@@ -52,13 +52,13 @@ We're always adding new models, but feel free to [file an issue](https://github.
5252torchtune provides the following finetuning recipes for training on one or more devices.
5353
5454
55- | Finetuning method | Devices | Recipe | Example config (s) |
55+ | Finetuning Method | Devices | Recipe | Example Config (s) |
5656| :-:| :-:| :-:| :-:|
57- | Full finetuning | 1-8 | [ full_finetune_single_device] ( recipes/full_finetune_single_device.py ) <br > [ full_finetune_distributed] ( recipes/full_finetune_distributed.py ) | [ Llama3.1 8B single-device] ( recipes/configs/llama3_1/8B_full_single_device.yaml ) <br > [ Llama 3.1 70B distributed] ( recipes/configs/llama3_1/70B_full.yaml )
58- | LoRA finetuning | 1-8 | [ lora_finetune_single_device] ( recipes/lora_finetune_single_device.py ) <br > [ lora_finetune_distributed] ( recipes/lora_finetune_distributed.py ) | [ Qwen2 0.5B single-device] ( recipes/configs/qwen2/0.5B_lora_single_device.yaml ) <br > [ Gemma 7B distributed] ( recipes/configs/gemma/7B_lora.yaml )
59- | QLoRA finetuning | 1-8 | [ lora_finetune_single_device] ( recipes/lora_finetune_single_device.py ) <br > [ lora_finetune_distributed] ( recipes/lora_finetune_distributed.py ) | [ Phi3 Mini single-device] ( recipes/configs/phi3/mini_qlora_single_device.yaml ) <br > [ Llama 3.1 405B distributed] ( recipes/configs/llama3_1/405B_qlora.yaml )
60- | DoRA/QDoRA finetuning | 1-8 | [ lora_finetune_single_device] ( recipes/lora_finetune_single_device.py ) <br > [ lora_finetune_distributed] ( recipes/lora_finetune_distributed.py ) | [ Llama3 8B QDoRA single-device] ( recipes/configs/llama3/8B_qdora_single_device.yaml ) <br > [ Llama3 8B DoRA distributed] ( recipes/configs/llama3/8B_dora.yaml )
61- | Quantization-aware training | 4-8 | [ qat_distributed] ( recipes/qat_distributed.py ) | [ Llama3 8B QAT] ( recipes/configs/llama3/8B_qat_full.yaml )
57+ | Full Finetuning | 1-8 | [ full_finetune_single_device] ( recipes/full_finetune_single_device.py ) <br > [ full_finetune_distributed] ( recipes/full_finetune_distributed.py ) | [ Llama3.1 8B single-device] ( recipes/configs/llama3_1/8B_full_single_device.yaml ) <br > [ Llama 3.1 70B distributed] ( recipes/configs/llama3_1/70B_full.yaml )
58+ | LoRA Finetuning | 1-8 | [ lora_finetune_single_device] ( recipes/lora_finetune_single_device.py ) <br > [ lora_finetune_distributed] ( recipes/lora_finetune_distributed.py ) | [ Qwen2 0.5B single-device] ( recipes/configs/qwen2/0.5B_lora_single_device.yaml ) <br > [ Gemma 7B distributed] ( recipes/configs/gemma/7B_lora.yaml )
59+ | QLoRA Finetuning | 1-8 | [ lora_finetune_single_device] ( recipes/lora_finetune_single_device.py ) <br > [ lora_finetune_distributed] ( recipes/lora_finetune_distributed.py ) | [ Phi3 Mini single-device] ( recipes/configs/phi3/mini_qlora_single_device.yaml ) <br > [ Llama 3.1 405B distributed] ( recipes/configs/llama3_1/405B_qlora.yaml )
60+ | DoRA/QDoRA Finetuning | 1-8 | [ lora_finetune_single_device] ( recipes/lora_finetune_single_device.py ) <br > [ lora_finetune_distributed] ( recipes/lora_finetune_distributed.py ) | [ Llama3 8B QDoRA single-device] ( recipes/configs/llama3/8B_qdora_single_device.yaml ) <br > [ Llama3 8B DoRA distributed] ( recipes/configs/llama3/8B_dora.yaml )
61+ | Quantization-Aware Training | 4-8 | [ qat_distributed] ( recipes/qat_distributed.py ) | [ Llama3 8B QAT] ( recipes/configs/llama3/8B_qat_full.yaml )
6262| Direct Preference Optimization |1-8 | [ lora_dpo_single_device] ( recipes/lora_dpo_single_device.py ) <br > [ lora_dpo_distributed] ( recipes/lora_dpo_distributed.py ) | [ Llama2 7B single-device] ( recipes/configs/llama2/7B_lora_dpo_single_device.yaml ) <br > [ Llama2 7B distributed] ( recipes/configs/llama2/7B_lora_dpo.yaml )
6363| Proximal Policy Optimization | 1 | [ ppo_full_finetune_single_device] ( recipes/ppo_full_finetune_single_device.py ) | [ Mistral 7B] ( recipes/configs/mistral/7B_full_ppo_low_memory.yaml )
6464| Knowledge Distillation | 1 | [ knowledge_distillation_single_device] ( recipes/knowledge_distillation_single_device.py ) | [ Qwen2 1.5B -> 0.5B] ( recipes/configs/qwen2/knowledge_distillation_single_device.yaml )
0 commit comments