diff --git a/.nojekyll b/.nojekyll index adb0eac46..ce3b02b59 100644 --- a/.nojekyll +++ b/.nojekyll @@ -1 +1 @@ -6fea3a42 \ No newline at end of file +2aa75668 \ No newline at end of file diff --git a/docs/custom_integrations.html b/docs/custom_integrations.html index 11dfc0712..44c037289 100644 --- a/docs/custom_integrations.html +++ b/docs/custom_integrations.html @@ -475,6 +475,7 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true});
  • Supported Models
  • Citation
    • +
  • DenseMixer
  • Grokfast @@ -609,25 +610,33 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true});

    Please see reference here

    +
    +

    DenseMixer

    +

    See DenseMixer

    +

    Simply add the following to your axolotl YAML config:

    +
    plugins:
+  - axolotl.integrations.densemixer.DenseMixerPlugin
    +

    Please see reference here

    +

    Grokfast

    See https://github.com/ironjr/grokfast

    Usage

    -
    plugins:
-  - axolotl.integrations.grokfast.GrokfastPlugin
-
-grokfast_alpha: 2.0
-grokfast_lamb: 0.98
    +
    plugins:
+  - axolotl.integrations.grokfast.GrokfastPlugin
+
+grokfast_alpha: 2.0
+grokfast_lamb: 0.98

    Citation

    -
    @article{lee2024grokfast,
-    title={{Grokfast}: Accelerated Grokking by Amplifying Slow Gradients},
-    author={Lee, Jaerin and Kang, Bong Gyun and Kim, Kihoon and Lee, Kyoung Mu},
-    journal={arXiv preprint arXiv:2405.20233},
-    year={2024}
-}
    +
    @article{lee2024grokfast,
+    title={{Grokfast}: Accelerated Grokking by Amplifying Slow Gradients},
+    author={Lee, Jaerin and Kang, Bong Gyun and Kim, Kihoon and Lee, Kyoung Mu},
+    journal={arXiv preprint arXiv:2405.20233},
+    year={2024}
+}

    Please see reference here

    @@ -635,145 +644,25 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true});

    Knowledge Distillation (KD)

    Usage

    -
    plugins:
-  - "axolotl.integrations.kd.KDPlugin"
-
-kd_trainer: True
-kd_ce_alpha: 0.1
-kd_alpha: 0.9
-kd_temperature: 1.0
-
-torch_compile: True  # torch>=2.5.1, recommended to reduce vram
-
-datasets:
-  - path: ...
-    type: "axolotl.integrations.kd.chat_template"
-    field_messages: "messages_combined"
-    logprobs_field: "llm_text_generation_vllm_logprobs"  # for kd only, field of logprobs
    +
    plugins:
+  - "axolotl.integrations.kd.KDPlugin"
+
+kd_trainer: True
+kd_ce_alpha: 0.1
+kd_alpha: 0.9
+kd_temperature: 1.0
+
+torch_compile: True  # torch>=2.5.1, recommended to reduce vram
+
+datasets:
+  - path: ...
+    type: "axolotl.integrations.kd.chat_template"
+    field_messages: "messages_combined"
+    logprobs_field: "llm_text_generation_vllm_logprobs"  # for kd only, field of logprobs

    An example dataset can be found at axolotl-ai-co/evolkit-logprobs-pipeline-75k-v2-sample

    Please see reference here

    -
    -

    Liger Kernels

    -

    Liger Kernel provides efficient Triton kernels for LLM training, offering:

    - -

    See https://github.com/linkedin/Liger-Kernel

    -
    -

    Usage

    -
    plugins:
-  - axolotl.integrations.liger.LigerPlugin
-liger_rope: true
-liger_rms_norm: true
-liger_glu_activation: true
-liger_layer_norm: true
-liger_fused_linear_cross_entropy: true
    -
    -
    -

    Supported Models

    -
      -
    • deepseek_v2
      • -
    • gemma
      • -
    • gemma2
      • -
    • gemma3
      • -
    • granite
      • -
    • jamba
      • -
    • llama
      • -
    • mistral
      • -
    • mixtral
      • -
    • mllama
      • -
    • mllama_text_model
      • -
    • olmo2
      • -
    • paligemma
      • -
    • phi3
      • -
    • qwen2
      • -
    • qwen2_5_vl
      • -
    • qwen2_vl
      • -
    -
    -
    -

    Citation

    -
    @article{hsu2024ligerkernelefficienttriton,
-      title={Liger Kernel: Efficient Triton Kernels for LLM Training},
-      author={Pin-Lun Hsu and Yun Dai and Vignesh Kothapalli and Qingquan Song and Shao Tang and Siyu Zhu and Steven Shimizu and Shivam Sahni and Haowen Ning and Yanning Chen},
-      year={2024},
-      eprint={2410.10989},
-      archivePrefix={arXiv},
-      primaryClass={cs.LG},
-      url={https://arxiv.org/abs/2410.10989},
-      journal={arXiv preprint arXiv:2410.10989},
-}
    -

    Please see reference here

    -
    -
    -
    -

    Language Model Evaluation Harness (LM Eval)

    -

    Run evaluation on model using the popular lm-evaluation-harness library.

    -

    See https://github.com/EleutherAI/lm-evaluation-harness

    -
    -

    Usage

    -
    plugins:
-  - axolotl.integrations.lm_eval.LMEvalPlugin
-
-lm_eval_tasks:
-  - gsm8k
-  - hellaswag
-  - arc_easy
-
-lm_eval_batch_size: # Batch size for evaluation
-output_dir: # Directory to save evaluation results
    -
    -
    -

    Citation

    -
    @misc{eval-harness,
-  author       = {Gao, Leo and Tow, Jonathan and Abbasi, Baber and Biderman, Stella and Black, Sid and DiPofi, Anthony and Foster, Charles and Golding, Laurence and Hsu, Jeffrey and Le Noac'h, Alain and Li, Haonan and McDonell, Kyle and Muennighoff, Niklas and Ociepa, Chris and Phang, Jason and Reynolds, Laria and Schoelkopf, Hailey and Skowron, Aviya and Sutawika, Lintang and Tang, Eric and Thite, Anish and Wang, Ben and Wang, Kevin and Zou, Andy},
-  title        = {A framework for few-shot language model evaluation},
-  month        = 07,
-  year         = 2024,
-  publisher    = {Zenodo},
-  version      = {v0.4.3},
-  doi          = {10.5281/zenodo.12608602},
-  url          = {https://zenodo.org/records/12608602}
-}
    -

    Please see reference here

    -
    -
    -
    -

    Spectrum

    -

    by Eric Hartford, Lucas Atkins, Fernando Fernandes, David Golchinfar

    -

    This plugin contains code to freeze the bottom fraction of modules in a model, based on the Signal-to-Noise Ratio (SNR).

    -

    See https://github.com/cognitivecomputations/spectrum

    -
    -

    Overview

    -

    Spectrum is a tool for scanning and evaluating the Signal-to-Noise Ratio (SNR) of layers in large language models. -By identifying the top n% of layers with the highest SNR, you can optimize training efficiency.

    -
    -
    -

    Usage

    -
    plugins:
-  - axolotl.integrations.spectrum.SpectrumPlugin
-
-spectrum_top_fraction: 0.5
-spectrum_model_name: meta-llama/Meta-Llama-3.1-8B
    -
    -
    -

    Citation

    -
    @misc{hartford2024spectrumtargetedtrainingsignal,
-      title={Spectrum: Targeted Training on Signal to Noise Ratio},
-      author={Eric Hartford and Lucas Atkins and Fernando Fernandes Neto and David Golchinfar},
-      year={2024},
-      eprint={2406.06623},
-      archivePrefix={arXiv},
-      primaryClass={cs.LG},
-      url={https://arxiv.org/abs/2406.06623},
-}
    -

    Please see reference here

    -
    -

    LLMCompressor

    Fine-tune sparsified models in Axolotl using Neural Magic’s LLMCompressor.

    @@ -784,34 +673,34 @@ By identifying the top n% of layers with the highest SNR, you can optimize train

    Requirements

    This will install all necessary dependencies to fine-tune sparsified models using the integration.

    -
    -

    Usage

    +
    +

    Usage

    To enable sparse fine-tuning with this integration, include the plugin in your Axolotl config:

    -
    plugins:
-  - axolotl.integrations.llm_compressor.LLMCompressorPlugin
-
-llmcompressor:
-  recipe:
-    finetuning_stage:
-      finetuning_modifiers:
-        ConstantPruningModifier:
-          targets: [
-            're:.*q_proj.weight',
-            're:.*k_proj.weight',
-            're:.*v_proj.weight',
-            're:.*o_proj.weight',
-            're:.*gate_proj.weight',
-            're:.*up_proj.weight',
-            're:.*down_proj.weight',
-          ]
-          start: 0
-  save_compressed: true
    +
    plugins:
+  - axolotl.integrations.llm_compressor.LLMCompressorPlugin
+
+llmcompressor:
+  recipe:
+    finetuning_stage:
+      finetuning_modifiers:
+        ConstantPruningModifier:
+          targets: [
+            're:.*q_proj.weight',
+            're:.*k_proj.weight',
+            're:.*v_proj.weight',
+            're:.*o_proj.weight',
+            're:.*gate_proj.weight',
+            're:.*up_proj.weight',
+            're:.*down_proj.weight',
+          ]
+          start: 0
+  save_compressed: true

    This plugin does not apply pruning or sparsification itself — it is intended for fine-tuning models that have already been sparsified.

    Pre-sparsified checkpoints can be: - Generated using LLMCompressor @@ -838,22 +727,22 @@ By identifying the top n% of layers with the highest SNR, you can optimize train

    After fine-tuning your sparse model, you can leverage vLLM for efficient inference. You can also use LLMCompressor to apply additional quantization to your fine-tuned sparse model before inference for even greater performance benefits.:

    -
    from vllm import LLM, SamplingParams
-
-prompts = [
-    "Hello, my name is",
-    "The president of the United States is",
-    "The capital of France is",
-    "The future of AI is",
-]
-sampling_params = SamplingParams(temperature=0.8, top_p=0.95)
-llm = LLM("path/to/your/sparse/model")
-outputs = llm.generate(prompts, sampling_params)
-
-for output in outputs:
-    prompt = output.prompt
-    generated_text = output.outputs[0].text
-    print(f"Prompt: {prompt!r}, Generated text: {generated_text!r}")
    +
    from vllm import LLM, SamplingParams
+
+prompts = [
+    "Hello, my name is",
+    "The president of the United States is",
+    "The capital of France is",
+    "The future of AI is",
+]
+sampling_params = SamplingParams(temperature=0.8, top_p=0.95)
+llm = LLM("path/to/your/sparse/model")
+outputs = llm.generate(prompts, sampling_params)
+
+for output in outputs:
+    prompt = output.prompt
+    generated_text = output.outputs[0].text
+    print(f"Prompt: {prompt!r}, Generated text: {generated_text!r}")

    For more details on vLLM’s capabilities and advanced configuration options, see the official vLLM documentation.

    @@ -863,6 +752,126 @@ sparse model before inference for even greater performance benefits.:

    Please see reference here

    +
    +

    Language Model Evaluation Harness (LM Eval)

    +

    Run evaluation on model using the popular lm-evaluation-harness library.

    +

    See https://github.com/EleutherAI/lm-evaluation-harness

    +
    +

    Usage

    +
    plugins:
+  - axolotl.integrations.lm_eval.LMEvalPlugin
+
+lm_eval_tasks:
+  - gsm8k
+  - hellaswag
+  - arc_easy
+
+lm_eval_batch_size: # Batch size for evaluation
+output_dir: # Directory to save evaluation results
    +
    +
    +

    Citation

    +
    @misc{eval-harness,
+  author       = {Gao, Leo and Tow, Jonathan and Abbasi, Baber and Biderman, Stella and Black, Sid and DiPofi, Anthony and Foster, Charles and Golding, Laurence and Hsu, Jeffrey and Le Noac'h, Alain and Li, Haonan and McDonell, Kyle and Muennighoff, Niklas and Ociepa, Chris and Phang, Jason and Reynolds, Laria and Schoelkopf, Hailey and Skowron, Aviya and Sutawika, Lintang and Tang, Eric and Thite, Anish and Wang, Ben and Wang, Kevin and Zou, Andy},
+  title        = {A framework for few-shot language model evaluation},
+  month        = 07,
+  year         = 2024,
+  publisher    = {Zenodo},
+  version      = {v0.4.3},
+  doi          = {10.5281/zenodo.12608602},
+  url          = {https://zenodo.org/records/12608602}
+}
    +

    Please see reference here

    +
    +
    +
    +

    Liger Kernels

    +

    Liger Kernel provides efficient Triton kernels for LLM training, offering:

    + +

    See https://github.com/linkedin/Liger-Kernel

    +
    +

    Usage

    +
    plugins:
+  - axolotl.integrations.liger.LigerPlugin
+liger_rope: true
+liger_rms_norm: true
+liger_glu_activation: true
+liger_layer_norm: true
+liger_fused_linear_cross_entropy: true
    +
    +
    +

    Supported Models

    +
      +
    • deepseek_v2
      • +
    • gemma
      • +
    • gemma2
      • +
    • gemma3
      • +
    • granite
      • +
    • jamba
      • +
    • llama
      • +
    • mistral
      • +
    • mixtral
      • +
    • mllama
      • +
    • mllama_text_model
      • +
    • olmo2
      • +
    • paligemma
      • +
    • phi3
      • +
    • qwen2
      • +
    • qwen2_5_vl
      • +
    • qwen2_vl
      • +
    +
    +
    +

    Citation

    +
    @article{hsu2024ligerkernelefficienttriton,
+      title={Liger Kernel: Efficient Triton Kernels for LLM Training},
+      author={Pin-Lun Hsu and Yun Dai and Vignesh Kothapalli and Qingquan Song and Shao Tang and Siyu Zhu and Steven Shimizu and Shivam Sahni and Haowen Ning and Yanning Chen},
+      year={2024},
+      eprint={2410.10989},
+      archivePrefix={arXiv},
+      primaryClass={cs.LG},
+      url={https://arxiv.org/abs/2410.10989},
+      journal={arXiv preprint arXiv:2410.10989},
+}
    +

    Please see reference here

    +
    +
    +
    +

    Spectrum

    +

    by Eric Hartford, Lucas Atkins, Fernando Fernandes, David Golchinfar

    +

    This plugin contains code to freeze the bottom fraction of modules in a model, based on the Signal-to-Noise Ratio (SNR).

    +

    See https://github.com/cognitivecomputations/spectrum

    +
    +

    Overview

    +

    Spectrum is a tool for scanning and evaluating the Signal-to-Noise Ratio (SNR) of layers in large language models. +By identifying the top n% of layers with the highest SNR, you can optimize training efficiency.

    +
    +
    +

    Usage

    +
    plugins:
+  - axolotl.integrations.spectrum.SpectrumPlugin
+
+spectrum_top_fraction: 0.5
+spectrum_model_name: meta-llama/Meta-Llama-3.1-8B
    +
    +
    +

    Citation

    +
    @misc{hartford2024spectrumtargetedtrainingsignal,
+      title={Spectrum: Targeted Training on Signal to Noise Ratio},
+      author={Eric Hartford and Lucas Atkins and Fernando Fernandes Neto and David Golchinfar},
+      year={2024},
+      eprint={2406.06623},
+      archivePrefix={arXiv},
+      primaryClass={cs.LG},
+      url={https://arxiv.org/abs/2406.06623},
+}
    +

    Please see reference here

    +
    +

    Adding a new integration

    Plugins can be used to customize the behavior of the training pipeline through hooks. See axolotl.integrations.BasePlugin for the possible hooks.

    @@ -903,10 +912,10 @@ Warning

    If you could not load your integration, please ensure you are pip installing in editable mode.

    -
    pip install -e .
    +
    pip install -e .

    and correctly spelled the integration name in the config file.

    -
    plugins:
-  - axolotl.integrations.your_integration_name.YourIntegrationPlugin
    +
    plugins:
+  - axolotl.integrations.your_integration_name.YourIntegrationPlugin
    diff --git a/search.json b/search.json index 67fbd84a4..4335f4077 100644 --- a/search.json +++ b/search.json @@ -3070,6 +3070,17 @@ "Custom Integrations" ] }, + { + "objectID": "docs/custom_integrations.html#densemixer", + "href": "docs/custom_integrations.html#densemixer", + "title": "Custom Integrations", + "section": "DenseMixer", + "text": "DenseMixer\nSee DenseMixer\nSimply add the following to your axolotl YAML config:\nplugins:\n - axolotl.integrations.densemixer.DenseMixerPlugin\nPlease see reference here", + "crumbs": [ + "Advanced Features", + "Custom Integrations" + ] + }, { "objectID": "docs/custom_integrations.html#grokfast", "href": "docs/custom_integrations.html#grokfast", @@ -3093,11 +3104,11 @@ ] }, { - "objectID": "docs/custom_integrations.html#liger-kernels", - "href": "docs/custom_integrations.html#liger-kernels", + "objectID": "docs/custom_integrations.html#llmcompressor", + "href": "docs/custom_integrations.html#llmcompressor", "title": "Custom Integrations", - "section": "Liger Kernels", - "text": "Liger Kernels\nLiger Kernel provides efficient Triton kernels for LLM training, offering:\n\n20% increase in multi-GPU training throughput\n60% reduction in memory usage\nCompatibility with both FSDP and DeepSpeed\n\nSee https://github.com/linkedin/Liger-Kernel\n\nUsage\nplugins:\n - axolotl.integrations.liger.LigerPlugin\nliger_rope: true\nliger_rms_norm: true\nliger_glu_activation: true\nliger_layer_norm: true\nliger_fused_linear_cross_entropy: true\n\n\nSupported Models\n\ndeepseek_v2\ngemma\ngemma2\ngemma3\ngranite\njamba\nllama\nmistral\nmixtral\nmllama\nmllama_text_model\nolmo2\npaligemma\nphi3\nqwen2\nqwen2_5_vl\nqwen2_vl\n\n\n\nCitation\n@article{hsu2024ligerkernelefficienttriton,\n title={Liger Kernel: Efficient Triton Kernels for LLM Training},\n author={Pin-Lun Hsu and Yun Dai and Vignesh Kothapalli and Qingquan Song and Shao Tang and Siyu Zhu and Steven Shimizu and Shivam Sahni and Haowen Ning and Yanning Chen},\n year={2024},\n eprint={2410.10989},\n archivePrefix={arXiv},\n primaryClass={cs.LG},\n url={https://arxiv.org/abs/2410.10989},\n journal={arXiv preprint arXiv:2410.10989},\n}\nPlease see reference here", + "section": "LLMCompressor", + "text": "LLMCompressor\nFine-tune sparsified models in Axolotl using Neural Magic’s LLMCompressor.\nThis integration enables fine-tuning of models sparsified using LLMCompressor within the Axolotl training framework. By combining LLMCompressor’s model compression capabilities with Axolotl’s distributed training pipelines, users can efficiently fine-tune sparse models at scale.\nIt uses Axolotl’s plugin system to hook into the fine-tuning flows while maintaining sparsity throughout training.\n\n\nRequirements\n\nAxolotl with llmcompressor extras:\npip install \"axolotl[llmcompressor]\"\nRequires llmcompressor >= 0.5.1\n\nThis will install all necessary dependencies to fine-tune sparsified models using the integration.\n\n\n\nUsage\nTo enable sparse fine-tuning with this integration, include the plugin in your Axolotl config:\nplugins:\n - axolotl.integrations.llm_compressor.LLMCompressorPlugin\n\nllmcompressor:\n recipe:\n finetuning_stage:\n finetuning_modifiers:\n ConstantPruningModifier:\n targets: [\n 're:.*q_proj.weight',\n 're:.*k_proj.weight',\n 're:.*v_proj.weight',\n 're:.*o_proj.weight',\n 're:.*gate_proj.weight',\n 're:.*up_proj.weight',\n 're:.*down_proj.weight',\n ]\n start: 0\n save_compressed: true\nThis plugin does not apply pruning or sparsification itself — it is intended for fine-tuning models that have already been sparsified.\nPre-sparsified checkpoints can be:\n- Generated using LLMCompressor\n- Downloaded from Neural Magic’s Hugging Face page\n- Any custom LLM with compatible sparsity patterns that you’ve created yourself\nTo learn more about writing and customizing LLMCompressor recipes, refer to the official documentation:\nhttps://github.com/vllm-project/llm-compressor/blob/main/README.md\n\n\nStorage Optimization with save_compressed\nSetting save_compressed: true in your configuration enables saving models in a compressed format, which:\n- Reduces disk space usage by approximately 40%\n- Maintains compatibility with vLLM for accelerated inference\n- Maintains compatibility with llmcompressor for further optimization (example: quantization)\nThis option is highly recommended when working with sparse models to maximize the benefits of model compression.\n\n\nExample Config\nSee examples/llama-3/sparse-finetuning.yaml for a complete example.\n\n\n\nInference with vLLM\nAfter fine-tuning your sparse model, you can leverage vLLM for efficient inference.\nYou can also use LLMCompressor to apply additional quantization to your fine-tuned\nsparse model before inference for even greater performance benefits.:\nfrom vllm import LLM, SamplingParams\n\nprompts = [\n \"Hello, my name is\",\n \"The president of the United States is\",\n \"The capital of France is\",\n \"The future of AI is\",\n]\nsampling_params = SamplingParams(temperature=0.8, top_p=0.95)\nllm = LLM(\"path/to/your/sparse/model\")\noutputs = llm.generate(prompts, sampling_params)\n\nfor output in outputs:\n prompt = output.prompt\n generated_text = output.outputs[0].text\n print(f\"Prompt: {prompt!r}, Generated text: {generated_text!r}\")\nFor more details on vLLM’s capabilities and advanced configuration options, see the official vLLM documentation.\n\n\nLearn More\nFor details on available sparsity and quantization schemes, fine-tuning recipes, and usage examples, visit the official LLMCompressor repository:\nhttps://github.com/vllm-project/llm-compressor\nPlease see reference here", "crumbs": [ "Advanced Features", "Custom Integrations" @@ -3115,22 +3126,22 @@ ] }, { - "objectID": "docs/custom_integrations.html#spectrum", - "href": "docs/custom_integrations.html#spectrum", + "objectID": "docs/custom_integrations.html#liger-kernels", + "href": "docs/custom_integrations.html#liger-kernels", "title": "Custom Integrations", - "section": "Spectrum", - "text": "Spectrum\nby Eric Hartford, Lucas Atkins, Fernando Fernandes, David Golchinfar\nThis plugin contains code to freeze the bottom fraction of modules in a model, based on the Signal-to-Noise Ratio (SNR).\nSee https://github.com/cognitivecomputations/spectrum\n\nOverview\nSpectrum is a tool for scanning and evaluating the Signal-to-Noise Ratio (SNR) of layers in large language models.\nBy identifying the top n% of layers with the highest SNR, you can optimize training efficiency.\n\n\nUsage\nplugins:\n - axolotl.integrations.spectrum.SpectrumPlugin\n\nspectrum_top_fraction: 0.5\nspectrum_model_name: meta-llama/Meta-Llama-3.1-8B\n\n\nCitation\n@misc{hartford2024spectrumtargetedtrainingsignal,\n title={Spectrum: Targeted Training on Signal to Noise Ratio},\n author={Eric Hartford and Lucas Atkins and Fernando Fernandes Neto and David Golchinfar},\n year={2024},\n eprint={2406.06623},\n archivePrefix={arXiv},\n primaryClass={cs.LG},\n url={https://arxiv.org/abs/2406.06623},\n}\nPlease see reference here", + "section": "Liger Kernels", + "text": "Liger Kernels\nLiger Kernel provides efficient Triton kernels for LLM training, offering:\n\n20% increase in multi-GPU training throughput\n60% reduction in memory usage\nCompatibility with both FSDP and DeepSpeed\n\nSee https://github.com/linkedin/Liger-Kernel\n\nUsage\nplugins:\n - axolotl.integrations.liger.LigerPlugin\nliger_rope: true\nliger_rms_norm: true\nliger_glu_activation: true\nliger_layer_norm: true\nliger_fused_linear_cross_entropy: true\n\n\nSupported Models\n\ndeepseek_v2\ngemma\ngemma2\ngemma3\ngranite\njamba\nllama\nmistral\nmixtral\nmllama\nmllama_text_model\nolmo2\npaligemma\nphi3\nqwen2\nqwen2_5_vl\nqwen2_vl\n\n\n\nCitation\n@article{hsu2024ligerkernelefficienttriton,\n title={Liger Kernel: Efficient Triton Kernels for LLM Training},\n author={Pin-Lun Hsu and Yun Dai and Vignesh Kothapalli and Qingquan Song and Shao Tang and Siyu Zhu and Steven Shimizu and Shivam Sahni and Haowen Ning and Yanning Chen},\n year={2024},\n eprint={2410.10989},\n archivePrefix={arXiv},\n primaryClass={cs.LG},\n url={https://arxiv.org/abs/2410.10989},\n journal={arXiv preprint arXiv:2410.10989},\n}\nPlease see reference here", "crumbs": [ "Advanced Features", "Custom Integrations" ] }, { - "objectID": "docs/custom_integrations.html#llmcompressor", - "href": "docs/custom_integrations.html#llmcompressor", + "objectID": "docs/custom_integrations.html#spectrum", + "href": "docs/custom_integrations.html#spectrum", "title": "Custom Integrations", - "section": "LLMCompressor", - "text": "LLMCompressor\nFine-tune sparsified models in Axolotl using Neural Magic’s LLMCompressor.\nThis integration enables fine-tuning of models sparsified using LLMCompressor within the Axolotl training framework. By combining LLMCompressor’s model compression capabilities with Axolotl’s distributed training pipelines, users can efficiently fine-tune sparse models at scale.\nIt uses Axolotl’s plugin system to hook into the fine-tuning flows while maintaining sparsity throughout training.\n\n\nRequirements\n\nAxolotl with llmcompressor extras:\npip install \"axolotl[llmcompressor]\"\nRequires llmcompressor >= 0.5.1\n\nThis will install all necessary dependencies to fine-tune sparsified models using the integration.\n\n\n\nUsage\nTo enable sparse fine-tuning with this integration, include the plugin in your Axolotl config:\nplugins:\n - axolotl.integrations.llm_compressor.LLMCompressorPlugin\n\nllmcompressor:\n recipe:\n finetuning_stage:\n finetuning_modifiers:\n ConstantPruningModifier:\n targets: [\n 're:.*q_proj.weight',\n 're:.*k_proj.weight',\n 're:.*v_proj.weight',\n 're:.*o_proj.weight',\n 're:.*gate_proj.weight',\n 're:.*up_proj.weight',\n 're:.*down_proj.weight',\n ]\n start: 0\n save_compressed: true\nThis plugin does not apply pruning or sparsification itself — it is intended for fine-tuning models that have already been sparsified.\nPre-sparsified checkpoints can be:\n- Generated using LLMCompressor\n- Downloaded from Neural Magic’s Hugging Face page\n- Any custom LLM with compatible sparsity patterns that you’ve created yourself\nTo learn more about writing and customizing LLMCompressor recipes, refer to the official documentation:\nhttps://github.com/vllm-project/llm-compressor/blob/main/README.md\n\n\nStorage Optimization with save_compressed\nSetting save_compressed: true in your configuration enables saving models in a compressed format, which:\n- Reduces disk space usage by approximately 40%\n- Maintains compatibility with vLLM for accelerated inference\n- Maintains compatibility with llmcompressor for further optimization (example: quantization)\nThis option is highly recommended when working with sparse models to maximize the benefits of model compression.\n\n\nExample Config\nSee examples/llama-3/sparse-finetuning.yaml for a complete example.\n\n\n\nInference with vLLM\nAfter fine-tuning your sparse model, you can leverage vLLM for efficient inference.\nYou can also use LLMCompressor to apply additional quantization to your fine-tuned\nsparse model before inference for even greater performance benefits.:\nfrom vllm import LLM, SamplingParams\n\nprompts = [\n \"Hello, my name is\",\n \"The president of the United States is\",\n \"The capital of France is\",\n \"The future of AI is\",\n]\nsampling_params = SamplingParams(temperature=0.8, top_p=0.95)\nllm = LLM(\"path/to/your/sparse/model\")\noutputs = llm.generate(prompts, sampling_params)\n\nfor output in outputs:\n prompt = output.prompt\n generated_text = output.outputs[0].text\n print(f\"Prompt: {prompt!r}, Generated text: {generated_text!r}\")\nFor more details on vLLM’s capabilities and advanced configuration options, see the official vLLM documentation.\n\n\nLearn More\nFor details on available sparsity and quantization schemes, fine-tuning recipes, and usage examples, visit the official LLMCompressor repository:\nhttps://github.com/vllm-project/llm-compressor\nPlease see reference here", + "section": "Spectrum", + "text": "Spectrum\nby Eric Hartford, Lucas Atkins, Fernando Fernandes, David Golchinfar\nThis plugin contains code to freeze the bottom fraction of modules in a model, based on the Signal-to-Noise Ratio (SNR).\nSee https://github.com/cognitivecomputations/spectrum\n\nOverview\nSpectrum is a tool for scanning and evaluating the Signal-to-Noise Ratio (SNR) of layers in large language models.\nBy identifying the top n% of layers with the highest SNR, you can optimize training efficiency.\n\n\nUsage\nplugins:\n - axolotl.integrations.spectrum.SpectrumPlugin\n\nspectrum_top_fraction: 0.5\nspectrum_model_name: meta-llama/Meta-Llama-3.1-8B\n\n\nCitation\n@misc{hartford2024spectrumtargetedtrainingsignal,\n title={Spectrum: Targeted Training on Signal to Noise Ratio},\n author={Eric Hartford and Lucas Atkins and Fernando Fernandes Neto and David Golchinfar},\n year={2024},\n eprint={2406.06623},\n archivePrefix={arXiv},\n primaryClass={cs.LG},\n url={https://arxiv.org/abs/2406.06623},\n}\nPlease see reference here", "crumbs": [ "Advanced Features", "Custom Integrations" diff --git a/sitemap.xml b/sitemap.xml index ac719c6f1..e3a39aca3 100644 --- a/sitemap.xml +++ b/sitemap.xml @@ -2,758 +2,758 @@ https://docs.axolotl.ai/docs/unsloth.html - 2025-07-07T19:24:01.499Z + 2025-07-07T21:05:27.508Z https://docs.axolotl.ai/docs/dataset-formats/conversation.html - 2025-07-07T19:24:01.495Z + 2025-07-07T21:05:27.502Z https://docs.axolotl.ai/docs/dataset-formats/stepwise_supervised.html - 2025-07-07T19:24:01.495Z + 2025-07-07T21:05:27.502Z https://docs.axolotl.ai/docs/dataset-formats/tokenized.html - 2025-07-07T19:24:01.495Z + 2025-07-07T21:05:27.503Z https://docs.axolotl.ai/docs/mac.html - 2025-07-07T19:24:01.498Z + 2025-07-07T21:05:27.507Z https://docs.axolotl.ai/docs/nccl.html - 2025-07-07T19:24:01.499Z + 2025-07-07T21:05:27.507Z https://docs.axolotl.ai/docs/multi-node.html - 2025-07-07T19:24:01.499Z + 2025-07-07T21:05:27.507Z https://docs.axolotl.ai/docs/docker.html - 2025-07-07T19:24:01.495Z + 2025-07-07T21:05:27.503Z https://docs.axolotl.ai/docs/lr_groups.html - 2025-07-07T19:24:01.498Z + 2025-07-07T21:05:27.507Z https://docs.axolotl.ai/docs/inference.html - 2025-07-07T19:24:01.498Z + 2025-07-07T21:05:27.507Z https://docs.axolotl.ai/docs/cli.html - 2025-07-07T19:24:01.495Z + 2025-07-07T21:05:27.502Z https://docs.axolotl.ai/docs/config-reference.html - 2025-07-07T19:27:44.889Z + 2025-07-07T21:08:49.663Z https://docs.axolotl.ai/docs/multi-gpu.html - 2025-07-07T19:24:01.498Z + 2025-07-07T21:05:27.507Z https://docs.axolotl.ai/docs/debugging.html - 2025-07-07T19:24:01.495Z + 2025-07-07T21:05:27.503Z https://docs.axolotl.ai/docs/multimodal.html - 2025-07-07T19:24:01.499Z + 2025-07-07T21:05:27.507Z https://docs.axolotl.ai/docs/api/cli.sweeps.html - 2025-07-07T19:27:30.521Z + 2025-07-07T21:08:36.131Z https://docs.axolotl.ai/docs/api/prompt_strategies.dpo.llama3.html - 2025-07-07T19:27:30.850Z + 2025-07-07T21:08:36.457Z https://docs.axolotl.ai/docs/api/utils.schedulers.html - 2025-07-07T19:27:31.238Z + 2025-07-07T21:08:36.840Z https://docs.axolotl.ai/docs/api/monkeypatch.llama_attn_hijack_xformers.html - 2025-07-07T19:27:31.052Z + 2025-07-07T21:08:36.655Z https://docs.axolotl.ai/docs/api/cli.cloud.modal_.html - 2025-07-07T19:27:30.570Z + 2025-07-07T21:08:36.180Z https://docs.axolotl.ai/docs/api/kernels.geglu.html - 2025-07-07T19:27:31.004Z + 2025-07-07T21:08:36.609Z https://docs.axolotl.ai/docs/api/core.trainers.utils.html - 2025-07-07T19:27:30.643Z + 2025-07-07T21:08:36.252Z https://docs.axolotl.ai/docs/api/core.datasets.chat.html - 2025-07-07T19:27:30.369Z + 2025-07-07T21:08:35.983Z https://docs.axolotl.ai/docs/api/utils.schemas.peft.html - 2025-07-07T19:27:31.353Z + 2025-07-07T21:08:36.954Z https://docs.axolotl.ai/docs/api/monkeypatch.btlm_attn_hijack_flash.html - 2025-07-07T19:27:31.115Z + 2025-07-07T21:08:36.718Z https://docs.axolotl.ai/docs/api/prompt_strategies.chat_template.html - 2025-07-07T19:27:30.752Z + 2025-07-07T21:08:36.360Z https://docs.axolotl.ai/docs/api/prompt_strategies.kto.user_defined.html - 2025-07-07T19:27:30.883Z + 2025-07-07T21:08:36.490Z https://docs.axolotl.ai/docs/api/cli.cloud.base.html - 2025-07-07T19:27:30.563Z + 2025-07-07T21:08:36.173Z https://docs.axolotl.ai/docs/api/kernels.swiglu.html - 2025-07-07T19:27:31.014Z + 2025-07-07T21:08:36.619Z https://docs.axolotl.ai/docs/api/prompt_strategies.stepwise_supervised.html - 2025-07-07T19:27:30.817Z + 2025-07-07T21:08:36.424Z https://docs.axolotl.ai/docs/api/prompt_strategies.bradley_terry.llama3.html - 2025-07-07T19:27:30.908Z + 2025-07-07T21:08:36.514Z https://docs.axolotl.ai/docs/api/prompt_strategies.completion.html - 2025-07-07T19:27:30.806Z + 2025-07-07T21:08:36.414Z https://docs.axolotl.ai/docs/api/kernels.utils.html - 2025-07-07T19:27:31.024Z + 2025-07-07T21:08:36.628Z https://docs.axolotl.ai/docs/api/common.datasets.html - 2025-07-07T19:27:31.568Z + 2025-07-07T21:08:37.165Z https://docs.axolotl.ai/docs/api/utils.schemas.datasets.html - 2025-07-07T19:27:31.344Z + 2025-07-07T21:08:36.945Z https://docs.axolotl.ai/docs/api/core.builders.rl.html - 2025-07-07T19:27:30.324Z + 2025-07-07T21:08:35.938Z https://docs.axolotl.ai/docs/api/evaluate.html - 2025-07-07T19:27:30.228Z + 2025-07-07T21:08:35.843Z https://docs.axolotl.ai/docs/api/kernels.quantize.html - 2025-07-07T19:27:31.022Z + 2025-07-07T21:08:36.626Z https://docs.axolotl.ai/docs/api/monkeypatch.llama_attn_hijack_flash.html - 2025-07-07T19:27:31.050Z + 2025-07-07T21:08:36.654Z https://docs.axolotl.ai/docs/api/core.trainers.mixins.rng_state_loader.html - 2025-07-07T19:27:30.687Z + 2025-07-07T21:08:36.295Z https://docs.axolotl.ai/docs/api/integrations.base.html - 2025-07-07T19:27:31.528Z + 2025-07-07T21:08:37.125Z https://docs.axolotl.ai/docs/api/cli.merge_lora.html - 2025-07-07T19:27:30.494Z + 2025-07-07T21:08:36.105Z https://docs.axolotl.ai/docs/api/cli.merge_sharded_fsdp_weights.html - 2025-07-07T19:27:30.506Z + 2025-07-07T21:08:36.117Z https://docs.axolotl.ai/docs/api/monkeypatch.transformers_fa_utils.html - 2025-07-07T19:27:31.132Z + 2025-07-07T21:08:36.734Z https://docs.axolotl.ai/docs/api/prompt_strategies.llama2_chat.html - 2025-07-07T19:27:30.800Z + 2025-07-07T21:08:36.408Z https://docs.axolotl.ai/docs/api/utils.collators.mm_chat.html - 2025-07-07T19:27:31.598Z + 2025-07-07T21:08:37.195Z https://docs.axolotl.ai/docs/api/utils.data.sft.html - 2025-07-07T19:27:31.279Z + 2025-07-07T21:08:36.880Z https://docs.axolotl.ai/docs/api/prompt_strategies.alpaca_instruct.html - 2025-07-07T19:27:30.768Z + 2025-07-07T21:08:36.375Z https://docs.axolotl.ai/docs/api/integrations.liger.args.html - 2025-07-07T19:27:31.543Z + 2025-07-07T21:08:37.141Z https://docs.axolotl.ai/docs/api/monkeypatch.mistral_attn_hijack_flash.html - 2025-07-07T19:27:31.067Z + 2025-07-07T21:08:36.669Z https://docs.axolotl.ai/docs/api/cli.vllm_serve.html - 2025-07-07T19:27:30.560Z + 2025-07-07T21:08:36.170Z https://docs.axolotl.ai/docs/api/monkeypatch.utils.html - 2025-07-07T19:27:31.113Z + 2025-07-07T21:08:36.716Z https://docs.axolotl.ai/docs/api/loaders.patch_manager.html - 2025-07-07T19:27:30.677Z + 2025-07-07T21:08:36.285Z https://docs.axolotl.ai/docs/api/utils.schemas.integrations.html - 2025-07-07T19:27:31.374Z + 2025-07-07T21:08:36.975Z https://docs.axolotl.ai/docs/api/utils.callbacks.perplexity.html - 2025-07-07T19:27:31.646Z + 2025-07-07T21:08:37.242Z https://docs.axolotl.ai/docs/api/cli.utils.html - 2025-07-07T19:27:30.553Z + 2025-07-07T21:08:36.163Z https://docs.axolotl.ai/docs/api/utils.schemas.config.html - 2025-07-07T19:27:31.314Z + 2025-07-07T21:08:36.915Z https://docs.axolotl.ai/docs/api/prompt_strategies.input_output.html - 2025-07-07T19:27:30.812Z + 2025-07-07T21:08:36.420Z https://docs.axolotl.ai/docs/api/utils.distributed.html - 2025-07-07T19:27:31.258Z + 2025-07-07T21:08:36.861Z https://docs.axolotl.ai/docs/api/monkeypatch.gradient_checkpointing.offload_disk.html - 2025-07-07T19:27:31.165Z + 2025-07-07T21:08:36.768Z https://docs.axolotl.ai/docs/api/monkeypatch.trainer_fsdp_optim.html - 2025-07-07T19:27:31.125Z + 2025-07-07T21:08:36.728Z https://docs.axolotl.ai/docs/api/core.builders.base.html - 2025-07-07T19:27:30.311Z + 2025-07-07T21:08:35.925Z https://docs.axolotl.ai/docs/api/core.trainers.trl.html - 2025-07-07T19:27:30.602Z + 2025-07-07T21:08:36.211Z https://docs.axolotl.ai/docs/api/cli.evaluate.html - 2025-07-07T19:27:30.425Z + 2025-07-07T21:08:36.038Z https://docs.axolotl.ai/docs/api/utils.optimizers.adopt.html - 2025-07-07T19:27:31.270Z + 2025-07-07T21:08:36.872Z https://docs.axolotl.ai/docs/api/utils.callbacks.qat.html - 2025-07-07T19:27:31.665Z + 2025-07-07T21:08:37.261Z https://docs.axolotl.ai/docs/api/core.trainers.dpo.trainer.html - 2025-07-07T19:27:30.618Z + 2025-07-07T21:08:36.228Z https://docs.axolotl.ai/docs/api/core.chat.format.shared.html - 2025-07-07T19:27:30.364Z + 2025-07-07T21:08:35.978Z https://docs.axolotl.ai/docs/api/monkeypatch.relora.html - 2025-07-07T19:27:31.075Z + 2025-07-07T21:08:36.678Z https://docs.axolotl.ai/docs/api/cli.config.html - 2025-07-07T19:27:30.471Z + 2025-07-07T21:08:36.082Z https://docs.axolotl.ai/docs/api/cli.preprocess.html - 2025-07-07T19:27:30.515Z + 2025-07-07T21:08:36.125Z https://docs.axolotl.ai/docs/api/core.trainers.base.html - 2025-07-07T19:27:30.585Z + 2025-07-07T21:08:36.195Z https://docs.axolotl.ai/docs/api/convert.html - 2025-07-07T19:27:30.253Z + 2025-07-07T21:08:35.867Z https://docs.axolotl.ai/docs/api/prompt_strategies.pygmalion.html - 2025-07-07T19:27:30.834Z + 2025-07-07T21:08:36.441Z https://docs.axolotl.ai/docs/api/utils.schemas.trl.html - 2025-07-07T19:27:31.357Z + 2025-07-07T21:08:36.957Z https://docs.axolotl.ai/docs/api/cli.args.html - 2025-07-07T19:27:30.445Z + 2025-07-07T21:08:36.058Z https://docs.axolotl.ai/docs/api/prompt_strategies.dpo.chat_template.html - 2025-07-07T19:27:30.840Z + 2025-07-07T21:08:36.447Z https://docs.axolotl.ai/docs/api/loaders.constants.html - 2025-07-07T19:27:30.678Z + 2025-07-07T21:08:36.287Z https://docs.axolotl.ai/docs/api/logging_config.html - 2025-07-07T19:27:30.305Z + 2025-07-07T21:08:35.919Z https://docs.axolotl.ai/docs/api/cli.inference.html - 2025-07-07T19:27:30.485Z + 2025-07-07T21:08:36.096Z https://docs.axolotl.ai/docs/api/utils.ctx_managers.sequence_parallel.html - 2025-07-07T19:27:30.718Z + 2025-07-07T21:08:36.326Z https://docs.axolotl.ai/docs/api/integrations.spectrum.args.html - 2025-07-07T19:27:31.550Z + 2025-07-07T21:08:37.147Z https://docs.axolotl.ai/docs/api/utils.schemas.training.html - 2025-07-07T19:27:31.326Z + 2025-07-07T21:08:36.927Z https://docs.axolotl.ai/docs/api/prompt_strategies.orcamini.html - 2025-07-07T19:27:30.828Z + 2025-07-07T21:08:36.435Z https://docs.axolotl.ai/docs/api/utils.freeze.html - 2025-07-07T19:27:31.196Z + 2025-07-07T21:08:36.798Z https://docs.axolotl.ai/docs/api/loaders.tokenizer.html - 2025-07-07T19:27:30.662Z + 2025-07-07T21:08:36.270Z https://docs.axolotl.ai/docs/api/utils.bench.html - 2025-07-07T19:27:31.188Z + 2025-07-07T21:08:36.790Z https://docs.axolotl.ai/docs/api/utils.quantization.html - 2025-07-07T19:27:31.300Z + 2025-07-07T21:08:36.901Z https://docs.axolotl.ai/docs/batch_vs_grad.html - 2025-07-07T19:24:01.495Z + 2025-07-07T21:05:27.502Z https://docs.axolotl.ai/docs/input_output.html - 2025-07-07T19:24:01.498Z + 2025-07-07T21:05:27.507Z https://docs.axolotl.ai/docs/sequence_parallelism.html - 2025-07-07T19:24:01.499Z + 2025-07-07T21:05:27.508Z https://docs.axolotl.ai/docs/reward_modelling.html - 2025-07-07T19:24:01.499Z + 2025-07-07T21:05:27.508Z https://docs.axolotl.ai/index.html - 2025-07-07T19:24:01.514Z + 2025-07-07T21:05:27.525Z https://docs.axolotl.ai/src/axolotl/integrations/LICENSE.html - 2025-07-07T19:24:01.518Z + 2025-07-07T21:05:27.529Z https://docs.axolotl.ai/FAQS.html - 2025-07-07T19:24:01.493Z + 2025-07-07T21:05:27.500Z https://docs.axolotl.ai/src/axolotl/integrations/cut_cross_entropy/ACKNOWLEDGEMENTS.html - 2025-07-07T19:24:01.518Z + 2025-07-07T21:05:27.529Z https://docs.axolotl.ai/TODO.html - 2025-07-07T19:24:01.493Z + 2025-07-07T21:05:27.500Z https://docs.axolotl.ai/examples/colab-notebooks/colab-axolotl-example.html - 2025-07-07T19:24:01.501Z + 2025-07-07T21:05:27.510Z https://docs.axolotl.ai/docs/torchao.html - 2025-07-07T19:24:01.499Z + 2025-07-07T21:05:27.508Z https://docs.axolotl.ai/docs/ray-integration.html - 2025-07-07T19:24:01.499Z + 2025-07-07T21:05:27.508Z https://docs.axolotl.ai/docs/quantize.html - 2025-07-07T19:24:01.499Z + 2025-07-07T21:05:27.507Z https://docs.axolotl.ai/docs/qat.html - 2025-07-07T19:24:01.499Z + 2025-07-07T21:05:27.507Z https://docs.axolotl.ai/docs/api/utils.lora.html - 2025-07-07T19:27:31.179Z + 2025-07-07T21:08:36.781Z https://docs.axolotl.ai/docs/api/prompt_strategies.alpaca_w_system.html - 2025-07-07T19:27:30.780Z + 2025-07-07T21:08:36.387Z https://docs.axolotl.ai/docs/api/monkeypatch.stablelm_attn_hijack_flash.html - 2025-07-07T19:27:31.122Z + 2025-07-07T21:08:36.725Z https://docs.axolotl.ai/docs/api/utils.collators.core.html - 2025-07-07T19:27:31.570Z + 2025-07-07T21:08:37.168Z https://docs.axolotl.ai/docs/api/prompt_strategies.metharme.html - 2025-07-07T19:27:30.824Z + 2025-07-07T21:08:36.431Z https://docs.axolotl.ai/docs/api/utils.callbacks.profiler.html - 2025-07-07T19:27:31.650Z + 2025-07-07T21:08:37.245Z https://docs.axolotl.ai/docs/api/utils.data.pretraining.html - 2025-07-07T19:27:31.271Z + 2025-07-07T21:08:36.873Z https://docs.axolotl.ai/docs/api/utils.callbacks.lisa.html - 2025-07-07T19:27:31.651Z + 2025-07-07T21:08:37.247Z https://docs.axolotl.ai/docs/api/utils.trainer.html - 2025-07-07T19:27:31.213Z + 2025-07-07T21:08:36.816Z https://docs.axolotl.ai/docs/api/integrations.cut_cross_entropy.args.html - 2025-07-07T19:27:31.531Z + 2025-07-07T21:08:37.129Z https://docs.axolotl.ai/docs/api/utils.schemas.model.html - 2025-07-07T19:27:31.321Z + 2025-07-07T21:08:36.922Z https://docs.axolotl.ai/docs/api/monkeypatch.data.batch_dataset_fetcher.html - 2025-07-07T19:27:31.135Z + 2025-07-07T21:08:36.737Z https://docs.axolotl.ai/docs/api/prompt_strategies.dpo.zephyr.html - 2025-07-07T19:27:30.862Z + 2025-07-07T21:08:36.469Z https://docs.axolotl.ai/docs/api/datasets.html - 2025-07-07T19:27:30.239Z + 2025-07-07T21:08:35.854Z https://docs.axolotl.ai/docs/api/utils.schemas.enums.html - 2025-07-07T19:27:31.385Z + 2025-07-07T21:08:36.985Z https://docs.axolotl.ai/docs/api/integrations.kd.trainer.html - 2025-07-07T19:27:31.540Z + 2025-07-07T21:08:37.137Z https://docs.axolotl.ai/docs/api/monkeypatch.lora_kernels.html - 2025-07-07T19:27:31.105Z + 2025-07-07T21:08:36.708Z https://docs.axolotl.ai/docs/api/utils.collators.batching.html - 2025-07-07T19:27:31.590Z + 2025-07-07T21:08:37.187Z https://docs.axolotl.ai/docs/api/core.trainers.grpo.sampler.html - 2025-07-07T19:27:30.642Z + 2025-07-07T21:08:36.251Z https://docs.axolotl.ai/docs/api/prompt_strategies.base.html - 2025-07-07T19:27:30.720Z + 2025-07-07T21:08:36.327Z https://docs.axolotl.ai/docs/api/monkeypatch.multipack.html - 2025-07-07T19:27:31.068Z + 2025-07-07T21:08:36.671Z https://docs.axolotl.ai/docs/api/prompt_strategies.orpo.chat_template.html - 2025-07-07T19:27:30.904Z + 2025-07-07T21:08:36.510Z https://docs.axolotl.ai/docs/api/prompt_strategies.dpo.passthrough.html - 2025-07-07T19:27:30.865Z + 2025-07-07T21:08:36.472Z https://docs.axolotl.ai/docs/api/core.chat.format.chatml.html - 2025-07-07T19:27:30.361Z + 2025-07-07T21:08:35.975Z https://docs.axolotl.ai/docs/api/core.trainers.mixins.scheduler.html - 2025-07-07T19:27:30.694Z + 2025-07-07T21:08:36.302Z https://docs.axolotl.ai/docs/api/utils.model_shard_quant.html - 2025-07-07T19:27:31.185Z + 2025-07-07T21:08:36.787Z https://docs.axolotl.ai/docs/api/prompt_strategies.kto.chatml.html - 2025-07-07T19:27:30.882Z + 2025-07-07T21:08:36.488Z https://docs.axolotl.ai/docs/api/utils.tokenization.html - 2025-07-07T19:27:31.172Z + 2025-07-07T21:08:36.775Z https://docs.axolotl.ai/docs/api/loaders.model.html - 2025-07-07T19:27:30.654Z + 2025-07-07T21:08:36.262Z https://docs.axolotl.ai/docs/api/utils.callbacks.mlflow_.html - 2025-07-07T19:27:31.655Z + 2025-07-07T21:08:37.250Z https://docs.axolotl.ai/docs/api/core.trainers.grpo.trainer.html - 2025-07-07T19:27:30.629Z + 2025-07-07T21:08:36.239Z https://docs.axolotl.ai/docs/api/cli.main.html - 2025-07-07T19:27:30.408Z + 2025-07-07T21:08:36.022Z https://docs.axolotl.ai/docs/api/utils.callbacks.comet_.html - 2025-07-07T19:27:31.658Z + 2025-07-07T21:08:37.254Z https://docs.axolotl.ai/docs/api/utils.chat_templates.html - 2025-07-07T19:27:31.174Z + 2025-07-07T21:08:36.776Z https://docs.axolotl.ai/docs/api/utils.schemas.utils.html - 2025-07-07T19:27:31.391Z + 2025-07-07T21:08:36.991Z https://docs.axolotl.ai/docs/api/common.architectures.html - 2025-07-07T19:27:31.551Z + 2025-07-07T21:08:37.149Z https://docs.axolotl.ai/docs/api/monkeypatch.llama_expand_mask.html - 2025-07-07T19:27:31.076Z + 2025-07-07T21:08:36.679Z https://docs.axolotl.ai/docs/api/prompt_strategies.alpaca_chat.html - 2025-07-07T19:27:30.766Z + 2025-07-07T21:08:36.374Z https://docs.axolotl.ai/docs/api/utils.samplers.multipack.html - 2025-07-07T19:27:31.640Z + 2025-07-07T21:08:37.236Z https://docs.axolotl.ai/docs/api/integrations.grokfast.optimizer.html - 2025-07-07T19:27:31.532Z + 2025-07-07T21:08:37.130Z https://docs.axolotl.ai/docs/api/prompt_strategies.dpo.chatml.html - 2025-07-07T19:27:30.861Z + 2025-07-07T21:08:36.467Z https://docs.axolotl.ai/docs/api/monkeypatch.mixtral.html - 2025-07-07T19:27:31.136Z + 2025-07-07T21:08:36.739Z https://docs.axolotl.ai/docs/api/train.html - 2025-07-07T19:27:30.218Z + 2025-07-07T21:08:35.833Z https://docs.axolotl.ai/docs/api/monkeypatch.llama_patch_multipack.html - 2025-07-07T19:27:31.116Z + 2025-07-07T21:08:36.719Z https://docs.axolotl.ai/docs/api/index.html - 2025-07-07T19:27:30.154Z + 2025-07-07T21:08:35.771Z https://docs.axolotl.ai/docs/api/loaders.adapter.html - 2025-07-07T19:27:30.669Z + 2025-07-07T21:08:36.277Z https://docs.axolotl.ai/docs/api/utils.schemas.multimodal.html - 2025-07-07T19:27:31.362Z + 2025-07-07T21:08:36.963Z https://docs.axolotl.ai/docs/api/kernels.lora.html - 2025-07-07T19:27:30.993Z + 2025-07-07T21:08:36.598Z https://docs.axolotl.ai/docs/api/prompt_strategies.kto.llama3.html - 2025-07-07T19:27:30.873Z + 2025-07-07T21:08:36.480Z https://docs.axolotl.ai/docs/api/cli.checks.html - 2025-07-07T19:27:30.452Z + 2025-07-07T21:08:36.065Z https://docs.axolotl.ai/docs/api/cli.quantize.html - 2025-07-07T19:27:30.575Z + 2025-07-07T21:08:36.185Z https://docs.axolotl.ai/docs/api/integrations.lm_eval.args.html - 2025-07-07T19:27:31.546Z + 2025-07-07T21:08:37.144Z https://docs.axolotl.ai/docs/api/core.chat.messages.html - 2025-07-07T19:27:30.360Z + 2025-07-07T21:08:35.974Z https://docs.axolotl.ai/docs/api/core.builders.causal.html - 2025-07-07T19:27:30.316Z + 2025-07-07T21:08:35.930Z https://docs.axolotl.ai/docs/api/core.trainers.relora.html - 2025-07-07T19:27:30.612Z + 2025-07-07T21:08:36.221Z https://docs.axolotl.ai/docs/api/models.mamba.modeling_mamba.html - 2025-07-07T19:27:31.569Z + 2025-07-07T21:08:37.166Z https://docs.axolotl.ai/docs/api/monkeypatch.gradient_checkpointing.offload_cpu.html - 2025-07-07T19:27:31.139Z + 2025-07-07T21:08:36.742Z https://docs.axolotl.ai/docs/api/core.trainers.mamba.html - 2025-07-07T19:27:30.607Z + 2025-07-07T21:08:36.217Z https://docs.axolotl.ai/docs/api/core.datasets.transforms.chat_builder.html - 2025-07-07T19:27:30.377Z + 2025-07-07T21:08:35.991Z https://docs.axolotl.ai/docs/api/loaders.processor.html - 2025-07-07T19:27:30.663Z + 2025-07-07T21:08:36.272Z https://docs.axolotl.ai/docs/api/core.chat.format.llama3x.html - 2025-07-07T19:27:30.363Z + 2025-07-07T21:08:35.976Z https://docs.axolotl.ai/docs/api/prompt_strategies.messages.chat.html - 2025-07-07T19:27:30.839Z + 2025-07-07T21:08:36.445Z https://docs.axolotl.ai/docs/api/cli.train.html - 2025-07-07T19:27:30.417Z + 2025-07-07T21:08:36.030Z https://docs.axolotl.ai/docs/api/core.trainers.mixins.optimizer.html - 2025-07-07T19:27:30.684Z + 2025-07-07T21:08:36.292Z https://docs.axolotl.ai/docs/api/utils.collators.mamba.html - 2025-07-07T19:27:31.593Z + 2025-07-07T21:08:37.190Z https://docs.axolotl.ai/docs/api/monkeypatch.unsloth_.html - 2025-07-07T19:27:31.133Z + 2025-07-07T21:08:36.736Z https://docs.axolotl.ai/docs/api/utils.dict.html - 2025-07-07T19:27:31.262Z + 2025-07-07T21:08:36.864Z https://docs.axolotl.ai/docs/api/prompt_strategies.user_defined.html - 2025-07-07T19:27:30.788Z + 2025-07-07T21:08:36.395Z https://docs.axolotl.ai/docs/api/core.training_args.html - 2025-07-07T19:27:30.337Z + 2025-07-07T21:08:35.951Z https://docs.axolotl.ai/docs/api/prompt_strategies.dpo.user_defined.html - 2025-07-07T19:27:30.864Z + 2025-07-07T21:08:36.471Z https://docs.axolotl.ai/docs/api/prompt_tokenizers.html - 2025-07-07T19:27:30.295Z + 2025-07-07T21:08:35.910Z https://docs.axolotl.ai/docs/api/common.const.html - 2025-07-07T19:27:31.553Z + 2025-07-07T21:08:37.150Z https://docs.axolotl.ai/docs/fsdp_qlora.html - 2025-07-07T19:24:01.496Z + 2025-07-07T21:05:27.503Z https://docs.axolotl.ai/docs/custom_integrations.html - 2025-07-07T19:24:01.495Z + 2025-07-07T21:05:27.502Z https://docs.axolotl.ai/docs/getting-started.html - 2025-07-07T19:24:01.496Z + 2025-07-07T21:05:27.503Z https://docs.axolotl.ai/docs/faq.html - 2025-07-07T19:24:01.495Z + 2025-07-07T21:05:27.503Z https://docs.axolotl.ai/docs/lora_optims.html - 2025-07-07T19:24:01.498Z + 2025-07-07T21:05:27.507Z https://docs.axolotl.ai/docs/rlhf.html - 2025-07-07T19:24:01.499Z + 2025-07-07T21:05:27.508Z https://docs.axolotl.ai/docs/amd_hpc.html - 2025-07-07T19:24:01.495Z + 2025-07-07T21:05:27.502Z https://docs.axolotl.ai/docs/installation.html - 2025-07-07T19:24:01.498Z + 2025-07-07T21:05:27.507Z https://docs.axolotl.ai/docs/multipack.html - 2025-07-07T19:24:01.499Z + 2025-07-07T21:05:27.507Z https://docs.axolotl.ai/docs/dataset_preprocessing.html - 2025-07-07T19:24:01.495Z + 2025-07-07T21:05:27.503Z https://docs.axolotl.ai/docs/dataset_loading.html - 2025-07-07T19:24:01.495Z + 2025-07-07T21:05:27.503Z https://docs.axolotl.ai/docs/dataset-formats/inst_tune.html - 2025-07-07T19:24:01.495Z + 2025-07-07T21:05:27.502Z https://docs.axolotl.ai/docs/dataset-formats/template_free.html - 2025-07-07T19:24:01.495Z + 2025-07-07T21:05:27.503Z https://docs.axolotl.ai/docs/dataset-formats/index.html - 2025-07-07T19:24:01.495Z + 2025-07-07T21:05:27.502Z https://docs.axolotl.ai/docs/dataset-formats/pretraining.html - 2025-07-07T19:24:01.495Z + 2025-07-07T21:05:27.502Z