-624" aria-hidden="true" tabindex="-1"></a><span class="fu">sdp_attention</span><span class="kw">:</span></span>
|
||||
<span id="cb1-625"><a href="#cb1-625" aria-hidden="true" tabindex="-1"></a><span class="co"># Optional[bool]. Shifted-sparse attention (only llama) - https://arxiv.org/pdf/2309.12307.pdf</span></span>
|
||||
<span id="cb1-626"><a href="#cb1-626" aria-hidden="true" tabindex="-1"></a><span class="fu">s2_attention</span><span class="kw">:</span></span>
|
||||
<span id="cb1-627"><a href="#cb1-627" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-628"><a href="#cb1-628" aria-hidden="true" tabindex="-1"></a><span class="co"># Optional[bool]. Whether to use low_cpu_mem_usage</span></span>
|
||||
<span id="cb1-629"><a href="#cb1-629" aria-hidden="true" tabindex="-1"></a><span class="fu">low_cpu_mem_usage</span><span class="kw">:</span></span>
|
||||
<span id="cb1-630"><a href="#cb1-630" aria-hidden="true" tabindex="-1"></a><span class="co"># Optional[str]. Resume from a specific checkpoint dir</span></span>
|
||||
<span id="cb1-631"><a href="#cb1-631" aria-hidden="true" tabindex="-1"></a><span class="fu">resume_from_checkpoint</span><span class="kw">:</span></span>
|
||||
<span id="cb1-632"><a href="#cb1-632" aria-hidden="true" tabindex="-1"></a><span class="co"># Optional[bool]. If resume_from_checkpoint isn't set and you simply want it to start where it left off.</span></span>
|
||||
<span id="cb1-633"><a href="#cb1-633" aria-hidden="true" tabindex="-1"></a><span class="co"># Be careful with this being turned on between different models.</span></span>
|
||||
<span id="cb1-634"><a href="#cb1-634" 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="cb1-635"><a href="#cb1-635" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-636"><a href="#cb1-636" aria-hidden="true" tabindex="-1"></a><span class="co">## Multimodal section</span></span>
|
||||
<span id="cb1-637"><a href="#cb1-637" aria-hidden="true" tabindex="-1"></a><span class="co"># int | tuple[int, int] | None . Size to resize images to, width x height.</span></span>
|
||||
<span id="cb1-638"><a href="#cb1-638" aria-hidden="true" tabindex="-1"></a><span class="co"># Will read from model/processor config if not set.</span></span>
|
||||
<span id="cb1-639"><a href="#cb1-639" aria-hidden="true" tabindex="-1"></a><span class="fu">image_size</span><span class="kw">:</span></span>
|
||||
<span id="cb1-640"><a href="#cb1-640" aria-hidden="true" tabindex="-1"></a><span class="co"># str. Algorithm to use for image resizing. "bilinear", "bicubic", "lanczos". Default is "bilinear".</span></span>
|
||||
<span id="cb1-641"><a href="#cb1-641" aria-hidden="true" tabindex="-1"></a><span class="fu">image_resize_algorithm</span><span class="kw">:</span><span class="at"> </span><span class="st">'bilinear'</span></span>
|
||||
<span id="cb1-642"><a href="#cb1-642" aria-hidden="true" tabindex="-1"></a><span class="co">## End of multimodal section</span></span>
|
||||
<span id="cb1-643"><a href="#cb1-643" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-644"><a href="#cb1-644" 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="cb1-645"><a href="#cb1-645" aria-hidden="true" tabindex="-1"></a><span class="fu">local_rank</span><span class="kw">:</span></span>
|
||||
<span id="cb1-646"><a href="#cb1-646" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-647"><a href="#cb1-647" aria-hidden="true" tabindex="-1"></a><span class="co"># Add or change special tokens.</span></span>
|
||||
<span id="cb1-648"><a href="#cb1-648" 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="cb1-649"><a href="#cb1-649" aria-hidden="true" tabindex="-1"></a><span class="fu">special_tokens</span><span class="kw">:</span></span>
|
||||
<span id="cb1-650"><a href="#cb1-650" aria-hidden="true" tabindex="-1"></a><span class="co"> # bos_token: "<s>"</span></span>
|
||||
<span id="cb1-651"><a href="#cb1-651" aria-hidden="true" tabindex="-1"></a><span class="co"> # eos_token: "</s>"</span></span>
|
||||
<span id="cb1-652"><a href="#cb1-652" aria-hidden="true" tabindex="-1"></a><span class="co"> # unk_token: "<unk>"</span></span>
|
||||
<span id="cb1-653"><a href="#cb1-653" aria-hidden="true" tabindex="-1"></a><span class="co"> # pad_token: "[PAD]"</span></span>
|
||||
<span id="cb1-654"><a href="#cb1-654" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-655"><a href="#cb1-655" aria-hidden="true" tabindex="-1"></a><span class="co"># Add extra tokens.</span></span>
|
||||
<span id="cb1-656"><a href="#cb1-656" aria-hidden="true" tabindex="-1"></a><span class="fu">tokens</span><span class="kw">:</span></span>
|
||||
<span id="cb1-657"><a href="#cb1-657" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-658"><a href="#cb1-658" aria-hidden="true" tabindex="-1"></a><span class="co"># Mapping token_id to new_token_string to override reserved added_tokens in the tokenizer.</span></span>
|
||||
<span id="cb1-659"><a href="#cb1-659" aria-hidden="true" tabindex="-1"></a><span class="co"># Only works for tokens that are not part of the base vocab (aka are added_tokens).</span></span>
|
||||
<span id="cb1-660"><a href="#cb1-660" aria-hidden="true" tabindex="-1"></a><span class="co"># Can be checked if they exist in tokenizer.json added_tokens.</span></span>
|
||||
<span id="cb1-661"><a href="#cb1-661" aria-hidden="true" tabindex="-1"></a><span class="fu">added_tokens_overrides</span><span class="kw">:</span><span class="co"> # Dict[int, str]</span></span>
|
||||
<span id="cb1-662"><a href="#cb1-662" aria-hidden="true" tabindex="-1"></a><span class="co"># 128041: "<|im_start|>"</span></span>
|
||||
<span id="cb1-663"><a href="#cb1-663" aria-hidden="true" tabindex="-1"></a><span class="co"># 128042: "<|im_end|>"</span></span>
|
||||
<span id="cb1-664"><a href="#cb1-664" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-665"><a href="#cb1-665" aria-hidden="true" tabindex="-1"></a><span class="co"># FSDP</span></span>
|
||||
<span id="cb1-666"><a href="#cb1-666" aria-hidden="true" tabindex="-1"></a><span class="fu">fsdp</span><span class="kw">:</span></span>
|
||||
<span id="cb1-667"><a href="#cb1-667" aria-hidden="true" tabindex="-1"></a><span class="fu">fsdp_config</span><span class="kw">:</span></span>
|
||||
<span id="cb1-668"><a href="#cb1-668" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-669"><a href="#cb1-669" aria-hidden="true" tabindex="-1"></a><span class="co"># Deepspeed config path. e.g., deepspeed_configs/zero3.json</span></span>
|
||||
<span id="cb1-670"><a href="#cb1-670" aria-hidden="true" tabindex="-1"></a><span class="fu">deepspeed</span><span class="kw">:</span></span>
|
||||
<span id="cb1-671"><a href="#cb1-671" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-672"><a href="#cb1-672" aria-hidden="true" tabindex="-1"></a><span class="co"># Advanced DDP Arguments</span></span>
|
||||
<span id="cb1-673"><a href="#cb1-673" aria-hidden="true" tabindex="-1"></a><span class="fu">ddp_timeout</span><span class="kw">:</span></span>
|
||||
<span id="cb1-674"><a href="#cb1-674" aria-hidden="true" tabindex="-1"></a><span class="fu">ddp_bucket_cap_mb</span><span class="kw">:</span></span>
|
||||
<span id="cb1-675"><a href="#cb1-675" aria-hidden="true" tabindex="-1"></a><span class="fu">ddp_broadcast_buffers</span><span class="kw">:</span></span>
|
||||
<span id="cb1-676"><a href="#cb1-676" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-677"><a href="#cb1-677" aria-hidden="true" tabindex="-1"></a><span class="co"># Sequence parallelism</span></span>
|
||||
<span id="cb1-678"><a href="#cb1-678" aria-hidden="true" tabindex="-1"></a><span class="co"># Set to a divisor of the number of GPUs available to split sequences into chunks of equal size.</span></span>
|
||||
<span id="cb1-679"><a href="#cb1-679" aria-hidden="true" tabindex="-1"></a><span class="co"># Use in long context training to prevent OOM when sequences cannot fit into a single GPU's VRAM.</span></span>
|
||||
<span id="cb1-680"><a href="#cb1-680" aria-hidden="true" tabindex="-1"></a><span class="co"># E.g., if 4 GPUs are available, set this value to 2 to split each sequence into two equal-sized</span></span>
|
||||
<span id="cb1-681"><a href="#cb1-681" aria-hidden="true" tabindex="-1"></a><span class="co"># subsequences, or set to 4 to split into four equal-sized subsequences.</span></span>
|
||||
<span id="cb1-682"><a href="#cb1-682" aria-hidden="true" tabindex="-1"></a><span class="co"># See https://axolotl-ai-cloud.github.io/axolotl/docs/sequence_parallelism.html for more details.</span></span>
|
||||
<span id="cb1-683"><a href="#cb1-683" aria-hidden="true" tabindex="-1"></a><span class="fu">sequence_parallel_degree</span><span class="kw">:</span></span>
|
||||
<span id="cb1-684"><a href="#cb1-684" aria-hidden="true" tabindex="-1"></a><span class="co"># Optional; strides across the key dimension. Larger values use more memory but should make training faster.</span></span>
|
||||
<span id="cb1-685"><a href="#cb1-685" aria-hidden="true" tabindex="-1"></a><span class="co"># Must evenly divide the number of KV heads in your model.</span></span>
|
||||
<span id="cb1-686"><a href="#cb1-686" aria-hidden="true" tabindex="-1"></a><span class="fu">heads_k_stride</span><span class="kw">:</span><span class="at"> </span><span class="dv">1</span></span>
|
||||
<span id="cb1-687"><a href="#cb1-687" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-688"><a href="#cb1-688" aria-hidden="true" tabindex="-1"></a><span class="co"># Path to torch distx for optim 'adamw_anyprecision'</span></span>
|
||||
<span id="cb1-689"><a href="#cb1-689" aria-hidden="true" tabindex="-1"></a><span class="fu">torchdistx_path</span><span class="kw">:</span></span>
|
||||
<span id="cb1-690"><a href="#cb1-690" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-691"><a href="#cb1-691" 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="cb1-692"><a href="#cb1-692" aria-hidden="true" tabindex="-1"></a><span class="fu">pretraining_dataset</span><span class="kw">:</span></span>
|
||||
<span id="cb1-693"><a href="#cb1-693" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-694"><a href="#cb1-694" aria-hidden="true" tabindex="-1"></a><span class="co"># Debug mode</span></span>
|
||||
<span id="cb1-695"><a href="#cb1-695" aria-hidden="true" tabindex="-1"></a><span class="fu">debug</span><span class="kw">:</span></span>
|
||||
<span id="cb1-696"><a href="#cb1-696" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-697"><a href="#cb1-697" aria-hidden="true" tabindex="-1"></a><span class="co"># Seed</span></span>
|
||||
<span id="cb1-698"><a href="#cb1-698" aria-hidden="true" tabindex="-1"></a><span class="fu">seed</span><span class="kw">:</span></span>
|
||||
<span id="cb1-699"><a href="#cb1-699" aria-hidden="true" tabindex="-1"></a></span>
|
||||
<span id="cb1-700"><a href="#cb1-700" aria-hidden="true" tabindex="-1"></a><span class="co"># Allow overwrite yml config using from cli</span></span>
|
||||
<span id="cb1-701"><a href="#cb1-701" 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>
|
||||
|
||||
|
||||
|
||||
|
||||
File diff suppressed because one or more lines are too long
338
sitemap.xml
338
sitemap.xml
@@ -2,678 +2,678 @@
|
||||
<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>2025-03-31T21:15:37.300Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:12.959Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/dataset-formats/stepwise_supervised.html</loc>
|
||||
<lastmod>2025-03-31T21:15:37.295Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:12.955Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/dataset-formats/template_free.html</loc>
|
||||
<lastmod>2025-03-31T21:15:37.295Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:12.955Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/dataset-formats/tokenized.html</loc>
|
||||
<lastmod>2025-03-31T21:15:37.295Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:12.955Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/nccl.html</loc>
|
||||
<lastmod>2025-03-31T21:15:37.299Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:12.958Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/amd_hpc.html</loc>
|
||||
<lastmod>2025-03-31T21:15:37.295Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:12.954Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/config.html</loc>
|
||||
<lastmod>2025-03-31T21:15:37.295Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:12.954Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/multi-gpu.html</loc>
|
||||
<lastmod>2025-03-31T21:15:37.299Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:12.958Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/installation.html</loc>
|
||||
<lastmod>2025-03-31T21:15:37.299Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:12.958Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/torchao.html</loc>
|
||||
<lastmod>2025-03-31T21:15:37.299Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:12.958Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/reward_modelling.html</loc>
|
||||
<lastmod>2025-03-31T21:15:37.299Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:12.958Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/input_output.html</loc>
|
||||
<lastmod>2025-03-31T21:15:37.299Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:12.958Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/multimodal.html</loc>
|
||||
<lastmod>2025-03-31T21:15:37.299Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:12.958Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/utils.callbacks.mlflow_.html</loc>
|
||||
<lastmod>2025-03-31T21:16:16.158Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.952Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/monkeypatch.trainer_fsdp_optim.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.739Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.538Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/monkeypatch.data.batch_dataset_fetcher.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.755Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.554Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/prompt_strategies.stepwise_supervised.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.444Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.245Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/monkeypatch.mistral_attn_hijack_flash.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.686Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.486Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/prompt_strategies.dpo.user_defined.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.491Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.291Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/integrations.liger.args.html</loc>
|
||||
<lastmod>2025-03-31T21:16:16.071Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.866Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/utils.schemas.training.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.932Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.730Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/datasets.html</loc>
|
||||
<lastmod>2025-03-31T21:16:14.934Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:42.739Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/kernels.geglu.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.626Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.427Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/monkeypatch.llama_attn_hijack_flash.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.670Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.471Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/cli.sweeps.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.273Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.076Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/utils.freeze.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.832Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.632Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/monkeypatch.multipack.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.688Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.488Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/cli.main.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.168Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:42.970Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/core.trainers.trl.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.353Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.155Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/prompt_strategies.dpo.passthrough.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.492Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.293Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/core.chat.format.llama3x.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.122Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:42.925Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/core.datasets.transforms.chat_builder.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.137Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:42.939Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/prompt_strategies.kto.user_defined.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.509Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.310Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/utils.collators.mamba.html</loc>
|
||||
<lastmod>2025-03-31T21:16:16.129Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.924Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/integrations.base.html</loc>
|
||||
<lastmod>2025-03-31T21:16:16.055Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.851Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/utils.bench.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.824Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.624Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/kernels.swiglu.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.636Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.437Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/core.chat.format.shared.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.124Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:42.926Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/integrations.cut_cross_entropy.args.html</loc>
|
||||
<lastmod>2025-03-31T21:16:16.059Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.855Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/core.datasets.chat.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.129Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:42.931Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/utils.callbacks.lisa.html</loc>
|
||||
<lastmod>2025-03-31T21:16:16.154Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.948Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/integrations.grokfast.optimizer.html</loc>
|
||||
<lastmod>2025-03-31T21:16:16.060Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.856Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/prompt_strategies.alpaca_chat.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.393Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.195Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/prompt_strategies.alpaca_instruct.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.395Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.196Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/prompt_strategies.kto.chatml.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.508Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.309Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/utils.schemas.integrations.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.979Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.776Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/utils.schemas.trl.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.961Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.759Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/prompt_tokenizers.html</loc>
|
||||
<lastmod>2025-03-31T21:16:14.990Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:42.794Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/utils.data.sft.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.909Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.707Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/utils.schedulers.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.874Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.673Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/utils.chat_templates.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.806Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.601Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/utils.models.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.788Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.584Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/prompt_strategies.dpo.chatml.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.488Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.288Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/utils.distributed.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.895Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.693Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/monkeypatch.utils.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.727Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.526Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/utils.schemas.utils.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.991Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.789Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/monkeypatch.llama_expand_mask.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.696Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.496Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/common.datasets.html</loc>
|
||||
<lastmod>2025-03-31T21:16:16.097Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.892Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/logging_config.html</loc>
|
||||
<lastmod>2025-03-31T21:16:14.995Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:42.799Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/kernels.quantize.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.643Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.444Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/monkeypatch.llama_patch_multipack.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.730Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.529Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/utils.schemas.model.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.927Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.725Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/monkeypatch.stablelm_attn_hijack_flash.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.735Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.535Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/monkeypatch.mixtral.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.756Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.555Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/utils.tokenization.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.796Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.591Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/integrations.kd.trainer.html</loc>
|
||||
<lastmod>2025-03-31T21:16:16.067Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.863Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/utils.schemas.datasets.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.950Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.747Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/utils.collators.core.html</loc>
|
||||
<lastmod>2025-03-31T21:16:16.099Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.895Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/monkeypatch.btlm_attn_hijack_flash.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.728Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.528Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/utils.optimizers.adopt.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.906Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.704Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/prompt_strategies.input_output.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.439Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.241Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/index.html</loc>
|
||||
<lastmod>2025-03-31T21:16:14.856Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:42.661Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/cli.cloud.modal_.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.319Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.121Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/prompt_strategies.dpo.llama3.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.477Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.278Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/cli.train.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.176Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:42.978Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/core.trainer_builder.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.010Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:42.814Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/utils.callbacks.perplexity.html</loc>
|
||||
<lastmod>2025-03-31T21:16:16.149Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.943Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/getting-started.html</loc>
|
||||
<lastmod>2025-03-31T21:15:37.296Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:12.955Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/inference.html</loc>
|
||||
<lastmod>2025-03-31T21:15:37.299Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:12.958Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/multipack.html</loc>
|
||||
<lastmod>2025-03-31T21:15:37.299Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:12.958Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/sequence_parallelism.html</loc>
|
||||
<lastmod>2025-03-31T21:15:37.299Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:12.958Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/multi-node.html</loc>
|
||||
<lastmod>2025-03-31T21:15:37.299Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:12.958Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/FAQS.html</loc>
|
||||
<lastmod>2025-03-31T21:15:37.294Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:12.953Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/src/axolotl/integrations/cut_cross_entropy/ACKNOWLEDGEMENTS.html</loc>
|
||||
<lastmod>2025-03-31T21:15:37.315Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:12.975Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/index.html</loc>
|
||||
<lastmod>2025-03-31T21:15:37.311Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:12.971Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/src/axolotl/integrations/LICENSE.html</loc>
|
||||
<lastmod>2025-03-31T21:15:37.314Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:12.975Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/TODO.html</loc>
|
||||
<lastmod>2025-03-31T21:15:37.294Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:12.953Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/lr_groups.html</loc>
|
||||
<lastmod>2025-03-31T21:15:37.299Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:12.958Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/debugging.html</loc>
|
||||
<lastmod>2025-03-31T21:15:37.296Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:12.955Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/faq.html</loc>
|
||||
<lastmod>2025-03-31T21:15:37.296Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:12.955Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/batch_vs_grad.html</loc>
|
||||
<lastmod>2025-03-31T21:15:37.295Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:12.954Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/lora_optims.html</loc>
|
||||
<lastmod>2025-03-31T21:15:37.299Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:12.958Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/utils.lora_embeddings.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.815Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.611Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/kernels.utils.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.645Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.446Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/prompt_strategies.chat_template.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.380Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.181Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/convert.html</loc>
|
||||
<lastmod>2025-03-31T21:16:14.948Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:42.752Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/common.const.html</loc>
|
||||
<lastmod>2025-03-31T21:16:16.080Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.876Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/cli.cloud.base.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.313Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.115Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/monkeypatch.relora.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.695Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.495Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/utils.lora.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.811Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.606Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/cli.merge_lora.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.248Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.050Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/prompt_strategies.bradley_terry.llama3.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.534Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.335Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/cli.merge_sharded_fsdp_weights.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.259Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.061Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/integrations.spectrum.args.html</loc>
|
||||
<lastmod>2025-03-31T21:16:16.077Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.873Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/models.mamba.modeling_mamba.html</loc>
|
||||
<lastmod>2025-03-31T21:16:16.098Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.893Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/common.architectures.html</loc>
|
||||
<lastmod>2025-03-31T21:16:16.079Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.874Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/utils.trainer.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.850Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.649Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/utils.callbacks.comet_.html</loc>
|
||||
<lastmod>2025-03-31T21:16:16.162Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.955Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/cli.vllm_serve.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.309Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.111Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/utils.schemas.multimodal.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.967Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.764Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/utils.gradient_checkpointing.unsloth.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.912Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.710Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/core.trainers.base.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.336Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.138Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/monkeypatch.unsloth_.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.747Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.546Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/utils.samplers.multipack.html</loc>
|
||||
<lastmod>2025-03-31T21:16:16.143Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.937Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/utils.callbacks.profiler.html</loc>
|
||||
<lastmod>2025-03-31T21:16:16.153Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.947Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/integrations.lm_eval.args.html</loc>
|
||||
<lastmod>2025-03-31T21:16:16.074Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.869Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/utils.data.pretraining.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.908Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.706Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/evaluate.html</loc>
|
||||
<lastmod>2025-03-31T21:16:14.927Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:42.731Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/utils.dict.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.898Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.697Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/cli.utils.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.305Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.107Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/prompt_strategies.pygmalion.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.462Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.262Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/core.training_args.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.096Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:42.899Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/cli.inference.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.239Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.041Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/kernels.lora.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.615Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.416Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/cli.evaluate.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.185Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:42.986Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/utils.collators.batching.html</loc>
|
||||
<lastmod>2025-03-31T21:16:16.126Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.920Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/prompt_strategies.completion.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.433Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.235Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/prompt_strategies.dpo.zephyr.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.489Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.290Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/prompt_strategies.metharme.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.451Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.252Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/prompt_strategies.orpo.chat_template.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.530Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.331Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/prompt_strategies.alpaca_w_system.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.407Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.208Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/utils.model_shard_quant.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.820Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.620Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/cli.config.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.225Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.027Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/utils.schemas.enums.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.985Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.783Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/cli.preprocess.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.267Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.070Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/core.chat.messages.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.119Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:42.922Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/prompt_strategies.dpo.chat_template.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.467Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.268Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/utils.schemas.peft.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.958Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.756Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/train.html</loc>
|
||||
<lastmod>2025-03-31T21:16:14.917Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:42.721Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/prompt_strategies.messages.chat.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.466Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.267Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/prompt_strategies.orcamini.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.455Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.256Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/utils.collators.mm_chat.html</loc>
|
||||
<lastmod>2025-03-31T21:16:16.134Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.929Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/prompt_strategies.kto.llama3.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.500Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.301Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/monkeypatch.attention.mllama.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.753Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.553Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/cli.checks.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.208Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.010Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/monkeypatch.transformers_fa_utils.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.745Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.544Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/monkeypatch.llama_attn_hijack_xformers.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.672Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.472Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/core.trainers.dpo.trainer.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.360Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.162Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/prompt_strategies.user_defined.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.415Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.216Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/cli.args.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.202Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.003Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/prompt_strategies.llama2_chat.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.428Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.229Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/utils.schemas.config.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.920Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.718Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/core.trainers.grpo.trainer.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.363Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.165Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/core.chat.format.chatml.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.121Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:42.923Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/monkeypatch.lora_kernels.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.719Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.518Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/api/prompt_strategies.base.html</loc>
|
||||
<lastmod>2025-03-31T21:16:15.365Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:43.167Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/rlhf.html</loc>
|
||||
<lastmod>2025-03-31T21:15:37.299Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:12.958Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/cli.html</loc>
|
||||
<lastmod>2025-03-31T21:15:37.295Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:12.954Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/unsloth.html</loc>
|
||||
<lastmod>2025-03-31T21:15:37.299Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:12.958Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/fsdp_qlora.html</loc>
|
||||
<lastmod>2025-03-31T21:15:37.296Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:12.955Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/dataset_preprocessing.html</loc>
|
||||
<lastmod>2025-03-31T21:15:37.296Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:12.955Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/custom_integrations.html</loc>
|
||||
<lastmod>2025-03-31T21:15:37.295Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:12.954Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/mac.html</loc>
|
||||
<lastmod>2025-03-31T21:15:37.299Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:12.958Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/docker.html</loc>
|
||||
<lastmod>2025-03-31T21:15:37.296Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:12.955Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/ray-integration.html</loc>
|
||||
<lastmod>2025-03-31T21:15:37.299Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:12.958Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/dataset-formats/index.html</loc>
|
||||
<lastmod>2025-03-31T21:15:37.295Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:12.955Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/dataset-formats/conversation.html</loc>
|
||||
<lastmod>2025-03-31T21:15:37.295Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:12.954Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/dataset-formats/pretraining.html</loc>
|
||||
<lastmod>2025-03-31T21:15:37.295Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:12.955Z</lastmod>
|
||||
</url>
|
||||
<url>
|
||||
<loc>https://axolotl-ai-cloud.github.io/axolotl/docs/dataset-formats/inst_tune.html</loc>
|
||||
<lastmod>2025-03-31T21:15:37.295Z</lastmod>
|
||||
<lastmod>2025-04-01T13:20:12.955Z</lastmod>
|
||||
</url>
|
||||
</urlset>
|
||||
|
||||
Reference in New Issue
Block a user