From f18c2bb1f8a1d1563e00da0d2253f3c45aee60a2 Mon Sep 17 00:00:00 2001 From: Quarto GHA Workflow Runner Date: Tue, 21 Apr 2026 14:23:11 +0000 Subject: [PATCH] Built site for gh-pages --- .nojekyll | 2 +- FAQS.html | 6 - docs/agents/grpo.html | 6 - docs/agents/model_architectures.html | 6 - docs/agents/new_model_support.html | 6 - docs/agents/preference_tuning.html | 6 - docs/agents/pretraining.html | 6 - docs/agents/reward_modelling.html | 6 - docs/agents/sft.html | 6 - docs/amd_hpc.html | 6 - docs/api/cli.args.html | 6 - docs/api/cli.art.html | 6 - docs/api/cli.checks.html | 6 - docs/api/cli.cloud.base.html | 6 - docs/api/cli.cloud.modal_.html | 6 - docs/api/cli.config.html | 6 - docs/api/cli.delinearize_llama4.html | 6 - docs/api/cli.evaluate.html | 6 - docs/api/cli.inference.html | 6 - docs/api/cli.main.html | 6 - docs/api/cli.merge_lora.html | 6 - docs/api/cli.merge_sharded_fsdp_weights.html | 6 - docs/api/cli.preprocess.html | 6 - docs/api/cli.quantize.html | 6 - docs/api/cli.train.html | 6 - docs/api/cli.utils.args.html | 6 - docs/api/cli.utils.fetch.html | 6 - docs/api/cli.utils.html | 6 - docs/api/cli.utils.load.html | 6 - docs/api/cli.utils.sweeps.html | 6 - docs/api/cli.utils.train.html | 6 - docs/api/cli.vllm_serve.html | 6 - docs/api/common.architectures.html | 6 - docs/api/common.const.html | 6 - docs/api/common.datasets.html | 6 - docs/api/convert.html | 6 - docs/api/core.builders.base.html | 6 - docs/api/core.builders.causal.html | 6 - docs/api/core.builders.rl.html | 6 - docs/api/core.chat.format.chatml.html | 6 - docs/api/core.chat.format.llama3x.html | 6 - docs/api/core.chat.format.shared.html | 6 - docs/api/core.chat.messages.html | 6 - docs/api/core.datasets.chat.html | 6 - ...core.datasets.transforms.chat_builder.html | 6 - docs/api/core.trainers.base.html | 6 - docs/api/core.trainers.dpo.trainer.html | 6 - docs/api/core.trainers.grpo.sampler.html | 6 - docs/api/core.trainers.grpo.trainer.html | 6 - docs/api/core.trainers.mamba.html | 6 - docs/api/core.trainers.mixins.optimizer.html | 6 - ...core.trainers.mixins.rng_state_loader.html | 6 - docs/api/core.trainers.mixins.scheduler.html | 6 - docs/api/core.trainers.trl.html | 6 - docs/api/core.trainers.utils.html | 6 - docs/api/core.training_args.html | 6 - docs/api/datasets.html | 6 - docs/api/evaluate.html | 6 - docs/api/index.html | 16 +- docs/api/integrations.base.html | 6 - .../integrations.cut_cross_entropy.args.html | 6 - docs/api/integrations.grokfast.optimizer.html | 6 - docs/api/integrations.kd.trainer.html | 6 - docs/api/integrations.liger.args.html | 6 - docs/api/integrations.lm_eval.args.html | 6 - docs/api/integrations.spectrum.args.html | 6 - docs/api/kernels.geglu.html | 6 - docs/api/kernels.lora.html | 6 - docs/api/kernels.quantize.html | 6 - docs/api/kernels.swiglu.html | 6 - docs/api/kernels.utils.html | 6 - docs/api/loaders.adapter.html | 6 - docs/api/loaders.constants.html | 6 - docs/api/loaders.model.html | 6 - docs/api/loaders.patch_manager.html | 6 - docs/api/loaders.processor.html | 6 - docs/api/loaders.tokenizer.html | 6 - docs/api/logging_config.html | 6 - docs/api/models.mamba.modeling_mamba.html | 6 - .../monkeypatch.btlm_attn_hijack_flash.html | 6 - ...onkeypatch.data.batch_dataset_fetcher.html | 6 - ...ch.gradient_checkpointing.offload_cpu.html | 6 - ...h.gradient_checkpointing.offload_disk.html | 6 - .../monkeypatch.llama_attn_hijack_flash.html | 6 - ...onkeypatch.llama_attn_hijack_xformers.html | 6 - docs/api/monkeypatch.lora_kernels.html | 6 - ...monkeypatch.mistral_attn_hijack_flash.html | 6 - docs/api/monkeypatch.mixtral.html | 6 - docs/api/monkeypatch.multipack.html | 6 - docs/api/monkeypatch.relora.html | 6 - ...onkeypatch.stablelm_attn_hijack_flash.html | 6 - docs/api/monkeypatch.trainer_fsdp_optim.html | 6 - .../monkeypatch.transformers_fa_utils.html | 6 - docs/api/monkeypatch.unsloth_.html | 1178 --------- docs/api/monkeypatch.utils.html | 6 - docs/api/prompt_strategies.alpaca_chat.html | 6 - .../prompt_strategies.alpaca_instruct.html | 6 - .../prompt_strategies.alpaca_w_system.html | 6 - docs/api/prompt_strategies.base.html | 6 - ...rompt_strategies.bradley_terry.llama3.html | 6 - docs/api/prompt_strategies.chat_template.html | 6 - docs/api/prompt_strategies.completion.html | 6 - .../prompt_strategies.dpo.chat_template.html | 6 - docs/api/prompt_strategies.dpo.chatml.html | 6 - docs/api/prompt_strategies.dpo.llama3.html | 6 - .../prompt_strategies.dpo.passthrough.html | 6 - .../prompt_strategies.dpo.user_defined.html | 6 - docs/api/prompt_strategies.dpo.zephyr.html | 6 - docs/api/prompt_strategies.input_output.html | 6 - docs/api/prompt_strategies.kto.chatml.html | 6 - docs/api/prompt_strategies.kto.llama3.html | 6 - .../prompt_strategies.kto.user_defined.html | 6 - docs/api/prompt_strategies.llama2_chat.html | 6 - docs/api/prompt_strategies.messages.chat.html | 6 - docs/api/prompt_strategies.metharme.html | 6 - docs/api/prompt_strategies.orcamini.html | 6 - .../prompt_strategies.orpo.chat_template.html | 6 - docs/api/prompt_strategies.pygmalion.html | 6 - ...prompt_strategies.stepwise_supervised.html | 6 - docs/api/prompt_strategies.user_defined.html | 6 - docs/api/prompt_tokenizers.html | 6 - docs/api/train.html | 6 - docs/api/utils.bench.html | 6 - docs/api/utils.callbacks.comet_.html | 6 - docs/api/utils.callbacks.lisa.html | 6 - docs/api/utils.callbacks.mlflow_.html | 6 - docs/api/utils.callbacks.perplexity.html | 6 - docs/api/utils.callbacks.profiler.html | 6 - docs/api/utils.callbacks.qat.html | 6 - docs/api/utils.chat_templates.html | 6 - docs/api/utils.collators.batching.html | 6 - docs/api/utils.collators.core.html | 6 - docs/api/utils.collators.mamba.html | 6 - docs/api/utils.collators.mm_chat.html | 6 - .../utils.ctx_managers.sequence_parallel.html | 6 - docs/api/utils.data.sft.html | 6 - docs/api/utils.data.streaming.html | 6 - docs/api/utils.dict.html | 6 - docs/api/utils.distributed.html | 6 - docs/api/utils.freeze.html | 6 - docs/api/utils.lora.html | 6 - docs/api/utils.model_shard_quant.html | 6 - docs/api/utils.optimizers.adopt.html | 6 - docs/api/utils.quantization.html | 6 - docs/api/utils.samplers.multipack.html | 6 - docs/api/utils.schedulers.html | 6 - docs/api/utils.schemas.config.html | 6 - docs/api/utils.schemas.datasets.html | 6 - docs/api/utils.schemas.enums.html | 6 - docs/api/utils.schemas.integrations.html | 6 - docs/api/utils.schemas.model.html | 6 - docs/api/utils.schemas.multimodal.html | 6 - docs/api/utils.schemas.peft.html | 6 - docs/api/utils.schemas.training.html | 6 - docs/api/utils.schemas.trl.html | 6 - docs/api/utils.schemas.utils.html | 6 - docs/api/utils.tokenization.html | 6 - docs/api/utils.trainer.html | 6 - docs/attention.html | 6 - docs/batch_vs_grad.html | 6 - docs/checkpoint_saving.html | 6 - docs/choosing_method.html | 6 - docs/cli.html | 6 - docs/config-reference.html | 1235 +++++----- docs/custom_integrations.html | 6 - docs/dataset-formats/conversation.html | 6 - docs/dataset-formats/index.html | 6 - docs/dataset-formats/inst_tune.html | 6 - docs/dataset-formats/pretraining.html | 6 - docs/dataset-formats/stepwise_supervised.html | 6 - docs/dataset-formats/template_free.html | 6 - docs/dataset-formats/tokenized.html | 6 - docs/dataset_loading.html | 6 - docs/dataset_preprocessing.html | 6 - docs/debugging.html | 19 +- docs/docker.html | 133 +- docs/ebft.html | 6 - docs/expert_quantization.html | 6 - docs/faq.html | 6 - docs/fsdp_qlora.html | 6 - docs/getting-started.html | 6 - docs/gradient_checkpointing.html | 6 - docs/grpo.html | 6 - docs/inference.html | 6 - docs/input_output.html | 6 - docs/installation.html | 179 +- docs/lora_optims.html | 6 - docs/lr_groups.html | 6 - docs/mac.html | 6 - docs/mixed_precision.html | 6 - docs/models/LiquidAI.html | 13 +- docs/models/apertus.html | 21 +- docs/models/arcee.html | 15 +- docs/models/devstral.html | 9 +- docs/models/gemma3n.html | 13 +- docs/models/gpt-oss.html | 11 +- docs/models/granite4.html | 15 +- docs/models/hunyuan.html | 15 +- docs/models/index.html | 6 - docs/models/internvl3_5.html | 8 +- docs/models/jamba.html | 6 - docs/models/kimi-linear.html | 6 - docs/models/llama-2.html | 6 - docs/models/llama-4.html | 6 - docs/models/magistral.html | 9 +- docs/models/magistral/think.html | 6 - docs/models/magistral/vision.html | 8 +- docs/models/mimo.html | 6 - docs/models/ministral.html | 6 - docs/models/ministral3.html | 8 +- docs/models/ministral3/think.html | 6 - docs/models/ministral3/vision.html | 8 +- docs/models/mistral-small.html | 8 +- docs/models/mistral.html | 6 - docs/models/olmo3.html | 6 - docs/models/orpheus.html | 6 - docs/models/phi.html | 6 - docs/models/plano.html | 6 - docs/models/qwen3-next.html | 8 +- docs/models/qwen3.html | 6 - docs/models/seed-oss.html | 15 +- docs/models/smolvlm2.html | 11 +- docs/models/trinity.html | 6 - docs/models/voxtral.html | 13 +- docs/multi-gpu.html | 6 - docs/multi-node.html | 6 - docs/multimodal.html | 6 - docs/multipack.html | 6 - docs/nccl.html | 6 - docs/nd_parallelism.html | 6 - docs/optimizations.html | 6 - docs/optimizers.html | 6 - docs/qat.html | 6 - docs/quantize.html | 6 - docs/ray-integration.html | 6 - docs/reward_modelling.html | 6 - docs/rlhf.html | 6 - docs/sequence_parallelism.html | 6 - docs/streaming.html | 6 - docs/telemetry.html | 6 - docs/torchao.html | 6 - docs/training_stability.html | 6 - docs/unsloth.html | 1281 ---------- docs/vllm_serving.html | 6 - .../colab-axolotl-example.html | 6 - index.html | 115 +- search.json | 2151 ++++++++--------- sitemap.xml | 1424 ++++++----- src/axolotl/integrations/LICENSE.html | 6 - .../cut_cross_entropy/ACKNOWLEDGEMENTS.html | 6 - 250 files changed, 2674 insertions(+), 6593 deletions(-) delete mode 100644 docs/api/monkeypatch.unsloth_.html delete mode 100644 docs/unsloth.html diff --git a/.nojekyll b/.nojekyll index 5361f5b47..d30a5b91b 100644 --- a/.nojekyll +++ b/.nojekyll @@ -1 +1 @@ -11c88725 \ No newline at end of file +1cf0992e \ No newline at end of file diff --git a/FAQS.html b/FAQS.html index 85ac56908..455a991fb 100644 --- a/FAQS.html +++ b/FAQS.html @@ -661,12 +661,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA - -
  • -
    - - Unsloth -
  • diff --git a/docs/agents/grpo.html b/docs/agents/grpo.html index 2681152f4..9b26a4231 100644 --- a/docs/agents/grpo.html +++ b/docs/agents/grpo.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/agents/model_architectures.html b/docs/agents/model_architectures.html index b0373a116..241216422 100644 --- a/docs/agents/model_architectures.html +++ b/docs/agents/model_architectures.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/agents/new_model_support.html b/docs/agents/new_model_support.html index 671db5ade..0e969ab01 100644 --- a/docs/agents/new_model_support.html +++ b/docs/agents/new_model_support.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/agents/preference_tuning.html b/docs/agents/preference_tuning.html index 0c3381039..033910dbd 100644 --- a/docs/agents/preference_tuning.html +++ b/docs/agents/preference_tuning.html @@ -661,12 +661,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/agents/pretraining.html b/docs/agents/pretraining.html index fe2caf9aa..a07fce09e 100644 --- a/docs/agents/pretraining.html +++ b/docs/agents/pretraining.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/agents/reward_modelling.html b/docs/agents/reward_modelling.html index 1504a6349..d4babe887 100644 --- a/docs/agents/reward_modelling.html +++ b/docs/agents/reward_modelling.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/agents/sft.html b/docs/agents/sft.html index 4a6080cc2..9f7d18389 100644 --- a/docs/agents/sft.html +++ b/docs/agents/sft.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/amd_hpc.html b/docs/amd_hpc.html index 83d66b3b2..aafc33d48 100644 --- a/docs/amd_hpc.html +++ b/docs/amd_hpc.html @@ -697,12 +697,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/cli.args.html b/docs/api/cli.args.html index 360de7c1e..ce24a26de 100644 --- a/docs/api/cli.args.html +++ b/docs/api/cli.args.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/cli.art.html b/docs/api/cli.art.html index 9790aef65..1996b1b60 100644 --- a/docs/api/cli.art.html +++ b/docs/api/cli.art.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/cli.checks.html b/docs/api/cli.checks.html index 9d2277f1b..d199c5ea0 100644 --- a/docs/api/cli.checks.html +++ b/docs/api/cli.checks.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/cli.cloud.base.html b/docs/api/cli.cloud.base.html index 24fe7624c..564f2e7d7 100644 --- a/docs/api/cli.cloud.base.html +++ b/docs/api/cli.cloud.base.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/cli.cloud.modal_.html b/docs/api/cli.cloud.modal_.html index c210733a3..fc3c25f45 100644 --- a/docs/api/cli.cloud.modal_.html +++ b/docs/api/cli.cloud.modal_.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/cli.config.html b/docs/api/cli.config.html index 15fc4aefb..d9dc5e775 100644 --- a/docs/api/cli.config.html +++ b/docs/api/cli.config.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/cli.delinearize_llama4.html b/docs/api/cli.delinearize_llama4.html index 7bf3c10a8..62527256d 100644 --- a/docs/api/cli.delinearize_llama4.html +++ b/docs/api/cli.delinearize_llama4.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/cli.evaluate.html b/docs/api/cli.evaluate.html index d45db111b..10a4ad3f8 100644 --- a/docs/api/cli.evaluate.html +++ b/docs/api/cli.evaluate.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/cli.inference.html b/docs/api/cli.inference.html index 1a1297816..6a278657d 100644 --- a/docs/api/cli.inference.html +++ b/docs/api/cli.inference.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/cli.main.html b/docs/api/cli.main.html index d1e0bc563..b0fc7f9d9 100644 --- a/docs/api/cli.main.html +++ b/docs/api/cli.main.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/cli.merge_lora.html b/docs/api/cli.merge_lora.html index c3c7ec5d5..3b1a098cb 100644 --- a/docs/api/cli.merge_lora.html +++ b/docs/api/cli.merge_lora.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/cli.merge_sharded_fsdp_weights.html b/docs/api/cli.merge_sharded_fsdp_weights.html index 6a13f77bf..a6833ce04 100644 --- a/docs/api/cli.merge_sharded_fsdp_weights.html +++ b/docs/api/cli.merge_sharded_fsdp_weights.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/cli.preprocess.html b/docs/api/cli.preprocess.html index 181bc1029..5f966ff26 100644 --- a/docs/api/cli.preprocess.html +++ b/docs/api/cli.preprocess.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/cli.quantize.html b/docs/api/cli.quantize.html index c2cf00a91..436358b07 100644 --- a/docs/api/cli.quantize.html +++ b/docs/api/cli.quantize.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/cli.train.html b/docs/api/cli.train.html index 6a6871ec5..906619b39 100644 --- a/docs/api/cli.train.html +++ b/docs/api/cli.train.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/cli.utils.args.html b/docs/api/cli.utils.args.html index 9d683903a..2eb1aec0a 100644 --- a/docs/api/cli.utils.args.html +++ b/docs/api/cli.utils.args.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/cli.utils.fetch.html b/docs/api/cli.utils.fetch.html index f0b0ba480..7811e3852 100644 --- a/docs/api/cli.utils.fetch.html +++ b/docs/api/cli.utils.fetch.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/cli.utils.html b/docs/api/cli.utils.html index aa0423cfc..758a86bba 100644 --- a/docs/api/cli.utils.html +++ b/docs/api/cli.utils.html @@ -661,12 +661,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/cli.utils.load.html b/docs/api/cli.utils.load.html index d27c2b954..4a3759d12 100644 --- a/docs/api/cli.utils.load.html +++ b/docs/api/cli.utils.load.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/cli.utils.sweeps.html b/docs/api/cli.utils.sweeps.html index bbcc9d3d3..68aefb0a4 100644 --- a/docs/api/cli.utils.sweeps.html +++ b/docs/api/cli.utils.sweeps.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/cli.utils.train.html b/docs/api/cli.utils.train.html index 187feac01..747fbf5fa 100644 --- a/docs/api/cli.utils.train.html +++ b/docs/api/cli.utils.train.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/cli.vllm_serve.html b/docs/api/cli.vllm_serve.html index 8b022a820..6a952f15b 100644 --- a/docs/api/cli.vllm_serve.html +++ b/docs/api/cli.vllm_serve.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/common.architectures.html b/docs/api/common.architectures.html index 03cec8109..1651c9218 100644 --- a/docs/api/common.architectures.html +++ b/docs/api/common.architectures.html @@ -661,12 +661,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/common.const.html b/docs/api/common.const.html index 2b0620700..94a86302d 100644 --- a/docs/api/common.const.html +++ b/docs/api/common.const.html @@ -661,12 +661,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/common.datasets.html b/docs/api/common.datasets.html index 8a667affb..03815136b 100644 --- a/docs/api/common.datasets.html +++ b/docs/api/common.datasets.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/convert.html b/docs/api/convert.html index 27daf3fc8..c5cb0820d 100644 --- a/docs/api/convert.html +++ b/docs/api/convert.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/core.builders.base.html b/docs/api/core.builders.base.html index 121ee3825..0589c2fc1 100644 --- a/docs/api/core.builders.base.html +++ b/docs/api/core.builders.base.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/core.builders.causal.html b/docs/api/core.builders.causal.html index a04129309..befa3267d 100644 --- a/docs/api/core.builders.causal.html +++ b/docs/api/core.builders.causal.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/core.builders.rl.html b/docs/api/core.builders.rl.html index de583248e..f1a78d4b8 100644 --- a/docs/api/core.builders.rl.html +++ b/docs/api/core.builders.rl.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/core.chat.format.chatml.html b/docs/api/core.chat.format.chatml.html index 89a651ae3..c40910309 100644 --- a/docs/api/core.chat.format.chatml.html +++ b/docs/api/core.chat.format.chatml.html @@ -661,12 +661,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/core.chat.format.llama3x.html b/docs/api/core.chat.format.llama3x.html index e3784d346..f59defbe5 100644 --- a/docs/api/core.chat.format.llama3x.html +++ b/docs/api/core.chat.format.llama3x.html @@ -661,12 +661,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/core.chat.format.shared.html b/docs/api/core.chat.format.shared.html index 449575bee..b99bc3e5c 100644 --- a/docs/api/core.chat.format.shared.html +++ b/docs/api/core.chat.format.shared.html @@ -661,12 +661,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/core.chat.messages.html b/docs/api/core.chat.messages.html index 8bdd25961..1594b4de4 100644 --- a/docs/api/core.chat.messages.html +++ b/docs/api/core.chat.messages.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/core.datasets.chat.html b/docs/api/core.datasets.chat.html index 7ff985de0..69a63c652 100644 --- a/docs/api/core.datasets.chat.html +++ b/docs/api/core.datasets.chat.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/core.datasets.transforms.chat_builder.html b/docs/api/core.datasets.transforms.chat_builder.html index 023cf8267..2bdd19e5d 100644 --- a/docs/api/core.datasets.transforms.chat_builder.html +++ b/docs/api/core.datasets.transforms.chat_builder.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/core.trainers.base.html b/docs/api/core.trainers.base.html index 2f88a9997..7a87857cb 100644 --- a/docs/api/core.trainers.base.html +++ b/docs/api/core.trainers.base.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/core.trainers.dpo.trainer.html b/docs/api/core.trainers.dpo.trainer.html index 4aafef517..3f768a695 100644 --- a/docs/api/core.trainers.dpo.trainer.html +++ b/docs/api/core.trainers.dpo.trainer.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/core.trainers.grpo.sampler.html b/docs/api/core.trainers.grpo.sampler.html index 7355688e7..9a13bec21 100644 --- a/docs/api/core.trainers.grpo.sampler.html +++ b/docs/api/core.trainers.grpo.sampler.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/core.trainers.grpo.trainer.html b/docs/api/core.trainers.grpo.trainer.html index d233df546..44a9dec91 100644 --- a/docs/api/core.trainers.grpo.trainer.html +++ b/docs/api/core.trainers.grpo.trainer.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/core.trainers.mamba.html b/docs/api/core.trainers.mamba.html index 3d86dc3a7..54d681c9c 100644 --- a/docs/api/core.trainers.mamba.html +++ b/docs/api/core.trainers.mamba.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/core.trainers.mixins.optimizer.html b/docs/api/core.trainers.mixins.optimizer.html index fd52e7de4..b94221985 100644 --- a/docs/api/core.trainers.mixins.optimizer.html +++ b/docs/api/core.trainers.mixins.optimizer.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/core.trainers.mixins.rng_state_loader.html b/docs/api/core.trainers.mixins.rng_state_loader.html index 73567ec93..5676960f9 100644 --- a/docs/api/core.trainers.mixins.rng_state_loader.html +++ b/docs/api/core.trainers.mixins.rng_state_loader.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/core.trainers.mixins.scheduler.html b/docs/api/core.trainers.mixins.scheduler.html index 8bd41625d..3e108c55b 100644 --- a/docs/api/core.trainers.mixins.scheduler.html +++ b/docs/api/core.trainers.mixins.scheduler.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/core.trainers.trl.html b/docs/api/core.trainers.trl.html index d0a755e3c..aa6ca081c 100644 --- a/docs/api/core.trainers.trl.html +++ b/docs/api/core.trainers.trl.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/core.trainers.utils.html b/docs/api/core.trainers.utils.html index 14886cb60..124be2523 100644 --- a/docs/api/core.trainers.utils.html +++ b/docs/api/core.trainers.utils.html @@ -661,12 +661,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/core.training_args.html b/docs/api/core.training_args.html index 314015780..67d146d57 100644 --- a/docs/api/core.training_args.html +++ b/docs/api/core.training_args.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/datasets.html b/docs/api/datasets.html index 39107f342..6beec978c 100644 --- a/docs/api/datasets.html +++ b/docs/api/datasets.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/evaluate.html b/docs/api/evaluate.html index 14820c0ca..b809f5f4b 100644 --- a/docs/api/evaluate.html +++ b/docs/api/evaluate.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/index.html b/docs/api/index.html index b1591720e..80ebae651 100644 --- a/docs/api/index.html +++ b/docs/api/index.html @@ -661,12 +661,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • @@ -1235,22 +1229,18 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); see https://github.com/huggingface/transformers/pull/35834 -monkeypatch.unsloth_ -module for patching with unsloth optimizations - - monkeypatch.data.batch_dataset_fetcher Monkey patches for the dataset fetcher to handle batches of packed indexes. - + monkeypatch.mixtral Patches to support multipack for mixtral - + monkeypatch.gradient_checkpointing.offload_cpu CPU offloaded checkpointing - + monkeypatch.gradient_checkpointing.offload_disk DISCO - DIsk-based Storage and Checkpointing with Optimized prefetching diff --git a/docs/api/integrations.base.html b/docs/api/integrations.base.html index 626a4a0e6..88edc7796 100644 --- a/docs/api/integrations.base.html +++ b/docs/api/integrations.base.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/integrations.cut_cross_entropy.args.html b/docs/api/integrations.cut_cross_entropy.args.html index 8326e1253..98512a5c4 100644 --- a/docs/api/integrations.cut_cross_entropy.args.html +++ b/docs/api/integrations.cut_cross_entropy.args.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/integrations.grokfast.optimizer.html b/docs/api/integrations.grokfast.optimizer.html index e956b528a..1b0d28e97 100644 --- a/docs/api/integrations.grokfast.optimizer.html +++ b/docs/api/integrations.grokfast.optimizer.html @@ -661,12 +661,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/integrations.kd.trainer.html b/docs/api/integrations.kd.trainer.html index ac8a11c05..f03530545 100644 --- a/docs/api/integrations.kd.trainer.html +++ b/docs/api/integrations.kd.trainer.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/integrations.liger.args.html b/docs/api/integrations.liger.args.html index fdb8606c7..1dcfb9d98 100644 --- a/docs/api/integrations.liger.args.html +++ b/docs/api/integrations.liger.args.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/integrations.lm_eval.args.html b/docs/api/integrations.lm_eval.args.html index 3d72be59b..3bcb390c7 100644 --- a/docs/api/integrations.lm_eval.args.html +++ b/docs/api/integrations.lm_eval.args.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/integrations.spectrum.args.html b/docs/api/integrations.spectrum.args.html index 34e243720..a0186b920 100644 --- a/docs/api/integrations.spectrum.args.html +++ b/docs/api/integrations.spectrum.args.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/kernels.geglu.html b/docs/api/kernels.geglu.html index a3a613080..9b8cf5fc5 100644 --- a/docs/api/kernels.geglu.html +++ b/docs/api/kernels.geglu.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/kernels.lora.html b/docs/api/kernels.lora.html index 351391e21..6bc81139b 100644 --- a/docs/api/kernels.lora.html +++ b/docs/api/kernels.lora.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/kernels.quantize.html b/docs/api/kernels.quantize.html index 967edc147..a2cbf26e3 100644 --- a/docs/api/kernels.quantize.html +++ b/docs/api/kernels.quantize.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/kernels.swiglu.html b/docs/api/kernels.swiglu.html index d0bacc428..071605607 100644 --- a/docs/api/kernels.swiglu.html +++ b/docs/api/kernels.swiglu.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/kernels.utils.html b/docs/api/kernels.utils.html index 0ae8885eb..250fd9592 100644 --- a/docs/api/kernels.utils.html +++ b/docs/api/kernels.utils.html @@ -661,12 +661,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/loaders.adapter.html b/docs/api/loaders.adapter.html index a452cd10c..e2ce542f1 100644 --- a/docs/api/loaders.adapter.html +++ b/docs/api/loaders.adapter.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/loaders.constants.html b/docs/api/loaders.constants.html index 80773c313..50c76031b 100644 --- a/docs/api/loaders.constants.html +++ b/docs/api/loaders.constants.html @@ -661,12 +661,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/loaders.model.html b/docs/api/loaders.model.html index 87dd915cd..82238cdcc 100644 --- a/docs/api/loaders.model.html +++ b/docs/api/loaders.model.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/loaders.patch_manager.html b/docs/api/loaders.patch_manager.html index 095a4a213..51496475a 100644 --- a/docs/api/loaders.patch_manager.html +++ b/docs/api/loaders.patch_manager.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/loaders.processor.html b/docs/api/loaders.processor.html index 151712866..03aaae9a7 100644 --- a/docs/api/loaders.processor.html +++ b/docs/api/loaders.processor.html @@ -661,12 +661,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/loaders.tokenizer.html b/docs/api/loaders.tokenizer.html index 6ab3e0904..194277055 100644 --- a/docs/api/loaders.tokenizer.html +++ b/docs/api/loaders.tokenizer.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/logging_config.html b/docs/api/logging_config.html index 76673ad72..2046710a3 100644 --- a/docs/api/logging_config.html +++ b/docs/api/logging_config.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/models.mamba.modeling_mamba.html b/docs/api/models.mamba.modeling_mamba.html index 644951b85..55f3a7d26 100644 --- a/docs/api/models.mamba.modeling_mamba.html +++ b/docs/api/models.mamba.modeling_mamba.html @@ -661,12 +661,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/monkeypatch.btlm_attn_hijack_flash.html b/docs/api/monkeypatch.btlm_attn_hijack_flash.html index ba8601973..657ed8bbf 100644 --- a/docs/api/monkeypatch.btlm_attn_hijack_flash.html +++ b/docs/api/monkeypatch.btlm_attn_hijack_flash.html @@ -661,12 +661,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/monkeypatch.data.batch_dataset_fetcher.html b/docs/api/monkeypatch.data.batch_dataset_fetcher.html index 0d285334f..0ee8fb678 100644 --- a/docs/api/monkeypatch.data.batch_dataset_fetcher.html +++ b/docs/api/monkeypatch.data.batch_dataset_fetcher.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/monkeypatch.gradient_checkpointing.offload_cpu.html b/docs/api/monkeypatch.gradient_checkpointing.offload_cpu.html index c3f9aec25..b79178fe0 100644 --- a/docs/api/monkeypatch.gradient_checkpointing.offload_cpu.html +++ b/docs/api/monkeypatch.gradient_checkpointing.offload_cpu.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/monkeypatch.gradient_checkpointing.offload_disk.html b/docs/api/monkeypatch.gradient_checkpointing.offload_disk.html index deb5aee15..04d6736c0 100644 --- a/docs/api/monkeypatch.gradient_checkpointing.offload_disk.html +++ b/docs/api/monkeypatch.gradient_checkpointing.offload_disk.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/monkeypatch.llama_attn_hijack_flash.html b/docs/api/monkeypatch.llama_attn_hijack_flash.html index 8162cdf74..c04044558 100644 --- a/docs/api/monkeypatch.llama_attn_hijack_flash.html +++ b/docs/api/monkeypatch.llama_attn_hijack_flash.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/monkeypatch.llama_attn_hijack_xformers.html b/docs/api/monkeypatch.llama_attn_hijack_xformers.html index e272b89e7..e79fac067 100644 --- a/docs/api/monkeypatch.llama_attn_hijack_xformers.html +++ b/docs/api/monkeypatch.llama_attn_hijack_xformers.html @@ -661,12 +661,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/monkeypatch.lora_kernels.html b/docs/api/monkeypatch.lora_kernels.html index 0ed8407ac..b33e6dbdb 100644 --- a/docs/api/monkeypatch.lora_kernels.html +++ b/docs/api/monkeypatch.lora_kernels.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/monkeypatch.mistral_attn_hijack_flash.html b/docs/api/monkeypatch.mistral_attn_hijack_flash.html index 75e658aab..ec38f3830 100644 --- a/docs/api/monkeypatch.mistral_attn_hijack_flash.html +++ b/docs/api/monkeypatch.mistral_attn_hijack_flash.html @@ -661,12 +661,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/monkeypatch.mixtral.html b/docs/api/monkeypatch.mixtral.html index 5161a4ac4..f929fc66a 100644 --- a/docs/api/monkeypatch.mixtral.html +++ b/docs/api/monkeypatch.mixtral.html @@ -661,12 +661,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/monkeypatch.multipack.html b/docs/api/monkeypatch.multipack.html index 45192896b..82ae2b744 100644 --- a/docs/api/monkeypatch.multipack.html +++ b/docs/api/monkeypatch.multipack.html @@ -661,12 +661,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/monkeypatch.relora.html b/docs/api/monkeypatch.relora.html index 1b126188b..e2eecfb6a 100644 --- a/docs/api/monkeypatch.relora.html +++ b/docs/api/monkeypatch.relora.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/monkeypatch.stablelm_attn_hijack_flash.html b/docs/api/monkeypatch.stablelm_attn_hijack_flash.html index fe3db4457..1c1eda79f 100644 --- a/docs/api/monkeypatch.stablelm_attn_hijack_flash.html +++ b/docs/api/monkeypatch.stablelm_attn_hijack_flash.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/monkeypatch.trainer_fsdp_optim.html b/docs/api/monkeypatch.trainer_fsdp_optim.html index 2fb07220c..b22d8e03a 100644 --- a/docs/api/monkeypatch.trainer_fsdp_optim.html +++ b/docs/api/monkeypatch.trainer_fsdp_optim.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/monkeypatch.transformers_fa_utils.html b/docs/api/monkeypatch.transformers_fa_utils.html index 83cff2f68..8e211c2e1 100644 --- a/docs/api/monkeypatch.transformers_fa_utils.html +++ b/docs/api/monkeypatch.transformers_fa_utils.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/monkeypatch.unsloth_.html b/docs/api/monkeypatch.unsloth_.html deleted file mode 100644 index 7d5a95bae..000000000 --- a/docs/api/monkeypatch.unsloth_.html +++ /dev/null @@ -1,1178 +0,0 @@ - - - - - - - - - -monkeypatch.unsloth_ – Axolotl - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
    -
    - - -
    - -
    - - -
    - - - -
    - - - - -
    -

    monkeypatch.unsloth_

    -

    monkeypatch.unsloth_

    -

    module for patching with unsloth optimizations

    - - -
    - -
    - -
    - - - - - \ No newline at end of file diff --git a/docs/api/monkeypatch.utils.html b/docs/api/monkeypatch.utils.html index b70879e72..46e6ee3c8 100644 --- a/docs/api/monkeypatch.utils.html +++ b/docs/api/monkeypatch.utils.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/prompt_strategies.alpaca_chat.html b/docs/api/prompt_strategies.alpaca_chat.html index 61a241151..b72c567aa 100644 --- a/docs/api/prompt_strategies.alpaca_chat.html +++ b/docs/api/prompt_strategies.alpaca_chat.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/prompt_strategies.alpaca_instruct.html b/docs/api/prompt_strategies.alpaca_instruct.html index c87a6bf96..64a70b400 100644 --- a/docs/api/prompt_strategies.alpaca_instruct.html +++ b/docs/api/prompt_strategies.alpaca_instruct.html @@ -661,12 +661,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/prompt_strategies.alpaca_w_system.html b/docs/api/prompt_strategies.alpaca_w_system.html index 7e9eee92a..22c6cccb7 100644 --- a/docs/api/prompt_strategies.alpaca_w_system.html +++ b/docs/api/prompt_strategies.alpaca_w_system.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/prompt_strategies.base.html b/docs/api/prompt_strategies.base.html index ac814121a..7041883f9 100644 --- a/docs/api/prompt_strategies.base.html +++ b/docs/api/prompt_strategies.base.html @@ -661,12 +661,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/prompt_strategies.bradley_terry.llama3.html b/docs/api/prompt_strategies.bradley_terry.llama3.html index 51c0fbd3b..c0932768a 100644 --- a/docs/api/prompt_strategies.bradley_terry.llama3.html +++ b/docs/api/prompt_strategies.bradley_terry.llama3.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/prompt_strategies.chat_template.html b/docs/api/prompt_strategies.chat_template.html index 209bc339a..1c013b943 100644 --- a/docs/api/prompt_strategies.chat_template.html +++ b/docs/api/prompt_strategies.chat_template.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/prompt_strategies.completion.html b/docs/api/prompt_strategies.completion.html index 127409e87..7a515ea70 100644 --- a/docs/api/prompt_strategies.completion.html +++ b/docs/api/prompt_strategies.completion.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/prompt_strategies.dpo.chat_template.html b/docs/api/prompt_strategies.dpo.chat_template.html index 9daea51f7..4d8bfd5d9 100644 --- a/docs/api/prompt_strategies.dpo.chat_template.html +++ b/docs/api/prompt_strategies.dpo.chat_template.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/prompt_strategies.dpo.chatml.html b/docs/api/prompt_strategies.dpo.chatml.html index 228a241ca..14d19b781 100644 --- a/docs/api/prompt_strategies.dpo.chatml.html +++ b/docs/api/prompt_strategies.dpo.chatml.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/prompt_strategies.dpo.llama3.html b/docs/api/prompt_strategies.dpo.llama3.html index d258a0cc9..36ed092e7 100644 --- a/docs/api/prompt_strategies.dpo.llama3.html +++ b/docs/api/prompt_strategies.dpo.llama3.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/prompt_strategies.dpo.passthrough.html b/docs/api/prompt_strategies.dpo.passthrough.html index 8f7ee5273..48fc2e1a7 100644 --- a/docs/api/prompt_strategies.dpo.passthrough.html +++ b/docs/api/prompt_strategies.dpo.passthrough.html @@ -661,12 +661,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/prompt_strategies.dpo.user_defined.html b/docs/api/prompt_strategies.dpo.user_defined.html index cc8a87411..4074955ab 100644 --- a/docs/api/prompt_strategies.dpo.user_defined.html +++ b/docs/api/prompt_strategies.dpo.user_defined.html @@ -661,12 +661,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/prompt_strategies.dpo.zephyr.html b/docs/api/prompt_strategies.dpo.zephyr.html index 73ea28ef9..2db251384 100644 --- a/docs/api/prompt_strategies.dpo.zephyr.html +++ b/docs/api/prompt_strategies.dpo.zephyr.html @@ -661,12 +661,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/prompt_strategies.input_output.html b/docs/api/prompt_strategies.input_output.html index 3a58bb4d6..a9f0c142a 100644 --- a/docs/api/prompt_strategies.input_output.html +++ b/docs/api/prompt_strategies.input_output.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/prompt_strategies.kto.chatml.html b/docs/api/prompt_strategies.kto.chatml.html index 0a2252668..66c53009e 100644 --- a/docs/api/prompt_strategies.kto.chatml.html +++ b/docs/api/prompt_strategies.kto.chatml.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/prompt_strategies.kto.llama3.html b/docs/api/prompt_strategies.kto.llama3.html index ab4d4313e..b63464b74 100644 --- a/docs/api/prompt_strategies.kto.llama3.html +++ b/docs/api/prompt_strategies.kto.llama3.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/prompt_strategies.kto.user_defined.html b/docs/api/prompt_strategies.kto.user_defined.html index ed2b14f87..8effa9aa8 100644 --- a/docs/api/prompt_strategies.kto.user_defined.html +++ b/docs/api/prompt_strategies.kto.user_defined.html @@ -661,12 +661,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/prompt_strategies.llama2_chat.html b/docs/api/prompt_strategies.llama2_chat.html index 7647ee07c..a31d34516 100644 --- a/docs/api/prompt_strategies.llama2_chat.html +++ b/docs/api/prompt_strategies.llama2_chat.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/prompt_strategies.messages.chat.html b/docs/api/prompt_strategies.messages.chat.html index 00811aaad..4b0029e52 100644 --- a/docs/api/prompt_strategies.messages.chat.html +++ b/docs/api/prompt_strategies.messages.chat.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/prompt_strategies.metharme.html b/docs/api/prompt_strategies.metharme.html index 78c13c0c4..aa612c7e9 100644 --- a/docs/api/prompt_strategies.metharme.html +++ b/docs/api/prompt_strategies.metharme.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/prompt_strategies.orcamini.html b/docs/api/prompt_strategies.orcamini.html index 4dd22d983..e6a2f3a9c 100644 --- a/docs/api/prompt_strategies.orcamini.html +++ b/docs/api/prompt_strategies.orcamini.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/prompt_strategies.orpo.chat_template.html b/docs/api/prompt_strategies.orpo.chat_template.html index 3355711fb..fa4cf939c 100644 --- a/docs/api/prompt_strategies.orpo.chat_template.html +++ b/docs/api/prompt_strategies.orpo.chat_template.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/prompt_strategies.pygmalion.html b/docs/api/prompt_strategies.pygmalion.html index 4a3bc73b2..05bd0a187 100644 --- a/docs/api/prompt_strategies.pygmalion.html +++ b/docs/api/prompt_strategies.pygmalion.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/prompt_strategies.stepwise_supervised.html b/docs/api/prompt_strategies.stepwise_supervised.html index 97daea234..d8e747e03 100644 --- a/docs/api/prompt_strategies.stepwise_supervised.html +++ b/docs/api/prompt_strategies.stepwise_supervised.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/prompt_strategies.user_defined.html b/docs/api/prompt_strategies.user_defined.html index 46afe34a8..6220251e7 100644 --- a/docs/api/prompt_strategies.user_defined.html +++ b/docs/api/prompt_strategies.user_defined.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/prompt_tokenizers.html b/docs/api/prompt_tokenizers.html index 8c62d9675..c1a6041aa 100644 --- a/docs/api/prompt_tokenizers.html +++ b/docs/api/prompt_tokenizers.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/train.html b/docs/api/train.html index d785a92fe..5635662a2 100644 --- a/docs/api/train.html +++ b/docs/api/train.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/utils.bench.html b/docs/api/utils.bench.html index c30ecaf51..e7b6ab70b 100644 --- a/docs/api/utils.bench.html +++ b/docs/api/utils.bench.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/utils.callbacks.comet_.html b/docs/api/utils.callbacks.comet_.html index c6f602624..e2ee816f5 100644 --- a/docs/api/utils.callbacks.comet_.html +++ b/docs/api/utils.callbacks.comet_.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/utils.callbacks.lisa.html b/docs/api/utils.callbacks.lisa.html index bb40e4071..cf3c11a97 100644 --- a/docs/api/utils.callbacks.lisa.html +++ b/docs/api/utils.callbacks.lisa.html @@ -661,12 +661,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/utils.callbacks.mlflow_.html b/docs/api/utils.callbacks.mlflow_.html index f4d767948..254401142 100644 --- a/docs/api/utils.callbacks.mlflow_.html +++ b/docs/api/utils.callbacks.mlflow_.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/utils.callbacks.perplexity.html b/docs/api/utils.callbacks.perplexity.html index bdd380ebf..557f64396 100644 --- a/docs/api/utils.callbacks.perplexity.html +++ b/docs/api/utils.callbacks.perplexity.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/utils.callbacks.profiler.html b/docs/api/utils.callbacks.profiler.html index b69afaea5..58f2d8313 100644 --- a/docs/api/utils.callbacks.profiler.html +++ b/docs/api/utils.callbacks.profiler.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/utils.callbacks.qat.html b/docs/api/utils.callbacks.qat.html index 885b9ae31..75d425301 100644 --- a/docs/api/utils.callbacks.qat.html +++ b/docs/api/utils.callbacks.qat.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/utils.chat_templates.html b/docs/api/utils.chat_templates.html index 20e01791d..f5ae5090a 100644 --- a/docs/api/utils.chat_templates.html +++ b/docs/api/utils.chat_templates.html @@ -661,12 +661,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/utils.collators.batching.html b/docs/api/utils.collators.batching.html index 22f974ac3..c2f666eef 100644 --- a/docs/api/utils.collators.batching.html +++ b/docs/api/utils.collators.batching.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/utils.collators.core.html b/docs/api/utils.collators.core.html index 69b224a10..897b9f699 100644 --- a/docs/api/utils.collators.core.html +++ b/docs/api/utils.collators.core.html @@ -661,12 +661,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/utils.collators.mamba.html b/docs/api/utils.collators.mamba.html index 48e0cbd8b..83944f998 100644 --- a/docs/api/utils.collators.mamba.html +++ b/docs/api/utils.collators.mamba.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/utils.collators.mm_chat.html b/docs/api/utils.collators.mm_chat.html index fd2f39b31..c18253c40 100644 --- a/docs/api/utils.collators.mm_chat.html +++ b/docs/api/utils.collators.mm_chat.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/utils.ctx_managers.sequence_parallel.html b/docs/api/utils.ctx_managers.sequence_parallel.html index 5d9263f99..add8dcd98 100644 --- a/docs/api/utils.ctx_managers.sequence_parallel.html +++ b/docs/api/utils.ctx_managers.sequence_parallel.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/utils.data.sft.html b/docs/api/utils.data.sft.html index 97b4acb48..150d91487 100644 --- a/docs/api/utils.data.sft.html +++ b/docs/api/utils.data.sft.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/utils.data.streaming.html b/docs/api/utils.data.streaming.html index 6b05e9b5c..ae6b42fd8 100644 --- a/docs/api/utils.data.streaming.html +++ b/docs/api/utils.data.streaming.html @@ -661,12 +661,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/utils.dict.html b/docs/api/utils.dict.html index caf8d3229..0dff30abd 100644 --- a/docs/api/utils.dict.html +++ b/docs/api/utils.dict.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/utils.distributed.html b/docs/api/utils.distributed.html index 60d38b7d3..a984482d1 100644 --- a/docs/api/utils.distributed.html +++ b/docs/api/utils.distributed.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/utils.freeze.html b/docs/api/utils.freeze.html index 256970e4a..a2dd93523 100644 --- a/docs/api/utils.freeze.html +++ b/docs/api/utils.freeze.html @@ -725,12 +725,6 @@ window.Quarto = { FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/utils.lora.html b/docs/api/utils.lora.html index 818affacd..471d8113b 100644 --- a/docs/api/utils.lora.html +++ b/docs/api/utils.lora.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/utils.model_shard_quant.html b/docs/api/utils.model_shard_quant.html index de0c7fb50..d5146dba9 100644 --- a/docs/api/utils.model_shard_quant.html +++ b/docs/api/utils.model_shard_quant.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/utils.optimizers.adopt.html b/docs/api/utils.optimizers.adopt.html index 2dbb0928c..d7236ce1d 100644 --- a/docs/api/utils.optimizers.adopt.html +++ b/docs/api/utils.optimizers.adopt.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/utils.quantization.html b/docs/api/utils.quantization.html index 23093e374..b04223268 100644 --- a/docs/api/utils.quantization.html +++ b/docs/api/utils.quantization.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/utils.samplers.multipack.html b/docs/api/utils.samplers.multipack.html index b6fec1508..f0a2583b4 100644 --- a/docs/api/utils.samplers.multipack.html +++ b/docs/api/utils.samplers.multipack.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/utils.schedulers.html b/docs/api/utils.schedulers.html index 5f316f31f..d69128171 100644 --- a/docs/api/utils.schedulers.html +++ b/docs/api/utils.schedulers.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/utils.schemas.config.html b/docs/api/utils.schemas.config.html index 2bad2e5a6..ab05dd163 100644 --- a/docs/api/utils.schemas.config.html +++ b/docs/api/utils.schemas.config.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/utils.schemas.datasets.html b/docs/api/utils.schemas.datasets.html index ad46eae8d..fe0a79573 100644 --- a/docs/api/utils.schemas.datasets.html +++ b/docs/api/utils.schemas.datasets.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/utils.schemas.enums.html b/docs/api/utils.schemas.enums.html index 7732e77d4..cf574f4fa 100644 --- a/docs/api/utils.schemas.enums.html +++ b/docs/api/utils.schemas.enums.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/utils.schemas.integrations.html b/docs/api/utils.schemas.integrations.html index 56871e2dc..8a7fdeea2 100644 --- a/docs/api/utils.schemas.integrations.html +++ b/docs/api/utils.schemas.integrations.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/utils.schemas.model.html b/docs/api/utils.schemas.model.html index 480586a36..559ebba9e 100644 --- a/docs/api/utils.schemas.model.html +++ b/docs/api/utils.schemas.model.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/utils.schemas.multimodal.html b/docs/api/utils.schemas.multimodal.html index 64e0541f9..5df7bc8cd 100644 --- a/docs/api/utils.schemas.multimodal.html +++ b/docs/api/utils.schemas.multimodal.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/utils.schemas.peft.html b/docs/api/utils.schemas.peft.html index 89bb07a90..4c4e236ea 100644 --- a/docs/api/utils.schemas.peft.html +++ b/docs/api/utils.schemas.peft.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/utils.schemas.training.html b/docs/api/utils.schemas.training.html index 876d09db5..a988aa328 100644 --- a/docs/api/utils.schemas.training.html +++ b/docs/api/utils.schemas.training.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/utils.schemas.trl.html b/docs/api/utils.schemas.trl.html index f0d2c95d1..6d020c01a 100644 --- a/docs/api/utils.schemas.trl.html +++ b/docs/api/utils.schemas.trl.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/utils.schemas.utils.html b/docs/api/utils.schemas.utils.html index fe0f9f841..3ca435063 100644 --- a/docs/api/utils.schemas.utils.html +++ b/docs/api/utils.schemas.utils.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/utils.tokenization.html b/docs/api/utils.tokenization.html index 41b7e2aad..2dbda523d 100644 --- a/docs/api/utils.tokenization.html +++ b/docs/api/utils.tokenization.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/api/utils.trainer.html b/docs/api/utils.trainer.html index 5c7bf728f..b08afd351 100644 --- a/docs/api/utils.trainer.html +++ b/docs/api/utils.trainer.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/attention.html b/docs/attention.html index 56d0083f3..9029e1c24 100644 --- a/docs/attention.html +++ b/docs/attention.html @@ -697,12 +697,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/batch_vs_grad.html b/docs/batch_vs_grad.html index 40c90fbc4..7822d1be5 100644 --- a/docs/batch_vs_grad.html +++ b/docs/batch_vs_grad.html @@ -662,12 +662,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/checkpoint_saving.html b/docs/checkpoint_saving.html index 205c5505a..8c5f23ce2 100644 --- a/docs/checkpoint_saving.html +++ b/docs/checkpoint_saving.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/choosing_method.html b/docs/choosing_method.html index 05b0e3bf2..22f2bf61a 100644 --- a/docs/choosing_method.html +++ b/docs/choosing_method.html @@ -697,12 +697,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/cli.html b/docs/cli.html index 54f283000..ba7912134 100644 --- a/docs/cli.html +++ b/docs/cli.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/config-reference.html b/docs/config-reference.html index 82f18d9ef..2c637fda7 100644 --- a/docs/config-reference.html +++ b/docs/config-reference.html @@ -697,12 +697,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • @@ -1817,637 +1811,630 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); # length generalization. scaling_softmax_bias: float | None -unsloth_cross_entropy_loss: bool | None -unsloth_lora_mlp: bool | None -unsloth_lora_qkv: bool | None -unsloth_lora_o: bool | None -unsloth_rms_norm: bool | None -unsloth_rope: bool | None - -# Apply custom LoRA autograd functions and activation function Triton kernels for speed -# and memory savings. See: https://docs.axolotl.ai/docs/lora_optims.html -lora_mlp_kernel: bool | None -# Apply custom LoRA autograd functions and activation function Triton kernels for speed -# and memory savings. See: https://docs.axolotl.ai/docs/lora_optims.html -lora_qkv_kernel: bool | None -# Apply custom LoRA autograd functions and activation function Triton kernels for speed -# and memory savings. See: https://docs.axolotl.ai/docs/lora_optims.html -lora_o_kernel: bool | None -# Apply custom LoRA autograd function for embedding layers. See: -# https://docs.axolotl.ai/docs/lora_optims.html -lora_embedding_kernel: bool | None - -# Whether to use chunked cross entropy loss for memory efficiency -chunked_cross_entropy: bool | None -# Number of chunks to use for chunked cross entropy loss -chunked_cross_entropy_num_chunks: int | None -# Enable Entropy-Aware Focal Training loss (EAFT) -use_eaft: bool | None -# Exponent for entropy weighting in EAFT (default: 1.0) -eaft_alpha: float | None = 1.0 -# Number of top logits for entropy approximation (default: 20) -eaft_k: int | None = 20 +# Apply custom LoRA autograd functions and activation function Triton kernels for speed +# and memory savings. See: https://docs.axolotl.ai/docs/lora_optims.html +lora_mlp_kernel: bool | None +# Apply custom LoRA autograd functions and activation function Triton kernels for speed +# and memory savings. See: https://docs.axolotl.ai/docs/lora_optims.html +lora_qkv_kernel: bool | None +# Apply custom LoRA autograd functions and activation function Triton kernels for speed +# and memory savings. See: https://docs.axolotl.ai/docs/lora_optims.html +lora_o_kernel: bool | None +# Apply custom LoRA autograd function for embedding layers. See: +# https://docs.axolotl.ai/docs/lora_optims.html +lora_embedding_kernel: bool | None + +# Whether to use chunked cross entropy loss for memory efficiency +chunked_cross_entropy: bool | None +# Number of chunks to use for chunked cross entropy loss +chunked_cross_entropy_num_chunks: int | None +# Enable Entropy-Aware Focal Training loss (EAFT) +use_eaft: bool | None +# Exponent for entropy weighting in EAFT (default: 1.0) +eaft_alpha: float | None = 1.0 +# Number of top logits for entropy approximation (default: 20) +eaft_k: int | None = 20 + +# Whether to use ALST tiled mlp for memory efficient long context +tiled_mlp: bool | None + +# Number of shards to use for ALST tiled mlp. If unset, it will be set based on +# seqlen/hidden_size +tiled_mlp_num_shards: int | None -# Whether to use ALST tiled mlp for memory efficient long context -tiled_mlp: bool | None - -# Number of shards to use for ALST tiled mlp. If unset, it will be set based on -# seqlen/hidden_size -tiled_mlp_num_shards: int | None - -# Whether to use original mlp for ALST tiled mlp. Otherwise uses a generic MLP based on -# llama. -tiled_mlp_use_original_mlp: bool | None = True - -llama4_linearized_experts: bool | None +# Whether to use original mlp for ALST tiled mlp. Otherwise uses a generic MLP based on +# llama. +tiled_mlp_use_original_mlp: bool | None = True + +llama4_linearized_experts: bool | None + +# Deepspeed config path. e.g., deepspeed_configs/zero3.json +deepspeed: str | dict[str, Any] | None +# Whether to use deepcompile for faster training with deepspeed +deepcompile: bool | None +# FSDP configuration +fsdp: list[str] | None -# Deepspeed config path. e.g., deepspeed_configs/zero3.json -deepspeed: str | dict[str, Any] | None -# Whether to use deepcompile for faster training with deepspeed -deepcompile: bool | None -# FSDP configuration -fsdp: list[str] | None - -# FSDP configuration options -fsdp_config: FSDPConfig | None - # For FSDPConfig: - # FSDP version - fsdp_version: int | None - # Enable activation checkpointing to reduce memory usage during forward passes - activation_checkpointing: bool | None - # Offload parameters to CPU to reduce GPU memory usage - offload_params: bool | None - # Synchronize module states across all processes - sync_module_states: bool | None - # Enable CPU RAM efficient loading to reduce memory usage during model loading - cpu_ram_efficient_loading: bool | None - # Disabling this enables swap memory usage for resource-constrained setups when - # offload_params is enabled. - cpu_offload_pin_memory: bool | None - # Use original parameters instead of flattened parameters - use_orig_params: bool | None - - # Type of state dict to use for saving/loading checkpoints - state_dict_type: Literal['FULL_STATE_DICT', 'LOCAL_STATE_DICT', 'SHARDED_STATE_DICT'] | None - # Final state dict type to use after training completion - final_state_dict_type: Literal['FULL_STATE_DICT', 'LOCAL_STATE_DICT', 'SHARDED_STATE_DICT'] | None - - # Policy for automatically wrapping modules with FSDP - auto_wrap_policy: Literal['TRANSFORMER_BASED_WRAP', 'SIZE_BASED_WRAP'] | None - # Class name of transformer layers to wrap (e.g., 'LlamaDecoderLayer') - transformer_layer_cls_to_wrap: str | None - - # Reshard parameters after forward pass to save memory - reshard_after_forward: bool | None - # Mixed precision policy for FSDP (e.g., 'fp16', 'bf16') - mixed_precision_policy: str | None - -# FSDP version -fsdp_version: int | None -fsdp_final_state_dict_type: Literal['FULL_STATE_DICT', 'LOCAL_STATE_DICT', 'SHARDED_STATE_DICT'] | None - -# How much of the dataset to set aside as evaluation. 1 = 100%, 0.50 = 50%, etc. 0 for -# no eval. -val_set_size: float | None = 0.0 - -# Number of devices to shard across. If not set, will use all available devices. -dp_shard_size: int | None -# Number of devices to replicate across. -dp_replicate_size: int | None -# Deprecated: use `context_parallel_size` instead -sequence_parallel_degree: int | None -# Set to a divisor of the number of GPUs available to split sequences into chunks of -# equal size. Use in long context training to prevent OOM when sequences cannot fit into -# a single GPU's VRAM. E.g., if 4 GPUs are available, set this value to 2 to split each -# sequence into two equal-sized subsequences, or set to 4 to split into four equal-sized -# subsequences. See https://docs.axolotl.ai/docs/sequence_parallelism.html for more -# details. -context_parallel_size: int | None -# Optional; strides across the key dimension. Larger values use more memory but should -# make training faster. Must evenly divide the number of KV heads in your model. -heads_k_stride: int | None -# One of 'varlen_llama3', 'batch_ring', 'batch_zigzag', 'batch_stripe'. Defaults to -# 'varlen_llama3' in the sample packing case, and 'batch_ring' in the non-sample packing -# case. -ring_attn_func: RingAttnFunc | None -# Number of tensor parallel processes in TP group. Only supported with DeepSpeed AutoTP. -tensor_parallel_size: int | None - -# Add or change special tokens. If you add tokens here, you don't need to add them to -# the `tokens` list. -special_tokens: SpecialTokensConfig | None - # For SpecialTokensConfig: - bos_token: str | None - eos_token: str | None - pad_token: str | None - unk_token: str | None - additional_special_tokens: list[str] | None +# FSDP configuration options +fsdp_config: FSDPConfig | None + # For FSDPConfig: + # FSDP version + fsdp_version: int | None + # Enable activation checkpointing to reduce memory usage during forward passes + activation_checkpointing: bool | None + # Offload parameters to CPU to reduce GPU memory usage + offload_params: bool | None + # Synchronize module states across all processes + sync_module_states: bool | None + # Enable CPU RAM efficient loading to reduce memory usage during model loading + cpu_ram_efficient_loading: bool | None + # Disabling this enables swap memory usage for resource-constrained setups when + # offload_params is enabled. + cpu_offload_pin_memory: bool | None + # Use original parameters instead of flattened parameters + use_orig_params: bool | None + + # Type of state dict to use for saving/loading checkpoints + state_dict_type: Literal['FULL_STATE_DICT', 'LOCAL_STATE_DICT', 'SHARDED_STATE_DICT'] | None + # Final state dict type to use after training completion + final_state_dict_type: Literal['FULL_STATE_DICT', 'LOCAL_STATE_DICT', 'SHARDED_STATE_DICT'] | None + + # Policy for automatically wrapping modules with FSDP + auto_wrap_policy: Literal['TRANSFORMER_BASED_WRAP', 'SIZE_BASED_WRAP'] | None + # Class name of transformer layers to wrap (e.g., 'LlamaDecoderLayer') + transformer_layer_cls_to_wrap: str | None + + # Reshard parameters after forward pass to save memory + reshard_after_forward: bool | None + # Mixed precision policy for FSDP (e.g., 'fp16', 'bf16') + mixed_precision_policy: str | None + +# FSDP version +fsdp_version: int | None +fsdp_final_state_dict_type: Literal['FULL_STATE_DICT', 'LOCAL_STATE_DICT', 'SHARDED_STATE_DICT'] | None + +# How much of the dataset to set aside as evaluation. 1 = 100%, 0.50 = 50%, etc. 0 for +# no eval. +val_set_size: float | None = 0.0 + +# Number of devices to shard across. If not set, will use all available devices. +dp_shard_size: int | None +# Number of devices to replicate across. +dp_replicate_size: int | None +# Deprecated: use `context_parallel_size` instead +sequence_parallel_degree: int | None +# Set to a divisor of the number of GPUs available to split sequences into chunks of +# equal size. Use in long context training to prevent OOM when sequences cannot fit into +# a single GPU's VRAM. E.g., if 4 GPUs are available, set this value to 2 to split each +# sequence into two equal-sized subsequences, or set to 4 to split into four equal-sized +# subsequences. See https://docs.axolotl.ai/docs/sequence_parallelism.html for more +# details. +context_parallel_size: int | None +# Optional; strides across the key dimension. Larger values use more memory but should +# make training faster. Must evenly divide the number of KV heads in your model. +heads_k_stride: int | None +# One of 'varlen_llama3', 'batch_ring', 'batch_zigzag', 'batch_stripe'. Defaults to +# 'varlen_llama3' in the sample packing case, and 'batch_ring' in the non-sample packing +# case. +ring_attn_func: RingAttnFunc | None +# Number of tensor parallel processes in TP group. Only supported with DeepSpeed AutoTP. +tensor_parallel_size: int | None + +# Add or change special tokens. If you add tokens here, you don't need to add them to +# the `tokens` list. +special_tokens: SpecialTokensConfig | None + # For SpecialTokensConfig: + bos_token: str | None + eos_token: str | None + pad_token: str | None + unk_token: str | None + additional_special_tokens: list[str] | None + +# Add extra tokens to the tokenizer +tokens: list[str] | None +# Mapping token_id to new_token_string to override reserved added_tokens in the +# tokenizer. Only works for tokens that are not part of the base vocab (aka are +# added_tokens). Can be checked if they exist in tokenizer.json added_tokens. +added_tokens_overrides: dict[int, str] | None -# Add extra tokens to the tokenizer -tokens: list[str] | None -# Mapping token_id to new_token_string to override reserved added_tokens in the -# tokenizer. Only works for tokens that are not part of the base vocab (aka are -# added_tokens). Can be checked if they exist in tokenizer.json added_tokens. -added_tokens_overrides: dict[int, str] | None +# Whether to use torch.compile and which backend to use. setting to `auto` will enable +# torch compile when torch>=2.6.0 +torch_compile: Literal['auto'] | bool | None +# Backend to use for torch.compile +torch_compile_backend: str | None +torch_compile_mode: Literal['default', 'reduce-overhead', 'max-autotune'] | None -# Whether to use torch.compile and which backend to use. setting to `auto` will enable -# torch compile when torch>=2.6.0 -torch_compile: Literal['auto'] | bool | None -# Backend to use for torch.compile -torch_compile_backend: str | None -torch_compile_mode: Literal['default', 'reduce-overhead', 'max-autotune'] | None - -# Maximum number of iterations to train for. It precedes num_epochs which means that if -# both are set, num_epochs will not be guaranteed. e.g., when 1 epoch is 1000 steps => -# `num_epochs: 2` and `max_steps: 100` will train for 100 steps -max_steps: int | None -# Number of warmup steps. Cannot use with warmup_ratio -warmup_steps: int | None -# Warmup ratio. Cannot use with warmup_steps -warmup_ratio: float | None -# Leave empty to eval at each epoch, integer for every N steps. float for fraction of -# total steps -eval_steps: int | float | None -# Number of times per epoch to run evals, mutually exclusive with eval_steps -evals_per_epoch: int | None -# Set to `no` to skip evaluation, `epoch` at end of each epoch, leave empty to infer -# from `eval_steps` -eval_strategy: str | None - -# Leave empty to save at each epoch, integer for every N steps. float for fraction of -# total steps -save_steps: int | float | None -# Number of times per epoch to save a checkpoint, mutually exclusive with save_steps -saves_per_epoch: int | None -# 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` -save_strategy: str | None -# Checkpoints saved at a time -save_total_limit: int | None -# Whether to checkpoint a model after the first step of training. Defaults to False. -save_first_step: bool | None - -# Logging frequency -logging_steps: int | None -# Stop training after this many evaluation losses have increased in a row. https://huggi -# ngface.co/transformers/v4.2.2/_modules/transformers/trainer_callback.html#EarlyStoppin -# gCallback -early_stopping_patience: int | None -load_best_model_at_end: bool | None = False -# Save only the model weights, skipping the optimizer. Using this means you can't resume -# from checkpoints. -save_only_model: bool | None = False -# Use tensorboard for logging -use_tensorboard: bool | None -# Enable the pytorch profiler to capture the first N steps of training to the -# output_dir. see https://pytorch.org/blog/understanding-gpu-memory-1/ for more -# information. Snapshots can be visualized @ https://pytorch.org/memory_viz -profiler_steps: int | None -# Which step to start the profiler at. Useful for only capturing a few steps mid-run. -profiler_steps_start: int | None = 0 -# bool of whether to report tokens per second at the end of training. This is not -# supported with pre-training datasets. -include_tokens_per_second: bool | None -# bool of whether to report tokens per second per-gpu during training by measuring -# throughput of non-padding tokens. -include_tkps: bool | None = True -# NEFT https://arxiv.org/abs/2310.05914, set this to a number (paper default is 5) to -# add noise to embeddings. Currently only supported on Llama and Mistral -neftune_noise_alpha: float | None - -# Parameter controlling the relative ratio loss weight in the ORPO loss. Passed to -# `beta` in `ORPOConfig` due to trl mapping. -orpo_alpha: float | None -# Target reward margin for the SimPO loss -simpo_gamma: float | None -# Weight of the BC regularizer -cpo_alpha: float | None +# Maximum number of iterations to train for. It precedes num_epochs which means that if +# both are set, num_epochs will not be guaranteed. e.g., when 1 epoch is 1000 steps => +# `num_epochs: 2` and `max_steps: 100` will train for 100 steps +max_steps: int | None +# Number of warmup steps. Cannot use with warmup_ratio +warmup_steps: int | None +# Warmup ratio. Cannot use with warmup_steps +warmup_ratio: float | None +# Leave empty to eval at each epoch, integer for every N steps. float for fraction of +# total steps +eval_steps: int | float | None +# Number of times per epoch to run evals, mutually exclusive with eval_steps +evals_per_epoch: int | None +# Set to `no` to skip evaluation, `epoch` at end of each epoch, leave empty to infer +# from `eval_steps` +eval_strategy: str | None + +# Leave empty to save at each epoch, integer for every N steps. float for fraction of +# total steps +save_steps: int | float | None +# Number of times per epoch to save a checkpoint, mutually exclusive with save_steps +saves_per_epoch: int | None +# 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` +save_strategy: str | None +# Checkpoints saved at a time +save_total_limit: int | None +# Whether to checkpoint a model after the first step of training. Defaults to False. +save_first_step: bool | None + +# Logging frequency +logging_steps: int | None +# Stop training after this many evaluation losses have increased in a row. https://huggi +# ngface.co/transformers/v4.2.2/_modules/transformers/trainer_callback.html#EarlyStoppin +# gCallback +early_stopping_patience: int | None +load_best_model_at_end: bool | None = False +# Save only the model weights, skipping the optimizer. Using this means you can't resume +# from checkpoints. +save_only_model: bool | None = False +# Use tensorboard for logging +use_tensorboard: bool | None +# Enable the pytorch profiler to capture the first N steps of training to the +# output_dir. see https://pytorch.org/blog/understanding-gpu-memory-1/ for more +# information. Snapshots can be visualized @ https://pytorch.org/memory_viz +profiler_steps: int | None +# Which step to start the profiler at. Useful for only capturing a few steps mid-run. +profiler_steps_start: int | None = 0 +# bool of whether to report tokens per second at the end of training. This is not +# supported with pre-training datasets. +include_tokens_per_second: bool | None +# bool of whether to report tokens per second per-gpu during training by measuring +# throughput of non-padding tokens. +include_tkps: bool | None = True +# NEFT https://arxiv.org/abs/2310.05914, set this to a number (paper default is 5) to +# add noise to embeddings. Currently only supported on Llama and Mistral +neftune_noise_alpha: float | None + +# Parameter controlling the relative ratio loss weight in the ORPO loss. Passed to +# `beta` in `ORPOConfig` due to trl mapping. +orpo_alpha: float | None +# Target reward margin for the SimPO loss +simpo_gamma: float | None +# Weight of the BC regularizer +cpo_alpha: float | None + +# Factor for desirable loss term in KTO loss +kto_desirable_weight: float | None +# Factor for undesirable loss term in KTO loss +kto_undesirable_weight: float | None +# The beta parameter for the RL training +rl_beta: float | None -# Factor for desirable loss term in KTO loss -kto_desirable_weight: float | None -# Factor for undesirable loss term in KTO loss -kto_undesirable_weight: float | None -# The beta parameter for the RL training -rl_beta: float | None - -# Defines the max memory usage per gpu on the system. Passed through to transformers -# when loading the model. -max_memory: dict[int | Literal['cpu', 'disk'], int | str] | None -# Limit the memory for all available GPUs to this amount (if an integer, expressed in -# gigabytes); default: unset -gpu_memory_limit: int | str | None -# Whether to use low_cpu_mem_usage -low_cpu_mem_usage: bool | None - -# The name of the chat template to use for training, following values are supported: -# tokenizer_default: Uses the chat template that is available in the -# tokenizer_config.json. If the chat template is not available in the tokenizer, it will -# raise an error. This is the default value. -# alpaca/inst/chatml/gemma/cohere/llama3/phi_3/deepseek_v2/jamba: These chat templates -# are available in the axolotl codebase at src/axolotl/utils/chat_templates.py. -# tokenizer_default_fallback_*: where * is the name of the chat template to fallback to. -# E.g. tokenizer_default_fallback_chatml. This is useful when the chat template is not -# available in the tokenizer. jinja: Uses a custom jinja template for the chat template. -# The custom jinja template should be provided in the chat_template_jinja field. The -# selected chat template will be saved to the tokenizer_config.json for easier -# inferencing -chat_template: ChatTemplate | Annotated[str, StringConstraints(pattern='^tokenizer_default_fallback_')] | None -# Custom jinja template or path to jinja file for chat template. This will be only used -# if chat_template is set to `jinja` or `null` (in which case chat_template is -# automatically set to `jinja`). Default is null. -chat_template_jinja: str | None -# Additional kwargs to pass to the chat template. This is useful for customizing the -# chat template. For example, you can pass `thinking=False` to add a generation prompt -# to the chat template. -chat_template_kwargs: dict[str, Any] | None -# Custom EOT (End-of-Turn) tokens to mask/unmask during training. These tokens mark the -# boundaries between conversation turns. For example: ['/INST', '</s>', -# '[/SYSTEM_PROMPT]']. If not specified, defaults to just the model's eos_token. This is -# useful for templates that use multiple delimiter tokens. -eot_tokens: list[str] | None -# Changes the default system message. Currently only supports chatml. -default_system_message: str | None +# Defines the max memory usage per gpu on the system. Passed through to transformers +# when loading the model. +max_memory: dict[int | Literal['cpu', 'disk'], int | str] | None +# Limit the memory for all available GPUs to this amount (if an integer, expressed in +# gigabytes); default: unset +gpu_memory_limit: int | str | None +# Whether to use low_cpu_mem_usage +low_cpu_mem_usage: bool | None + +# The name of the chat template to use for training, following values are supported: +# tokenizer_default: Uses the chat template that is available in the +# tokenizer_config.json. If the chat template is not available in the tokenizer, it will +# raise an error. This is the default value. +# alpaca/inst/chatml/gemma/cohere/llama3/phi_3/deepseek_v2/jamba: These chat templates +# are available in the axolotl codebase at src/axolotl/utils/chat_templates.py. +# tokenizer_default_fallback_*: where * is the name of the chat template to fallback to. +# E.g. tokenizer_default_fallback_chatml. This is useful when the chat template is not +# available in the tokenizer. jinja: Uses a custom jinja template for the chat template. +# The custom jinja template should be provided in the chat_template_jinja field. The +# selected chat template will be saved to the tokenizer_config.json for easier +# inferencing +chat_template: ChatTemplate | Annotated[str, StringConstraints(pattern='^tokenizer_default_fallback_')] | None +# Custom jinja template or path to jinja file for chat template. This will be only used +# if chat_template is set to `jinja` or `null` (in which case chat_template is +# automatically set to `jinja`). Default is null. +chat_template_jinja: str | None +# Additional kwargs to pass to the chat template. This is useful for customizing the +# chat template. For example, you can pass `thinking=False` to add a generation prompt +# to the chat template. +chat_template_kwargs: dict[str, Any] | None +# Custom EOT (End-of-Turn) tokens to mask/unmask during training. These tokens mark the +# boundaries between conversation turns. For example: ['/INST', '</s>', +# '[/SYSTEM_PROMPT]']. If not specified, defaults to just the model's eos_token. This is +# useful for templates that use multiple delimiter tokens. +eot_tokens: list[str] | None +# Changes the default system message. Currently only supports chatml. +default_system_message: str | None + +# Token index or indices to adjust embedding weights to the mean of the other tokens. +# This is useful when the model has untrained embeddings. +fix_untrained_tokens: int | list[int] | None + +is_preprocess: bool | None +preprocess_iterable: bool | None -# Token index or indices to adjust embedding weights to the mean of the other tokens. -# This is useful when the model has untrained embeddings. -fix_untrained_tokens: int | list[int] | None - -is_preprocess: bool | None -preprocess_iterable: bool | None - -# Total number of tokens - internal use -total_num_tokens: int | None -total_supervised_tokens: int | None -# You can set these packing optimizations AFTER starting a training at least once. The -# trainer will provide recommended values for these values. -sample_packing_eff_est: float | None -axolotl_config_path: str | None - +# Total number of tokens - internal use +total_num_tokens: int | None +total_supervised_tokens: int | None +# You can set these packing optimizations AFTER starting a training at least once. The +# trainer will provide recommended values for these values. +sample_packing_eff_est: float | None +axolotl_config_path: str | None + +# Internal use only - Used to identify which the model is based on +is_falcon_derived_model: bool | None +# Internal use only - Used to identify which the model is based on +is_llama_derived_model: bool | None +# Internal use only - Used to identify which the model is based on. Please note that if +# you set this to true, `padding_side` will be set to 'left' by default +is_mistral_derived_model: bool | None # Internal use only - Used to identify which the model is based on -is_falcon_derived_model: bool | None -# Internal use only - Used to identify which the model is based on -is_llama_derived_model: bool | None -# Internal use only - Used to identify which the model is based on. Please note that if -# you set this to true, `padding_side` will be set to 'left' by default -is_mistral_derived_model: bool | None -# Internal use only - Used to identify which the model is based on -is_qwen_derived_model: bool | None - -# Add plugins to extend the pipeline. See `src/axolotl/integrations` for the available -# plugins or doc below for more details. -# https://docs.axolotl.ai/docs/custom_integrations.html -plugins: list[str] | None -# Enable sample generation during training for monitoring -generate_samples: bool | None = False -# Number of samples to generate at each interval -num_generation_samples: int | None = 3 -# Maximum new tokens to generate per sample -generation_max_new_tokens: int | None = 50 -# Temperature for sample generation (0.0 = greedy) -generation_temperature: float | None = 0.7 -# Nucleus sampling parameter for generation -generation_top_p: float | None -# Top-k sampling parameter for generation -generation_top_k: int | None -# Ratio of input to use as prompt (0.0-1.0) -generation_prompt_ratio: float | None = 0.5 -# Whether to use sampling (vs greedy decoding) -generation_do_sample: bool | None = True - -# This is the huggingface model that contains *.pt, *.safetensors, or *.bin files. This -# can also be a relative path to a model on disk -base_model: str (required) -# If the base_model repo on hf hub doesn't include configuration .json files, You can -# set that here, or leave this empty to default to base_model -base_model_config: str | None -# transformers config class (e.g., 'LlamaConfig', 'MistralConfig'). Defaults to -# AutoConfig. -cls_model_config: str | None -# Optional tokenizer configuration path in case you want to use a different tokenizer -# than the one defined in the base model -tokenizer_config: str | None -# use_fast option for tokenizer loading from_pretrained, default to True -tokenizer_use_fast: bool | None -# Whether to use the legacy tokenizer setting, defaults to True -tokenizer_legacy: bool | None -# Whether to use mistral-common tokenizer. If set to True, it will use the mistral- -# common tokenizer. -tokenizer_use_mistral_common: bool | None -# Corresponding tokenizer for the model AutoTokenizer is a good choice -tokenizer_type: str | None -# transformers processor class -processor_type: str | None -# Whether to save jinja files for tokenizer, transformers default is True -tokenizer_save_jinja_files: bool | None = True -# Trust remote code for untrusted source -trust_remote_code: bool | None +is_qwen_derived_model: bool | None + +# Add plugins to extend the pipeline. See `src/axolotl/integrations` for the available +# plugins or doc below for more details. +# https://docs.axolotl.ai/docs/custom_integrations.html +plugins: list[str] | None +# Enable sample generation during training for monitoring +generate_samples: bool | None = False +# Number of samples to generate at each interval +num_generation_samples: int | None = 3 +# Maximum new tokens to generate per sample +generation_max_new_tokens: int | None = 50 +# Temperature for sample generation (0.0 = greedy) +generation_temperature: float | None = 0.7 +# Nucleus sampling parameter for generation +generation_top_p: float | None +# Top-k sampling parameter for generation +generation_top_k: int | None +# Ratio of input to use as prompt (0.0-1.0) +generation_prompt_ratio: float | None = 0.5 +# Whether to use sampling (vs greedy decoding) +generation_do_sample: bool | None = True + +# This is the huggingface model that contains *.pt, *.safetensors, or *.bin files. This +# can also be a relative path to a model on disk +base_model: str (required) +# If the base_model repo on hf hub doesn't include configuration .json files, You can +# set that here, or leave this empty to default to base_model +base_model_config: str | None +# transformers config class (e.g., 'LlamaConfig', 'MistralConfig'). Defaults to +# AutoConfig. +cls_model_config: str | None +# Optional tokenizer configuration path in case you want to use a different tokenizer +# than the one defined in the base model +tokenizer_config: str | None +# use_fast option for tokenizer loading from_pretrained, default to True +tokenizer_use_fast: bool | None +# Whether to use the legacy tokenizer setting, defaults to True +tokenizer_legacy: bool | None +# Whether to use mistral-common tokenizer. If set to True, it will use the mistral- +# common tokenizer. +tokenizer_use_mistral_common: bool | None +# Corresponding tokenizer for the model AutoTokenizer is a good choice +tokenizer_type: str | None +# transformers processor class +processor_type: str | None +# Whether to save jinja files for tokenizer, transformers default is True +tokenizer_save_jinja_files: bool | None = True +# Trust remote code for untrusted source +trust_remote_code: bool | None + +# Don't move the model to the device before sharding. Set to `false` to revert to legacy +# behavior. +experimental_skip_move_to_device: bool | None = True + +# Use custom kernels, e.g. MegaBlocks. +use_kernels: bool | None -# Don't move the model to the device before sharding. Set to `false` to revert to legacy -# behavior. -experimental_skip_move_to_device: bool | None = True - -# Use custom kernels, e.g. MegaBlocks. -use_kernels: bool | None - -# Model loading quantization config -model_quantization_config: Literal['Mxfp4Config', 'FineGrainedFP8Config'] | None -# kwargs for model quantization config -model_quantization_config_kwargs: dict[str, Any] | None - -# Where to save the full-finetuned model to -output_dir: str = ./model-out -# push checkpoints to hub -hub_model_id: str | None -# how to push checkpoints to hub -hub_strategy: str | None -# branch/revision to push to on hub (default: main) -hub_revision: str | None -# Whether to save the model using safetensors format. Defaults to True. -save_safetensors: bool | None = True - -# This will attempt to quantize the model down to 8 bits and use adam 8 bit optimizer -load_in_8bit: bool | None = False -# Use bitsandbytes 4 bit -load_in_4bit: bool | None = False - -# If you want to use 'lora', 'qlora', or 'llama-adapter', or leave blank to train all -# parameters in original model -adapter: Literal['lora', 'qlora', 'llama-adapter'] | None -# If you already have a lora model trained that you want to load, put that here. This -# means after training, if you want to test the model, you should set this to the value -# of `output_dir`. Note that if you merge an adapter to the base model, a new -# subdirectory `merged` will be created under the `output_dir`. -lora_model_dir: str | None -lora_r: int | None -lora_alpha: int | None -lora_fan_in_fan_out: bool | None -lora_target_modules: str | list[str] | None -lora_target_parameters: str | list[str] | None -# If true, will target all linear modules -lora_target_linear: bool | None -# If you added new tokens to the tokenizer, you may need to save some LoRA modules -# because they need to know the new tokens. For LLaMA and Mistral, you need to save -# `embed_tokens` and `lm_head`. It may vary for other models. `embed_tokens` converts -# tokens to embeddings, and `lm_head` converts embeddings to token probabilities. -lora_modules_to_save: list[str] | None -lora_dropout: float | None = 0.0 -# The layer indices to transform, otherwise, apply to all layers -peft_layers_to_transform: list[int] | None -peft_layers_pattern: list[str] | None - -peft: PeftConfig | None - # For PeftConfig: - # Configuration options for loftq initialization for LoRA - loftq_config: LoftQConfig | None - # For LoftQConfig: - # typically 4 bits - loftq_bits: int = 4 - -# Whether to use DoRA. -peft_use_dora: bool | None -# Whether to use RSLoRA. -peft_use_rslora: bool | None -# List of layer indices to replicate. -peft_layer_replication: list[tuple[int, int]] | None -# How to initialize LoRA weights. Default to True which is MS original implementation. -peft_init_lora_weights: bool | str | None -# A list of token indices to fine-tune on the `embed_tokens` layer. Otherwise, a dict -# mapping an embedding layer name to its trainable token indices. See -# https://huggingface.co/docs/peft/v0.17.0/en/developer_guides/lora#efficiently-train- -# tokens-alongside-lora -peft_trainable_token_indices: list[int] | dict[str, list[int]] | None -# Whether to tie adapter weights for tied model weights. See -# https://github.com/huggingface/peft/issues/2864 -peft_ensure_weight_tying: bool | None -# Whether to upcast the LoRA adapter to fp32. This is enabled by default in PEFT. -peft_autocast_adapter_dtype: bool | None - -# load qlora model in sharded format for FSDP using answer.ai technique. -qlora_sharded_model_loading: bool | None = False -# Do the LoRA/PEFT loading on CPU -- this is required if the base model is so large it -# takes up most or all of the available GPU VRAM, e.g. during a model and LoRA merge -lora_on_cpu: bool | None -# Whether you are training a 4-bit GPTQ quantized model -gptq: bool | None -# optional overrides to the bnb 4bit quantization configuration -bnb_config_kwargs: dict[str, Any] | None - -# loraplus learning rate ratio lr_B / lr_A. Recommended value is 2^4. -loraplus_lr_ratio: float | None -# loraplus learning rate for lora embedding layers. Default value is 1e-6. -loraplus_lr_embedding: float | None = 1e-06 - -merge_lora: bool | None -# Method to use for LoRA merging. 'memory_efficient' (default) processes shards -# individually to reduce memory usage, 'legacy' loads the full model into memory. -merge_method: Literal['legacy', 'memory_efficient'] | None = memory_efficient - -# Whether to use ReLoRA. Use with jagged_restart_*steps options. -relora: bool | None -# threshold for optimizer magnitude when pruning -relora_prune_ratio: float | None -# True to perform lora weight merges on cpu during restarts, for modest gpu memory -# savings -relora_cpu_offload: bool | None +# Model loading quantization config +model_quantization_config: Literal['Mxfp4Config', 'FineGrainedFP8Config'] | None +# kwargs for model quantization config +model_quantization_config_kwargs: dict[str, Any] | None + +# Where to save the full-finetuned model to +output_dir: str = ./model-out +# push checkpoints to hub +hub_model_id: str | None +# how to push checkpoints to hub +hub_strategy: str | None +# branch/revision to push to on hub (default: main) +hub_revision: str | None +# Whether to save the model using safetensors format. Defaults to True. +save_safetensors: bool | None = True + +# This will attempt to quantize the model down to 8 bits and use adam 8 bit optimizer +load_in_8bit: bool | None = False +# Use bitsandbytes 4 bit +load_in_4bit: bool | None = False + +# If you want to use 'lora', 'qlora', or 'llama-adapter', or leave blank to train all +# parameters in original model +adapter: Literal['lora', 'qlora', 'llama-adapter'] | None +# If you already have a lora model trained that you want to load, put that here. This +# means after training, if you want to test the model, you should set this to the value +# of `output_dir`. Note that if you merge an adapter to the base model, a new +# subdirectory `merged` will be created under the `output_dir`. +lora_model_dir: str | None +lora_r: int | None +lora_alpha: int | None +lora_fan_in_fan_out: bool | None +lora_target_modules: str | list[str] | None +lora_target_parameters: str | list[str] | None +# If true, will target all linear modules +lora_target_linear: bool | None +# If you added new tokens to the tokenizer, you may need to save some LoRA modules +# because they need to know the new tokens. For LLaMA and Mistral, you need to save +# `embed_tokens` and `lm_head`. It may vary for other models. `embed_tokens` converts +# tokens to embeddings, and `lm_head` converts embeddings to token probabilities. +lora_modules_to_save: list[str] | None +lora_dropout: float | None = 0.0 +# The layer indices to transform, otherwise, apply to all layers +peft_layers_to_transform: list[int] | None +peft_layers_pattern: list[str] | None + +peft: PeftConfig | None + # For PeftConfig: + # Configuration options for loftq initialization for LoRA + loftq_config: LoftQConfig | None + # For LoftQConfig: + # typically 4 bits + loftq_bits: int = 4 + +# Whether to use DoRA. +peft_use_dora: bool | None +# Whether to use RSLoRA. +peft_use_rslora: bool | None +# List of layer indices to replicate. +peft_layer_replication: list[tuple[int, int]] | None +# How to initialize LoRA weights. Default to True which is MS original implementation. +peft_init_lora_weights: bool | str | None +# A list of token indices to fine-tune on the `embed_tokens` layer. Otherwise, a dict +# mapping an embedding layer name to its trainable token indices. See +# https://huggingface.co/docs/peft/v0.17.0/en/developer_guides/lora#efficiently-train- +# tokens-alongside-lora +peft_trainable_token_indices: list[int] | dict[str, list[int]] | None +# Whether to tie adapter weights for tied model weights. See +# https://github.com/huggingface/peft/issues/2864 +peft_ensure_weight_tying: bool | None +# Whether to upcast the LoRA adapter to fp32. This is enabled by default in PEFT. +peft_autocast_adapter_dtype: bool | None + +# load qlora model in sharded format for FSDP using answer.ai technique. +qlora_sharded_model_loading: bool | None = False +# Do the LoRA/PEFT loading on CPU -- this is required if the base model is so large it +# takes up most or all of the available GPU VRAM, e.g. during a model and LoRA merge +lora_on_cpu: bool | None +# Whether you are training a 4-bit GPTQ quantized model +gptq: bool | None +# optional overrides to the bnb 4bit quantization configuration +bnb_config_kwargs: dict[str, Any] | None + +# loraplus learning rate ratio lr_B / lr_A. Recommended value is 2^4. +loraplus_lr_ratio: float | None +# loraplus learning rate for lora embedding layers. Default value is 1e-6. +loraplus_lr_embedding: float | None = 1e-06 + +merge_lora: bool | None +# Method to use for LoRA merging. 'memory_efficient' (default) processes shards +# individually to reduce memory usage, 'legacy' loads the full model into memory. +merge_method: Literal['legacy', 'memory_efficient'] | None = memory_efficient + +# Whether to use ReLoRA. Use with jagged_restart_*steps options. +relora: bool | None +# threshold for optimizer magnitude when pruning +relora_prune_ratio: float | None +# True to perform lora weight merges on cpu during restarts, for modest gpu memory +# savings +relora_cpu_offload: bool | None + +# how often to reset for jagged restarts +jagged_restart_steps: int | None +# how many warmup steps to take after reset for jagged restarts +jagged_restart_warmup_steps: int | None +# how many anneal steps to take before reset for jagged restarts +jagged_restart_anneal_steps: int | None -# how often to reset for jagged restarts -jagged_restart_steps: int | None -# how many warmup steps to take after reset for jagged restarts -jagged_restart_warmup_steps: int | None -# how many anneal steps to take before reset for jagged restarts -jagged_restart_anneal_steps: int | None - -# If greater than 1, backpropagation will be skipped and the gradients will be -# accumulated for the given number of steps. -gradient_accumulation_steps: int | None = 1 -# The number of samples to include in each batch. This is the number of samples sent to -# each GPU. Batch size per gpu = micro_batch_size * gradient_accumulation_steps -micro_batch_size: int | None = 1 -# Total batch size, we do not recommended setting this manually -batch_size: int | None -# per gpu micro batch size for evals, defaults to value of micro_batch_size -eval_batch_size: int | None - -# whether to find batch size that fits in memory. Passed to underlying transformers -# Trainer -auto_find_batch_size: bool | None +# If greater than 1, backpropagation will be skipped and the gradients will be +# accumulated for the given number of steps. +gradient_accumulation_steps: int | None = 1 +# The number of samples to include in each batch. This is the number of samples sent to +# each GPU. Batch size per gpu = micro_batch_size * gradient_accumulation_steps +micro_batch_size: int | None = 1 +# Total batch size, we do not recommended setting this manually +batch_size: int | None +# per gpu micro batch size for evals, defaults to value of micro_batch_size +eval_batch_size: int | None + +# whether to find batch size that fits in memory. Passed to underlying transformers +# Trainer +auto_find_batch_size: bool | None + +# Whether to mask out or include the human's prompt from the training labels +train_on_inputs: bool | None = False +# Group similarly sized data to minimize padding. May be slower to start, as it must +# download and sort the entire dataset. Note that training loss may have an oscillating +# pattern with this enabled. +group_by_length: bool | None -# Whether to mask out or include the human's prompt from the training labels -train_on_inputs: bool | None = False -# Group similarly sized data to minimize padding. May be slower to start, as it must -# download and sort the entire dataset. Note that training loss may have an oscillating -# pattern with this enabled. -group_by_length: bool | None - -learning_rate: str | float (required) -embedding_lr: float | None -embedding_lr_scale: float | None -# Specify weight decay -weight_decay: float | None = 0.0 -# Specify optimizer -optimizer: OptimizerNames | CustomSupportedOptimizers | None = OptimizerNames.ADAMW_TORCH_FUSED -# Dictionary of arguments to pass to the optimizer -optim_args: str | dict[str, Any] | None -# 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 -optim_target_modules: list[str] | Literal['all_linear'] | None -# Path to torch distx for optim 'adamw_anyprecision' -torchdistx_path: str | None -lr_scheduler: SchedulerType | Literal['one_cycle'] | Literal['rex'] | None = SchedulerType.COSINE -# Specify a scheduler and kwargs to use with the optimizer -lr_scheduler_kwargs: dict[str, Any] | None -lr_quadratic_warmup: bool | None -# decay lr to some percentage of the peak lr, e.g. cosine_min_lr_ratio=0.1 for 10% of -# peak lr -cosine_min_lr_ratio: float | None -# 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 -cosine_constant_lr_ratio: float | None -# Learning rate div factor -lr_div_factor: float | None - -lr_groups: list[LrGroup] | None - # For LrGroup: - name: str (required) - modules: list[str] (required) - lr: float (required) - -# adamw hyperparams -adam_epsilon: float | None -# only used for CAME Optimizer -adam_epsilon2: float | None -# adamw hyperparams -adam_beta1: float | None -# adamw hyperparams -adam_beta2: float | None -# only used for CAME Optimizer -adam_beta3: float | None - -# Dion Optimizer learning rate -dion_lr: float | None -# Dion Optimizer momentum -dion_momentum: float | None -# Dion Optimizer: r/d fraction for low-rank approximation. Used to compute the low-rank -# dimension. -dion_rank_fraction: float | None = 1.0 -# Dion Optimizer: Round up the low-rank dimension to a multiple of this number. This may -# be useful to ensure even sharding. -dion_rank_multiple_of: int | None = 1 - -# Gradient clipping max norm -max_grad_norm: float | None -num_epochs: float = 1.0 - -use_wandb: bool | None -# Set the name of your wandb run -wandb_name: str | None -# Set the ID of your wandb run -wandb_run_id: str | None -# "offline" to save run metadata locally and not sync to the server, "disabled" to turn -# off wandb -wandb_mode: str | None -# Your wandb project name -wandb_project: str | None -# A wandb Team name if using a Team -wandb_entity: str | None -wandb_watch: str | None -# "checkpoint" to log model to wandb Artifacts every `save_steps` or "end" to log only -# at the end of training -wandb_log_model: str | None - -use_mlflow: bool | None -# URI to mlflow -mlflow_tracking_uri: str | None -# Your experiment name -mlflow_experiment_name: str | None -# Your run name -mlflow_run_name: str | None -# set to true to copy each saved checkpoint on each save to mlflow artifact registry -hf_mlflow_log_artifacts: bool | None - -# Enable or disable Comet integration. -use_comet: bool | None -# API key for Comet. Recommended to set via `comet login`. -comet_api_key: str | None -# Workspace name in Comet. Defaults to the user's default workspace. -comet_workspace: str | None -# Project name in Comet. Defaults to Uncategorized. -comet_project_name: str | None -# Identifier for the experiment. Used to append data to an existing experiment or -# control the key of new experiments. Default to a random key. -comet_experiment_key: str | None -# Create a new experiment ("create") or log to an existing one ("get"). Default -# ("get_or_create") auto-selects based on configuration. -comet_mode: str | None -# Set to True to log data to Comet server, or False for offline storage. Default is -# True. -comet_online: bool | None -# Dictionary for additional configuration settings, see the doc for more details. -comet_experiment_config: dict[str, Any] | None - -use_trackio: bool | None -# Your trackio project name -trackio_project_name: str | None -# Set the name of your trackio run -trackio_run_name: str | None -# Hugging Face Space ID to sync dashboard to (optional, runs locally if not provided) -trackio_space_id: str | None +learning_rate: str | float (required) +embedding_lr: float | None +embedding_lr_scale: float | None +# Specify weight decay +weight_decay: float | None = 0.0 +# Specify optimizer +optimizer: OptimizerNames | CustomSupportedOptimizers | None = OptimizerNames.ADAMW_TORCH_FUSED +# Dictionary of arguments to pass to the optimizer +optim_args: str | dict[str, Any] | None +# 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 +optim_target_modules: list[str] | Literal['all_linear'] | None +# Path to torch distx for optim 'adamw_anyprecision' +torchdistx_path: str | None +lr_scheduler: SchedulerType | Literal['one_cycle'] | Literal['rex'] | None = SchedulerType.COSINE +# Specify a scheduler and kwargs to use with the optimizer +lr_scheduler_kwargs: dict[str, Any] | None +lr_quadratic_warmup: bool | None +# decay lr to some percentage of the peak lr, e.g. cosine_min_lr_ratio=0.1 for 10% of +# peak lr +cosine_min_lr_ratio: float | None +# 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 +cosine_constant_lr_ratio: float | None +# Learning rate div factor +lr_div_factor: float | None + +lr_groups: list[LrGroup] | None + # For LrGroup: + name: str (required) + modules: list[str] (required) + lr: float (required) + +# adamw hyperparams +adam_epsilon: float | None +# only used for CAME Optimizer +adam_epsilon2: float | None +# adamw hyperparams +adam_beta1: float | None +# adamw hyperparams +adam_beta2: float | None +# only used for CAME Optimizer +adam_beta3: float | None + +# Dion Optimizer learning rate +dion_lr: float | None +# Dion Optimizer momentum +dion_momentum: float | None +# Dion Optimizer: r/d fraction for low-rank approximation. Used to compute the low-rank +# dimension. +dion_rank_fraction: float | None = 1.0 +# Dion Optimizer: Round up the low-rank dimension to a multiple of this number. This may +# be useful to ensure even sharding. +dion_rank_multiple_of: int | None = 1 + +# Gradient clipping max norm +max_grad_norm: float | None +num_epochs: float = 1.0 + +use_wandb: bool | None +# Set the name of your wandb run +wandb_name: str | None +# Set the ID of your wandb run +wandb_run_id: str | None +# "offline" to save run metadata locally and not sync to the server, "disabled" to turn +# off wandb +wandb_mode: str | None +# Your wandb project name +wandb_project: str | None +# A wandb Team name if using a Team +wandb_entity: str | None +wandb_watch: str | None +# "checkpoint" to log model to wandb Artifacts every `save_steps` or "end" to log only +# at the end of training +wandb_log_model: str | None + +use_mlflow: bool | None +# URI to mlflow +mlflow_tracking_uri: str | None +# Your experiment name +mlflow_experiment_name: str | None +# Your run name +mlflow_run_name: str | None +# set to true to copy each saved checkpoint on each save to mlflow artifact registry +hf_mlflow_log_artifacts: bool | None + +# Enable or disable Comet integration. +use_comet: bool | None +# API key for Comet. Recommended to set via `comet login`. +comet_api_key: str | None +# Workspace name in Comet. Defaults to the user's default workspace. +comet_workspace: str | None +# Project name in Comet. Defaults to Uncategorized. +comet_project_name: str | None +# Identifier for the experiment. Used to append data to an existing experiment or +# control the key of new experiments. Default to a random key. +comet_experiment_key: str | None +# Create a new experiment ("create") or log to an existing one ("get"). Default +# ("get_or_create") auto-selects based on configuration. +comet_mode: str | None +# Set to True to log data to Comet server, or False for offline storage. Default is +# True. +comet_online: bool | None +# Dictionary for additional configuration settings, see the doc for more details. +comet_experiment_config: dict[str, Any] | None + +use_trackio: bool | None +# Your trackio project name +trackio_project_name: str | None +# Set the name of your trackio run +trackio_run_name: str | None +# Hugging Face Space ID to sync dashboard to (optional, runs locally if not provided) +trackio_space_id: str | None + +# Enable OpenTelemetry metrics collection and Prometheus export +use_otel_metrics: bool | None = False +# Host to bind the OpenTelemetry metrics server to +otel_metrics_host: str | None = localhost +# Port for the Prometheus metrics HTTP server +otel_metrics_port: int | None = 8000 -# Enable OpenTelemetry metrics collection and Prometheus export -use_otel_metrics: bool | None = False -# Host to bind the OpenTelemetry metrics server to -otel_metrics_host: str | None = localhost -# Port for the Prometheus metrics HTTP server -otel_metrics_port: int | None = 8000 +# the number of activate layers in LISA +lisa_n_layers: int | None +# how often to switch layers in LISA +lisa_step_interval: int | None +# path under the model to access the layers +lisa_layers_attribute: str | None = model.layers -# the number of activate layers in LISA -lisa_n_layers: int | None -# how often to switch layers in LISA -lisa_step_interval: int | None -# path under the model to access the layers -lisa_layers_attribute: str | None = model.layers +gradio_title: str | None +gradio_share: bool | None +gradio_server_name: str | None +gradio_server_port: int | None +gradio_max_new_tokens: int | None +gradio_temperature: float | None -gradio_title: str | None -gradio_share: bool | None -gradio_server_name: str | None -gradio_server_port: int | None -gradio_max_new_tokens: int | None -gradio_temperature: float | None - -use_ray: bool = False -ray_run_name: str | None -ray_num_workers: int = 1 -resources_per_worker: dict - -# The size of the image to resize to. It can be an integer (resized into padded-square -# image) or a tuple (width, height).If not provided, we will attempt to load from -# preprocessor.size, otherwise, images won't be resized. -image_size: int | tuple[int, int] | None -# The resampling algorithm to use for image resizing. Default is bilinear. Please refer -# to PIL.Image.Resampling for more details. -image_resize_algorithm: Literal['bilinear', 'bicubic', 'lanczos'] | Resampling | None - -# optional overrides to the base model configuration -overrides_of_model_config: dict[str, Any] | None -# optional overrides the base model loading from_pretrained -overrides_of_model_kwargs: dict[str, Any] | None -# If you want to specify the type of model to load, AutoModelForCausalLM is a good -# choice too -type_of_model: str | None -# You can specify to choose a specific model revision from huggingface hub -revision_of_model: str | None - -max_packed_sequence_len: int | None -rope_scaling: Any | None -noisy_embedding_alpha: float | None -dpo_beta: float | None -evaluation_strategy: str | None -eval_table_size: int | None -eval_max_new_tokens: int | None -dpo_use_logits_to_keep: bool | None -dpo_generate_during_eval: bool | None -dpo_norm_loss: bool | None -rpo_alpha: float | None
    +use_ray: bool = False +ray_run_name: str | None +ray_num_workers: int = 1 +resources_per_worker: dict + +# The size of the image to resize to. It can be an integer (resized into padded-square +# image) or a tuple (width, height).If not provided, we will attempt to load from +# preprocessor.size, otherwise, images won't be resized. +image_size: int | tuple[int, int] | None +# The resampling algorithm to use for image resizing. Default is bilinear. Please refer +# to PIL.Image.Resampling for more details. +image_resize_algorithm: Literal['bilinear', 'bicubic', 'lanczos'] | Resampling | None + +# optional overrides to the base model configuration +overrides_of_model_config: dict[str, Any] | None +# optional overrides the base model loading from_pretrained +overrides_of_model_kwargs: dict[str, Any] | None +# If you want to specify the type of model to load, AutoModelForCausalLM is a good +# choice too +type_of_model: str | None +# You can specify to choose a specific model revision from huggingface hub +revision_of_model: str | None + +max_packed_sequence_len: int | None +rope_scaling: Any | None +noisy_embedding_alpha: float | None +dpo_beta: float | None +evaluation_strategy: str | None +eval_table_size: int | None +eval_max_new_tokens: int | None +dpo_use_logits_to_keep: bool | None +dpo_generate_during_eval: bool | None +dpo_norm_loss: bool | None +rpo_alpha: float | None diff --git a/docs/custom_integrations.html b/docs/custom_integrations.html index 231b0a13f..ca3efc1b9 100644 --- a/docs/custom_integrations.html +++ b/docs/custom_integrations.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/dataset-formats/conversation.html b/docs/dataset-formats/conversation.html index c36f5f978..0ec4f3547 100644 --- a/docs/dataset-formats/conversation.html +++ b/docs/dataset-formats/conversation.html @@ -697,12 +697,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/dataset-formats/index.html b/docs/dataset-formats/index.html index 6d6778ced..6c8f86e2d 100644 --- a/docs/dataset-formats/index.html +++ b/docs/dataset-formats/index.html @@ -697,12 +697,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/dataset-formats/inst_tune.html b/docs/dataset-formats/inst_tune.html index ac2b8680b..1f6ef125c 100644 --- a/docs/dataset-formats/inst_tune.html +++ b/docs/dataset-formats/inst_tune.html @@ -697,12 +697,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/dataset-formats/pretraining.html b/docs/dataset-formats/pretraining.html index bd56394a5..4ee44dee9 100644 --- a/docs/dataset-formats/pretraining.html +++ b/docs/dataset-formats/pretraining.html @@ -662,12 +662,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/dataset-formats/stepwise_supervised.html b/docs/dataset-formats/stepwise_supervised.html index 6ed98af11..e1bec5050 100644 --- a/docs/dataset-formats/stepwise_supervised.html +++ b/docs/dataset-formats/stepwise_supervised.html @@ -697,12 +697,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/dataset-formats/template_free.html b/docs/dataset-formats/template_free.html index 007288060..bc3630886 100644 --- a/docs/dataset-formats/template_free.html +++ b/docs/dataset-formats/template_free.html @@ -697,12 +697,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/dataset-formats/tokenized.html b/docs/dataset-formats/tokenized.html index 6869457a6..759c2879d 100644 --- a/docs/dataset-formats/tokenized.html +++ b/docs/dataset-formats/tokenized.html @@ -697,12 +697,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/dataset_loading.html b/docs/dataset_loading.html index 7cf8cd263..3a385237b 100644 --- a/docs/dataset_loading.html +++ b/docs/dataset_loading.html @@ -697,12 +697,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/dataset_preprocessing.html b/docs/dataset_preprocessing.html index 8e42249c4..68b327f98 100644 --- a/docs/dataset_preprocessing.html +++ b/docs/dataset_preprocessing.html @@ -662,12 +662,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/debugging.html b/docs/debugging.html index 49a259eff..35530faf5 100644 --- a/docs/debugging.html +++ b/docs/debugging.html @@ -697,12 +697,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • @@ -919,8 +913,9 @@ If you prefer to watch a video, rather than read, you can skip to the

    Setup

    Make sure you have an editable install of Axolotl, which ensures that changes you make to the code are reflected at runtime. Run the following commands from the root of this project:

    -
    pip3 install packaging
-pip3 install --no-build-isolation -e '.[flash-attn,deepspeed]'
    +
    export UV_TORCH_BACKEND=cu128  # or cu130
+uv sync --extra flash-attn --extra deepspeed --group dev --group test
+source .venv/bin/activate

    Remote Hosts

    If you developing on a remote host, you can easily use VSCode to debug remotely. To do so, you will need to follow this remote - SSH guide. You can also see the video below on Docker and Remote SSH debugging.

    @@ -1034,14 +1029,14 @@ You may not want to delete these folders. For example, if you are debugging mode If you already have axolotl cloned on your host, make sure you have the latest changes and change into the root of the project.

    Next, run the desired docker image and mount the current directory. Below is a docker command you can run to do this:2

    -
    docker run --privileged --gpus '"all"' --shm-size 10g --rm -it --name axolotl --ipc=host --ulimit memlock=-1 --ulimit stack=67108864 --mount type=bind,src="${PWD}",target=/workspace/axolotl -v ${HOME}/.cache/huggingface:/root/.cache/huggingface axolotlai/axolotl:main-py3.10-cu118-2.0.1
    +
    docker run --privileged --gpus '"all"' --shm-size 10g --rm -it --name axolotl --ipc=host --ulimit memlock=-1 --ulimit stack=67108864 --mount type=bind,src="${PWD}",target=/workspace/axolotl -v ${HOME}/.cache/huggingface:/root/.cache/huggingface axolotlai/axolotl-uv:main-latest

    [!Tip] To understand which containers are available, see the Docker section of the README and the DockerHub repo. For details of how the Docker containers are built, see axolotl’s Docker CI builds.

    -

    You will now be in the container. Next, perform an editable install of Axolotl:

    -
    pip3 install packaging
-pip3 install --no-build-isolation -e '.[flash-attn,deepspeed]'
    +

    You will now be in the container. Next, install Axolotl with dev dependencies:

    +
    uv sync --extra flash-attn --extra deepspeed --group dev --group test
+source .venv/bin/activate

    Attach To Container

    diff --git a/docs/docker.html b/docs/docker.html index 62001ed89..3c5300f06 100644 --- a/docs/docker.html +++ b/docs/docker.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • @@ -833,7 +827,8 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); -

    This section describes the different Docker images that are released by AxolotlAI at Docker Hub.

    +

    This section describes the different Docker images that are released by AxolotlAI at +Docker Hub.

    @@ -844,24 +839,61 @@ Important
    -

    For Blackwell GPUs, please use the tags with PyTorch 2.7.1 and CUDA 12.8.

    +

    For Blackwell GPUs, please use the tags with PyTorch 2.9.1 and CUDA 12.8.

    +
    +
    +
    +
    +
    + +
    +
    +Tip +
    +
    +
    +

    Each image below is available in a uv variant that uses uv with +a relocatable venv (/workspace/axolotl-venv) instead of Miniconda + pip. Append -uv to the image name +(e.g. axolotlai/axolotl-base-uv). Tags follow the same format. We recommend the uv images for new deployments.

    Base

    -

    The base image is the most minimal image that can install Axolotl. It is based on the nvidia/cuda image. It includes python, torch, git, git-lfs, awscli, pydantic, and more.

    +

    The base image is the most minimal image that can install Axolotl. It is based on the nvidia/cuda image. +It includes python, torch, git, git-lfs, awscli, pydantic, and more.

    Image

    -
    axolotlai/axolotl-base
    -

    Link: Docker Hub

    + + + + + + + + + + + + + + + + + + + + +
    VariantImageDocker Hub
    pipaxolotlai/axolotl-baseLink
    uvaxolotlai/axolotl-base-uvLink

    Tags format

    -
    main-base-py{python_version}-cu{cuda_version}-{pytorch_version}
    +
    main-base-py{python_version}-cu{cuda_version}-{pytorch_version}

    Tags examples:

      -
    • main-base-py3.11-cu128-2.8.0
    • main-base-py3.11-cu128-2.9.1
      • +
    • main-base-py3.12-cu128-2.10.0
      • +
    • main-base-py3.12-cu130-2.9.1
      • +
    • main-base-py3.12-cu130-2.10.0
    @@ -870,22 +902,41 @@ Important

    The main image is the image that is used to run Axolotl. It is based on the axolotlai/axolotl-base image and includes the Axolotl codebase, dependencies, and more.

    Image

    -
    axolotlai/axolotl
    -

    Link: Docker Hub

    + + + + + + + + + + + + + + + + + + + + +
    VariantImageDocker Hub
    pipaxolotlai/axolotlLink
    uvaxolotlai/axolotl-uvLink

    Tags format

    -
    # on push to main
-main-py{python_version}-cu{cuda_version}-{pytorch_version}
-
-# latest main (currently torch 2.6.0, python 3.11, cuda 12.4)
-main-latest
-
-# nightly build
-{branch}-{date_in_YYYYMMDD}-py{python_version}-cu{cuda_version}-{pytorch_version}
-
-# tagged release
-{version}
    +
    # on push to main
+main-py{python_version}-cu{cuda_version}-{pytorch_version}
+
+# latest main (currently torch 2.9.1, python 3.11, cuda 12.8)
+main-latest
+
+# nightly build
+{branch}-{date_in_YYYYMMDD}-py{python_version}-cu{cuda_version}-{pytorch_version}
+
+# tagged release
+{version}
    @@ -901,11 +952,12 @@ Tip

    Tags examples:

      -
    • main-py3.11-cu128-2.8.0
    • main-py3.11-cu128-2.9.1
      • +
    • main-py3.12-cu128-2.10.0
      • +
    • main-py3.12-cu130-2.9.1
      • +
    • main-py3.12-cu130-2.10.0
    • main-latest
      • -
    • main-20250303-py3.11-cu124-2.6.0
      • -
    • main-20250303-py3.11-cu126-2.6.0
      • +
    • main-20260315-py3.11-cu128-2.9.1
    • 0.12.0
    @@ -928,8 +980,27 @@ Tip

    Image

    -
    axolotlai/axolotl-cloud
    -

    Link: Docker Hub

    + + + + + + + + + + + + + + + + + + + + +
    VariantImageDocker Hub
    pipaxolotlai/axolotl-cloudLink
    uvaxolotlai/axolotl-cloud-uvLink

    Tags format

    diff --git a/docs/ebft.html b/docs/ebft.html index a49150984..123cd3f92 100644 --- a/docs/ebft.html +++ b/docs/ebft.html @@ -697,12 +697,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/expert_quantization.html b/docs/expert_quantization.html index c5d98d371..a56b3742e 100644 --- a/docs/expert_quantization.html +++ b/docs/expert_quantization.html @@ -697,12 +697,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/faq.html b/docs/faq.html index 777eaa3d4..9cf9937ac 100644 --- a/docs/faq.html +++ b/docs/faq.html @@ -697,12 +697,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/fsdp_qlora.html b/docs/fsdp_qlora.html index 1118c1e42..c1d9c76fc 100644 --- a/docs/fsdp_qlora.html +++ b/docs/fsdp_qlora.html @@ -662,12 +662,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/getting-started.html b/docs/getting-started.html index 12af41f46..6ab6e0462 100644 --- a/docs/getting-started.html +++ b/docs/getting-started.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/gradient_checkpointing.html b/docs/gradient_checkpointing.html index e6678e8b8..c106e4d57 100644 --- a/docs/gradient_checkpointing.html +++ b/docs/gradient_checkpointing.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/grpo.html b/docs/grpo.html index 80a3b8db6..b184c268d 100644 --- a/docs/grpo.html +++ b/docs/grpo.html @@ -697,12 +697,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/inference.html b/docs/inference.html index ec3e9a3de..e8770b479 100644 --- a/docs/inference.html +++ b/docs/inference.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/input_output.html b/docs/input_output.html index 0961b3b39..036972c34 100644 --- a/docs/input_output.html +++ b/docs/input_output.html @@ -662,12 +662,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/installation.html b/docs/installation.html index 04c3dd8bb..6aebb81ae 100644 --- a/docs/installation.html +++ b/docs/installation.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • @@ -787,12 +781,11 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true});
    @@ -840,25 +831,11 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); -
    -

    2 Installation Methods

    -
    -
    -
    - -
    -
    -Important -
    -
    -
    -

    Please make sure to have Pytorch installed before installing Axolotl in your local environment.

    -

    Follow the instructions at: https://pytorch.org/get-started/locally/

    -
    -
    +
    +

    2 Installation

    @@ -872,50 +849,33 @@ Important

    For Blackwell GPUs, please use Pytorch 2.9.1 and CUDA 12.8.

    -
    -

    2.1 PyPI Installation (Recommended)

    -
    pip3 install -U packaging setuptools wheel ninja
-pip3 install --no-build-isolation axolotl[flash-attn,deepspeed]
    -

    We use --no-build-isolation in order to detect the installed PyTorch version (if -installed) in order not to clobber it, and so that we set the correct version of -dependencies that are specific to the PyTorch version or other installed -co-dependencies.

    +
    +

    2.1 Quick Install

    +

    Axolotl uses uv as its package manager. uv is a fast, reliable Python package installer and resolver built in Rust.

    +

    Install uv if not already installed:

    +
    curl -LsSf https://astral.sh/uv/install.sh | sh
+source $HOME/.local/bin/env
    +

    Choose your CUDA version (e.g. cu128, cu130), create a venv, and install:

    +
    export UV_TORCH_BACKEND=cu128  # or cu130
+uv venv --no-project --relocatable
+source .venv/bin/activate
+uv pip install --no-build-isolation axolotl[flash-attn,deepspeed]
    -
    -

    2.2 uv Installation

    -

    uv is a fast, reliable Python package installer and resolver built in Rust. It offers significant performance improvements over pip and provides better dependency resolution, making it an excellent choice for complex environments.

    -

    Install uv if not already installed

    -
    curl -LsSf https://astral.sh/uv/install.sh | sh
-source $HOME/.local/bin/env
    -

    Choose your CUDA version to use with PyTorch; e.g. cu124, cu126, cu128, -then create the venv and activate

    -
    export UV_TORCH_BACKEND=cu126
-uv venv --no-project --relocatable
-source .venv/bin/activate
    -

    Install PyTorch -- PyTorch 2.6.0 recommended

    -
    uv pip install packaging setuptools wheel
-uv pip install torch==2.6.0
-uv pip install awscli pydantic
    -

    Install axolotl from PyPi

    -
    uv pip install --no-build-isolation axolotl[deepspeed,flash-attn]
-
-# optionally install with vLLM if you're using torch==2.6.0 and want to train w/ GRPO
-uv pip install --no-build-isolation axolotl[deepspeed,flash-attn,vllm]
    -
    -
    -

    2.3 Edge/Development Build

    +
    +

    2.2 Edge/Development Build

    For the latest features between releases:

    -
    git clone https://github.com/axolotl-ai-cloud/axolotl.git
-cd axolotl
-pip3 install -U packaging setuptools wheel ninja
-pip3 install --no-build-isolation -e '.[flash-attn,deepspeed]'
    +
    git clone https://github.com/axolotl-ai-cloud/axolotl.git
+cd axolotl
+export UV_TORCH_BACKEND=cu128  # or cu130
+uv sync --extra flash-attn --extra deepspeed
+source .venv/bin/activate
    +

    uv sync creates a .venv, installs exact pinned versions from uv.lock, and sets up an editable install automatically.

    -
    -

    2.4 Docker

    -
    docker run --gpus '"all"' --rm -it axolotlai/axolotl:main-latest
    +
    +

    2.3 Docker

    +
    docker run --gpus '"all"' --rm -it --ipc=host axolotlai/axolotl-uv:main-latest

    For development with Docker:

    -
    docker compose up -d
    +
    docker compose up -d
    @@ -926,12 +886,12 @@ then create the venv and activate

    -
    docker run --privileged --gpus '"all"' --shm-size 10g --rm -it \
-  --name axolotl --ipc=host \
-  --ulimit memlock=-1 --ulimit stack=67108864 \
-  --mount type=bind,src="${PWD}",target=/workspace/axolotl \
-  -v ${HOME}/.cache/huggingface:/root/.cache/huggingface \
-  axolotlai/axolotl:main-latest
    +
    docker run --privileged --gpus '"all"' --shm-size 10g --rm -it \
+  --name axolotl --ipc=host \
+  --ulimit memlock=-1 --ulimit stack=67108864 \
+  --mount type=bind,src="${PWD}",target=/workspace/axolotl \
+  -v ${HOME}/.cache/huggingface:/root/.cache/huggingface \
+  axolotlai/axolotl-uv:main-latest
    @@ -944,7 +904,7 @@ Important
    -

    For Blackwell GPUs, please use axolotlai/axolotl:main-py3.11-cu128-2.9.1 or the cloud variant axolotlai/axolotl-cloud:main-py3.11-cu128-2.9.1.

    +

    For Blackwell GPUs, please use axolotlai/axolotl-uv:main-py3.11-cu128-2.9.1 or the cloud variant axolotlai/axolotl-cloud-uv:main-py3.11-cu128-2.9.1.

    Please refer to the Docker documentation for more information on the different Docker images that are available.

    @@ -956,7 +916,7 @@ Important

    3.1 Cloud GPU Providers

    For providers supporting Docker:

      -
    • Use axolotlai/axolotl-cloud:main-latest
      • +
    • Use axolotlai/axolotl-cloud-uv:main-latest
    • Available on:
      • RunPod
        • @@ -978,8 +938,8 @@ Important

        4 Platform-Specific Instructions

        4.1 macOS

        -
        pip3 install --no-build-isolation -e '.'
        -

        See Section 6 for Mac-specific issues.

        +
        uv pip install --no-build-isolation -e '.'
        +

        See Section 7 for Mac-specific issues.

        4.2 Windows

        @@ -998,23 +958,46 @@ Important
    -
    -

    5 Environment Managers

    -
    -

    5.1 Conda/Pip venv

    -
      -

    1. Install Python ≥3.11

      2. -

    2. Install PyTorch: https://pytorch.org/get-started/locally/

      3. -

    3. Install Axolotl:

      -
      pip3 install -U packaging setuptools wheel ninja
-pip3 install --no-build-isolation -e '.[flash-attn,deepspeed]'
      4. -

    4. (Optional) Login to Hugging Face:

      -
      hf auth login
      5. -
    +
    +

    5 Migrating from pip to uv

    +

    If you have an existing pip-based Axolotl installation, you can migrate to uv:

    +
    # Install uv
+curl -LsSf https://astral.sh/uv/install.sh | sh
+source $HOME/.local/bin/env
+
+# Create a fresh venv (recommended for a clean start)
+export UV_TORCH_BACKEND=cu128  # or cu130
+uv venv --no-project --relocatable
+source .venv/bin/activate
+
+# Reinstall axolotl
+uv pip install --no-build-isolation axolotl[flash-attn,deepspeed]
    +
    +

    6 Using pip (Alternative)

    +

    If you are unable to install uv, you can still use pip directly.

    +
    +
    +
    + +
    +
    +Important +
    +
    +
    +

    Please make sure to have PyTorch installed before installing Axolotl with pip.

    +

    Follow the instructions at: https://pytorch.org/get-started/locally/

    +
    +
    +
    pip3 install -U packaging setuptools wheel ninja
+pip3 install --no-build-isolation axolotl[flash-attn,deepspeed]
    +

    For editable/development installs:

    +
    pip3 install -U packaging setuptools wheel ninja
+pip3 install --no-build-isolation -e '.[flash-attn,deepspeed]'
    -
    -

    6 Troubleshooting

    +
    +

    7 Troubleshooting

    If you encounter installation issues, see our FAQ and Debugging Guide.

    diff --git a/docs/lora_optims.html b/docs/lora_optims.html index 1ef931140..a4f102e54 100644 --- a/docs/lora_optims.html +++ b/docs/lora_optims.html @@ -697,12 +697,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/lr_groups.html b/docs/lr_groups.html index d45802a34..81a41bbf9 100644 --- a/docs/lr_groups.html +++ b/docs/lr_groups.html @@ -697,12 +697,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/mac.html b/docs/mac.html index 6ef3463f9..cc873c65e 100644 --- a/docs/mac.html +++ b/docs/mac.html @@ -662,12 +662,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/mixed_precision.html b/docs/mixed_precision.html index 8740e7f7b..6576b0f0a 100644 --- a/docs/mixed_precision.html +++ b/docs/mixed_precision.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • diff --git a/docs/models/LiquidAI.html b/docs/models/LiquidAI.html index f3ea1bc9c..455326c92 100644 --- a/docs/models/LiquidAI.html +++ b/docs/models/LiquidAI.html @@ -696,12 +696,6 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true}); FSDP + QLoRA
    -
    • -
  • -
    - - Unsloth -
  • @@ -827,8 +821,7 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true});

  • Install Axolotl following the installation guide.

    Here is an example of how to install from pip:

    # Ensure you have a compatible version of Pytorch installed
-pip3 install packaging setuptools wheel ninja
-pip3 install --no-build-isolation 'axolotl[flash-attn]>=0.12.0'
    • +uv pip install --no-build-isolation 'axolotl[flash-attn]>=0.12.0'

  • Run one of the finetuning examples below.

    LFM2

    # FFT SFT (1x48GB @ 25GiB)
@@ -837,7 +830,7 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true});
 
    # LoRA SFT (1x48GB @ 2.7GiB)
 axolotl train examples/LiquidAI/lfm2-vl-lora.yaml
    
 
    LFM2-MoE
    
-
    pip install git+https://github.com/huggingface/transformers.git@0c9a72e4576fe4c84077f066e585129c97bfd4e6
+
    uv pip install git+https://github.com/huggingface/transformers.git@0c9a72e4576fe4c84077f066e585129c97bfd4e6
 
 # LoRA SFT (1x48GB @ 16.2GiB)
 axolotl train examples/LiquidAI/lfm2-8b-a1b-lora.yaml
    • @@ -846,7 +839,7 @@ gtag('config', 'G-9KYCVJBNMQ', { 'anonymize_ip': true});

    TIPS