LoRA, QLoRA, DPO, PPO, QAT, knowledge distillation, and more Out-of-the-box memory efficiency,...
此外,它还不断添加新的模型,以满足不同场景下的需求。 可定制的微调配方:Torchtune提供了多种微调配方,如全微调、LoRA、QLoRA、DPO、PPO、QAT和知识蒸馏等。这些配方可以应用于一个或多个设备上的训练,为开发者提供了灵活的选择。 高效的内存和性能优化:Torchtune利用最新的PyTorch API,实现了高效的内存管理和性能...
Proximal Policy Optimization 1 ppo_full_finetune_single_device Mistral 7B Knowledge Distillation 1 knowledge_distillation_single_device Qwen2 1.5B -> 0.5B The above configs are just examples to get you started. If you see a model above not listed here, we likely still support it. If you're...
Proximal Policy Optimization 1 ppo_full_finetune_single_device Mistral 7B Knowledge Distillation 1 knowledge_distillation_single_device Qwen2 1.5B -> 0.5B The above configs are just examples to get you started. If you see a model above not listed here, we likely still support it. If you're...
ppo_full_finetune_single_device mistral/7B_full_ppo_low_memory lora_finetune_distributed llama2/7B_lora llama2/13B_lora llama2/70B_lora llama2/7B_qlora llama2/70B_qlora llama3/8B_dora llama3/70B_lora llama3_1/70B_lora llama3_3/70B_lora llama3_3/70B_qlora llama3/8B_lora llama3...
效果：1. 简化微调流程：TorchTune 封装了微调过程中常见的操作，减少代码量，降低出错概率。2. 提升...
ppo_full_finetune_single_device.py qat_distributed.py quantization.md quantize.py tests torchtune .flake8 .gitignore .pre-commit-config.yaml CODE_OF_CONDUCT.md CONTRIBUTING.md LICENSE MANIFEST.in README.md pyproject.toml version.txt Breadcrumbs torchtune /recipes / qat_distributed.py Latest commi...
recipes/ppo_full_finetune_single_device.py Outdated @@ -119,6 +119,12 @@ def __init__(self, cfg: DictConfig) -> None: self._log_every_n_steps = cfg.get("log_every_n_steps", 1) self._log_peak_memory_stats = cfg.get("log_peak_memory_stats", False) if self._log_peak...
2 changes: 2 additions & 0 deletions 2 recipes/ppo_full_finetune_single_device.py Original file line numberDiff line numberDiff line change @@ -935,6 +935,7 @@ def train(self) -> None: curr_epoch == 0 and self.profiler_profile_memory and idx == self.profiler_wait_steps + self...
test_ppo_full_finetune_single_device.py test_qat_distributed.py utils.py regression_tests torchtune __init__.py cache_artifacts.sh common.py conftest.py test_import_recipes.py test_utils.py torchtune .flake8 .gitignore .pre-commit-config.yaml CODE_OF_CONDUCT.md CON...