release v0.6.1

Former-commit-id: a59d823f554505b2e649e6e111b9dee8306d3ad8

README.md

@@ -5,7 +5,7 @@
 [![GitHub last commit](https://img.shields.io/github/last-commit/hiyouga/LLaMA-Factory)](https://github.com/hiyouga/LLaMA-Factory/commits/main)
 [![PyPI](https://img.shields.io/pypi/v/llmtuner)](https://pypi.org/project/llmtuner/)
 [![Downloads](https://static.pepy.tech/badge/llmtuner)](https://pypi.org/project/llmtuner/)
-[![Citation](https://img.shields.io/badge/citation-26-green)](#projects-using-llama-factory)
+[![Citation](https://img.shields.io/badge/citation-27-green)](#projects-using-llama-factory)
 [![GitHub pull request](https://img.shields.io/badge/PRs-welcome-blue)](https://github.com/hiyouga/LLaMA-Factory/pulls)
 [![Discord](https://dcbadge.vercel.app/api/server/rKfvV9r9FK?compact=true&style=flat)](https://discord.gg/rKfvV9r9FK)
 [![Twitter](https://img.shields.io/twitter/follow/llamafactory_ai)](https://twitter.com/llamafactory_ai)
@@ -72,14 +72,14 @@ Compared to ChatGLM's [P-Tuning](https://github.com/THUDM/ChatGLM2-6B/tree/main/
 
 [24/03/20] We supported **FSDP+QLoRA** that fine-tunes a 70B model on 2x24GB GPUs. See `examples/fsdp_qlora` for usage.
 
-[24/03/13] We supported **[LoRA+](https://arxiv.org/abs/2402.12354)**. Try `loraplus_lr_ratio=16.0` to enable LoRA+ algorithm.
+[24/03/13] We supported **[LoRA+](https://arxiv.org/abs/2402.12354)**. See `examples/extras/loraplus` for usage.
 
-[24/03/07] We supported gradient low-rank projection (**[GaLore](https://arxiv.org/abs/2403.03507)**) algorithm. Try `--use_galore` to use the memory-efficient optimizer.
+[24/03/07] We supported gradient low-rank projection (**[GaLore](https://arxiv.org/abs/2403.03507)**) algorithm. See `examples/extras/galore` for usage.
-
-[24/03/07] We integrated **[vLLM](https://github.com/vllm-project/vllm)** for faster and concurrent inference. Try `--infer_backend vllm` to enjoy **270%** inference speed. (LoRA is not yet supported, merge it first.)
 
 <details><summary>Full Changelog</summary>
 
+[24/03/07] We integrated **[vLLM](https://github.com/vllm-project/vllm)** for faster and concurrent inference. Try `--infer_backend vllm` to enjoy **270%** inference speed. (LoRA is not yet supported, merge it first.)
+
 [24/02/28] We supported weight-decomposed LoRA (**[DoRA](https://arxiv.org/abs/2402.09353)**). Try `--use_dora` to activate DoRA training.
 
 [24/02/15] We supported **block expansion** proposed by [LLaMA Pro](https://github.com/TencentARC/LLaMA-Pro). See `examples/extras/llama_pro` for usage.
@@ -451,7 +451,7 @@ CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \
 ```
 
 > [!TIP]
-> Use `--adapter_name_or_path path_to_sft_checkpoint,path_to_ppo_checkpoint` to infer the fine-tuned model.
+> Use `--adapter_name_or_path path_to_sft_checkpoint,path_to_ppo_checkpoint` to infer the fine-tuned model if `--create_new_adapter` was enabled.
 
 > [!WARNING]
 > Use `--per_device_train_batch_size=1` for LLaMA-2 models in fp16 PPO training.
@@ -482,7 +482,7 @@ CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \
 ```
 
 > [!TIP]
-> Use `--adapter_name_or_path path_to_sft_checkpoint,path_to_dpo_checkpoint` to infer the fine-tuned model.
+> Use `--adapter_name_or_path path_to_sft_checkpoint,path_to_dpo_checkpoint` to infer the fine-tuned model if `--create_new_adapter` was enabled.
 
 ### Distributed Training
 
@@ -570,7 +570,7 @@ deepspeed --num_gpus 8 src/train_bash.py \
 ### Merge LoRA weights and export model
 
 ```bash
-CUDA_VISIBLE_DEVICES=0 python src/export_model.py \
+CUDA_VISIBLE_DEVICES= python src/export_model.py \
     --model_name_or_path path_to_llama_model \
     --adapter_name_or_path path_to_checkpoint \
     --template default \
@@ -586,7 +586,7 @@ CUDA_VISIBLE_DEVICES=0 python src/export_model.py \
 > [!TIP]
 > Use `--model_name_or_path path_to_export` solely to use the exported model.
 > 
-> Use `--export_quantization_bit 4` and `--export_quantization_dataset data/c4_demo.json` to quantize the model with AutoGPTQ after merging the LoRA weights.
+> Use `CUDA_VISIBLE_DEVICES=0`, `--export_quantization_bit 4` and `--export_quantization_dataset data/c4_demo.json` to quantize the model with AutoGPTQ after merging the LoRA weights.
 
 ### Inference with OpenAI-style API
 
@@ -662,19 +662,23 @@ CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \
 
 ### Dockerize Training
 
-#### Get ready
+#### Use Docker
-
-Necessary dockerized environment is needed, such as Docker or Docker Compose.
-
-#### Docker support
 
 ```bash
 docker build -f ./Dockerfile -t llama-factory:latest .
 
-docker run --gpus=all -v ./hf_cache:/root/.cache/huggingface/ -v ./data:/app/data -v ./output:/app/output -p 7860:7860 --shm-size 16G --name llama_factory -d llama-factory:latest
+docker run --gpus=all \
+    -v ./hf_cache:/root/.cache/huggingface/ \
+    -v ./data:/app/data \
+    -v ./output:/app/output \
+    -e CUDA_VISIBLE_DEVICES=0 \
+    -p 7860:7860 \
+    --shm-size 16G \
+    --name llama_factory \
+    -d llama-factory:latest
 ```
 
-#### Docker Compose support
+#### Use Docker Compose
 
 ```bash
 docker compose -f ./docker-compose.yml up -d
@@ -682,7 +686,7 @@ docker compose -f ./docker-compose.yml up -d
 
 > [!TIP]
 > Details about volume:
-> * hf_cache: Utilize Huggingface cache on the host machine. Reassignable if a cache already exists in a different directory.
+> * hf_cache: Utilize Hugging Face cache on the host machine. Reassignable if a cache already exists in a different directory.
 > * data: Place datasets on this dir of the host machine so that they can be selected on LLaMA Board GUI.
 > * output: Set export dir to this location so that the merged result can be accessed directly on the host machine.
 
@@ -709,6 +713,7 @@ docker compose -f ./docker-compose.yml up -d
 1. Huang et al. Key-Point-Driven Data Synthesis with its Enhancement on Mathematical Reasoning. 2024. [[arxiv]](https://arxiv.org/abs/2403.02333)
 1. Duan et al. Negating Negatives: Alignment without Human Positive Samples via Distributional Dispreference Optimization. 2024. [[arxiv]](https://arxiv.org/abs/2403.03419)
 1. Xie and Schwertfeger. Empowering Robotics with Large Language Models: osmAG Map Comprehension with LLMs. 2024. [[arxiv]](https://arxiv.org/abs/2403.08228)
+1. Hongbin Na. CBT-LLM: A Chinese Large Language Model for Cognitive Behavioral Therapy-based Mental Health Question Answering. 2024. [[arxiv]](https://arxiv.org/abs/2403.16008)
 1. **[StarWhisper](https://github.com/Yu-Yang-Li/StarWhisper)**: A large language model for Astronomy, based on ChatGLM2-6B and Qwen-14B.
 1. **[DISC-LawLLM](https://github.com/FudanDISC/DISC-LawLLM)**: A large language model specialized in Chinese legal domain, based on Baichuan-13B, is capable of retrieving and reasoning on legal knowledge.
 1. **[Sunsimiao](https://github.com/thomas-yanxin/Sunsimiao)**: A large language model specialized in Chinese medical domain, based on Baichuan-7B and ChatGLM-6B.

README_zh.md

@@ -5,7 +5,7 @@
 [![GitHub last commit](https://img.shields.io/github/last-commit/hiyouga/LLaMA-Factory)](https://github.com/hiyouga/LLaMA-Factory/commits/main)
 [![PyPI](https://img.shields.io/pypi/v/llmtuner)](https://pypi.org/project/llmtuner/)
 [![Downloads](https://static.pepy.tech/badge/llmtuner)](https://pypi.org/project/llmtuner/)
-[![Citation](https://img.shields.io/badge/citation-26-green)](#使用了-llama-factory-的项目)
+[![Citation](https://img.shields.io/badge/citation-27-green)](#使用了-llama-factory-的项目)
 [![GitHub pull request](https://img.shields.io/badge/PRs-welcome-blue)](https://github.com/hiyouga/LLaMA-Factory/pulls)
 [![Discord](https://dcbadge.vercel.app/api/server/rKfvV9r9FK?compact=true&style=flat)](https://discord.gg/rKfvV9r9FK)
 [![Twitter](https://img.shields.io/twitter/follow/llamafactory_ai)](https://twitter.com/llamafactory_ai)
@@ -72,14 +72,14 @@ https://github.com/hiyouga/LLaMA-Factory/assets/16256802/ec36a9dd-37f4-4f72-81bd
 
 [24/03/20] 我们支持了能在 2x24GB GPU 上微调 70B 模型的 **FSDP+QLoRA**。详细用法请参照 `examples/fsdp_qlora`。
 
-[24/03/13] 我们支持了 **[LoRA+](https://arxiv.org/abs/2402.12354)**。请使用 `loraplus_lr_ratio=16.0` 参数开启 LoRA+ 方法。
+[24/03/13] 我们支持了 **[LoRA+](https://arxiv.org/abs/2402.12354)**。详细用法请参照 `examples/extras/loraplus`。
 
-[24/03/07] 我们支持了梯度低秩投影（**[GaLore](https://arxiv.org/abs/2403.03507)**）算法。请使用 `--use_galore` 参数切换显存高效的优化器。
+[24/03/07] 我们支持了梯度低秩投影（**[GaLore](https://arxiv.org/abs/2403.03507)**）算法。详细用法请参照 `examples/extras/galore`。
-
-[24/03/07] 我们集成了 **[vLLM](https://github.com/vllm-project/vllm)** 以实现极速并发推理。请使用 `--infer_backend vllm` 来获得 **270%** 的推理速度。（尚不支持 LoRA，请先合并权重。）
 
 <details><summary>展开日志</summary>
 
+[24/03/07] 我们集成了 **[vLLM](https://github.com/vllm-project/vllm)** 以实现极速并发推理。请使用 `--infer_backend vllm` 来获得 **270%** 的推理速度。（尚不支持 LoRA，请先合并权重。）
+
 [24/02/28] 我们支持了 **[DoRA](https://arxiv.org/abs/2402.09353)** 微调。请使用 `--use_dora` 参数进行 DoRA 微调。
 
 [24/02/15] 我们支持了 [LLaMA Pro](https://github.com/TencentARC/LLaMA-Pro) 提出的**块扩展**方法。详细用法请参照 `examples/extras/llama_pro`。
@@ -450,7 +450,7 @@ CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \
 ```
 
 > [!TIP]
-> 使用 `--adapter_name_or_path path_to_sft_checkpoint,path_to_ppo_checkpoint` 来进行微调模型的推理。
+> 如果开启了 `--create_new_adapter`，则使用 `--adapter_name_or_path path_to_sft_checkpoint,path_to_ppo_checkpoint` 来进行微调模型的推理。
 
 > [!WARNING]
 > 如果使用 fp16 精度进行 LLaMA-2 模型的 PPO 训练，请使用 `--per_device_train_batch_size=1`。
@@ -481,7 +481,7 @@ CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \
 ```
 
 > [!TIP]
-> 使用 `--adapter_name_or_path path_to_sft_checkpoint,path_to_dpo_checkpoint` 来进行微调模型的推理。
+> 如果开启了 `--create_new_adapter`，则使用 `--adapter_name_or_path path_to_sft_checkpoint,path_to_dpo_checkpoint` 来进行微调模型的推理。
 
 ### 多 GPU 分布式训练
 
@@ -569,7 +569,7 @@ deepspeed --num_gpus 8 src/train_bash.py \
 ### 合并 LoRA 权重并导出模型
 
 ```bash
-CUDA_VISIBLE_DEVICES=0 python src/export_model.py \
+CUDA_VISIBLE_DEVICES= python src/export_model.py \
     --model_name_or_path path_to_llama_model \
     --adapter_name_or_path path_to_checkpoint \
     --template default \
@@ -585,7 +585,7 @@ CUDA_VISIBLE_DEVICES=0 python src/export_model.py \
 > [!TIP]
 > 仅使用 `--model_name_or_path path_to_export` 来加载导出后的模型。
 > 
-> 合并 LoRA 权重之后可再次使用 `--export_quantization_bit 4` 和 `--export_quantization_dataset data/c4_demo.json` 基于 AutoGPTQ 量化模型。
+> 合并 LoRA 权重之后可再次使用 `CUDA_VISIBLE_DEVICES=0`、`--export_quantization_bit 4` 和 `--export_quantization_dataset data/c4_demo.json` 基于 AutoGPTQ 量化模型。
 
 ### 使用 OpenAI 风格 API 推理
 
@@ -659,6 +659,36 @@ CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \
 > [!TIP]
 > 我们建议在量化模型的预测中使用 `--per_device_eval_batch_size=1` 和 `--max_target_length 128`。
 
+### 使用容器
+
+#### 使用 Docker
+
+```bash
+docker build -f ./Dockerfile -t llama-factory:latest .
+
+docker run --gpus=all \
+    -v ./hf_cache:/root/.cache/huggingface/ \
+    -v ./data:/app/data \
+    -v ./output:/app/output \
+    -e CUDA_VISIBLE_DEVICES=0 \
+    -p 7860:7860 \
+    --shm-size 16G \
+    --name llama_factory \
+    -d llama-factory:latest
+```
+
+#### 使用 Docker Compose
+
+```bash
+docker compose -f ./docker-compose.yml up -d
+```
+
+> [!TIP]
+> 数据卷详情：
+> * hf_cache：使用宿主机的 Hugging Face 缓存文件夹，允许更改为新的目录。
+> * data：宿主机中存放数据集的文件夹路径。
+> * output：将导出目录设置为该路径后，即可在宿主机中访问导出后的模型。
+
 ## 使用了 LLaMA Factory 的项目
 
 1. Wang et al. ESRL: Efficient Sampling-based Reinforcement Learning for Sequence Generation. 2023. [[arxiv]](https://arxiv.org/abs/2308.02223)
@@ -682,6 +712,7 @@ CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \
 1. Huang et al. Key-Point-Driven Data Synthesis with its Enhancement on Mathematical Reasoning. 2024. [[arxiv]](https://arxiv.org/abs/2403.02333)
 1. Duan et al. Negating Negatives: Alignment without Human Positive Samples via Distributional Dispreference Optimization. 2024. [[arxiv]](https://arxiv.org/abs/2403.03419)
 1. Xie and Schwertfeger. Empowering Robotics with Large Language Models: osmAG Map Comprehension with LLMs. 2024. [[arxiv]](https://arxiv.org/abs/2403.08228)
+1. Hongbin Na. CBT-LLM: A Chinese Large Language Model for Cognitive Behavioral Therapy-based Mental Health Question Answering. 2024. [[arxiv]](https://arxiv.org/abs/2403.16008)
 1. **[StarWhisper](https://github.com/Yu-Yang-Li/StarWhisper)**: 天文大模型 StarWhisper，基于 ChatGLM2-6B 和 Qwen-14B 在天文数据上微调而得。
 1. **[DISC-LawLLM](https://github.com/FudanDISC/DISC-LawLLM)**: 中文法律领域大模型 DISC-LawLLM，基于 Baichuan-13B 微调而得，具有法律推理和知识检索能力。
 1. **[Sunsimiao](https://github.com/thomas-yanxin/Sunsimiao)**: 孙思邈中文医疗大模型 Sumsimiao，基于 Baichuan-7B 和 ChatGLM-6B 在中文医疗数据上微调而得。

docker-compose.yml

@@ -10,6 +10,8 @@ services:
       - ./hf_cache:/root/.cache/huggingface/
       - ./data:/app/data
       - ./output:/app/output
+    environment:
+      - CUDA_VISIBLE_DEVICES=0
     ports:
       - "7860:7860"
     ipc: host

examples/accelerate/fsdp_config.yaml

@@ -9,7 +9,7 @@ fsdp_config:
   fsdp_forward_prefetch: false
   fsdp_offload_params: true
   fsdp_sharding_strategy: FULL_SHARD
-  fsdp_state_dict_type: SHARDED_STATE_DICT
+  fsdp_state_dict_type: FULL_STATE_DICT
   fsdp_sync_module_states: true
   fsdp_use_orig_params: false
 machine_rank: 0

examples/extras/galore/adamw.sh

@@ -1,31 +0,0 @@
-#!/bin/bash
-
-CUDA_VISIBLE_DEVICES=0 python ../../../src/train_bash.py \
-    --stage sft \
-    --do_train \
-    --model_name_or_path meta-llama/Llama-2-7b-hf \
-    --dataset alpaca_gpt4_en,glaive_toolcall \
-    --dataset_dir ../../../data \
-    --template default \
-    --finetuning_type full \
-    --output_dir ../../../saves/LLaMA2-7B/galore/sft \
-    --overwrite_cache \
-    --overwrite_output_dir \
-    --cutoff_len 1024 \
-    --preprocessing_num_workers 16 \
-    --per_device_train_batch_size 1 \
-    --per_device_eval_batch_size 1 \
-    --gradient_accumulation_steps 1 \
-    --lr_scheduler_type cosine \
-    --logging_steps 10 \
-    --warmup_steps 20 \
-    --save_steps 100 \
-    --eval_steps 100 \
-    --evaluation_strategy steps \
-    --load_best_model_at_end \
-    --learning_rate 5e-5 \
-    --num_train_epochs 3.0 \
-    --max_samples 3000 \
-    --val_size 0.1 \
-    --plot_loss \
-    --fp16

examples/extras/galore/adamw_8bit_bf16.sh

@@ -1,32 +0,0 @@
-#!/bin/bash
-
-CUDA_VISIBLE_DEVICES=0 python ../../../src/train_bash.py \
-    --stage sft \
-    --do_train \
-    --model_name_or_path meta-llama/Llama-2-7b-hf \
-    --dataset alpaca_gpt4_en,glaive_toolcall \
-    --dataset_dir ../../../data \
-    --template default \
-    --finetuning_type full \
-    --optim adamw_8bit \
-    --output_dir ../../../saves/LLaMA2-7B/galore/sft \
-    --overwrite_cache \
-    --overwrite_output_dir \
-    --cutoff_len 1024 \
-    --preprocessing_num_workers 16 \
-    --per_device_train_batch_size 1 \
-    --per_device_eval_batch_size 1 \
-    --gradient_accumulation_steps 1 \
-    --lr_scheduler_type cosine \
-    --logging_steps 10 \
-    --warmup_steps 20 \
-    --save_steps 100 \
-    --eval_steps 100 \
-    --evaluation_strategy steps \
-    --load_best_model_at_end \
-    --learning_rate 5e-5 \
-    --num_train_epochs 3.0 \
-    --max_samples 3000 \
-    --val_size 0.1 \
-    --plot_loss \
-    --pure_bf16

examples/extras/galore/galore_adamw_8bit_bf16.sh

@@ -1,36 +0,0 @@
-#!/bin/bash
-
-CUDA_VISIBLE_DEVICES=0 python ../../../src/train_bash.py \
-    --stage sft \
-    --do_train \
-    --model_name_or_path meta-llama/Llama-2-7b-hf \
-    --dataset alpaca_gpt4_en,glaive_toolcall \
-    --dataset_dir ../../../data \
-    --template default \
-    --finetuning_type full \
-    --optim adamw_8bit \
-    --use_galore \
-    --galore_layerwise \
-    --galore_target mlp,self_attn \
-    --galore_rank 128 \
-    --output_dir ../../../saves/LLaMA2-7B/galore/sft \
-    --overwrite_cache \
-    --overwrite_output_dir \
-    --cutoff_len 1024 \
-    --preprocessing_num_workers 16 \
-    --per_device_train_batch_size 1 \
-    --per_device_eval_batch_size 1 \
-    --gradient_accumulation_steps 1 \
-    --lr_scheduler_type cosine \
-    --logging_steps 10 \
-    --warmup_steps 20 \
-    --save_steps 100 \
-    --eval_steps 100 \
-    --evaluation_strategy steps \
-    --load_best_model_at_end \
-    --learning_rate 5e-5 \
-    --num_train_epochs 3.0 \
-    --max_samples 3000 \
-    --val_size 0.1 \
-    --plot_loss \
-    --pure_bf16

examples/extras/galore/sft.sh

@@ -32,4 +32,4 @@ CUDA_VISIBLE_DEVICES=0 python ../../../src/train_bash.py \
     --max_samples 3000 \
     --val_size 0.1 \
     --plot_loss \
-    --fp16
+    --pure_bf16

examples/fsdp_qlora/README.md

@@ -1,5 +1,5 @@
 ```bash
-pip install git+https://github.com/huggingface/transformers.git
+pip install "transformers>=4.39.1"
 pip install "accelerate>=0.28.0"
 pip install "bitsandbytes>=0.43.0"
 ```

requirements.txt

@@ -2,7 +2,7 @@ torch>=1.13.1
 transformers>=4.37.2
 datasets>=2.14.3
 accelerate>=0.27.2
-peft>=0.9.0
+peft>=0.10.0
 trl>=0.8.1
 gradio>=3.38.0,<4.0.0
 scipy

src/llmtuner/__init__.py

@@ -7,5 +7,5 @@ from .train import export_model, run_exp
 from .webui import create_ui, create_web_demo
 
 
-__version__ = "0.6.0"
+__version__ = "0.6.1"
 __all__ = ["create_app", "ChatModel", "Evaluator", "export_model", "run_exp", "create_ui", "create_web_demo"]

src/llmtuner/data/template.py

@@ -689,6 +689,8 @@ _register_template(
 
 _register_template(
     name="vanilla",
+    format_separator=EmptyFormatter(slots=["\n"]),
+    efficient_eos=True,
 )
 
 

src/llmtuner/eval/template.py

@@ -1,14 +1,10 @@
 from dataclasses import dataclass
-from typing import TYPE_CHECKING, Dict, List, Tuple
+from typing import Dict, List, Sequence, Tuple
 
 from ..data import Role
 from ..extras.constants import CHOICES
 
 
-if TYPE_CHECKING:
-    from datasets import Dataset
-
-
 @dataclass
 class EvalTemplate:
     system: str
@@ -16,22 +12,29 @@ class EvalTemplate:
     answer: str
     prefix: str
 
-    def parse_example(self, example: Dict[str, str]) -> Tuple[str, str]:
+    def _parse_example(self, example: Dict[str, str]) -> Tuple[str, str]:
+        r"""
+        input: a dict with keys {"question", "A", "B", "C", "D", "answer"}
+        output: a tuple of (prompt, response)
+        """
         candidates = [self.choice.format(choice=ch, content=example[ch]) for ch in CHOICES if ch in example]
         return "".join([example["question"]] + candidates + [self.answer]), example["answer"]
 
     def format_example(
-        self, target_data: Dict[str, str], support_set: "Dataset", subject_name: str
+        self, target_data: Dict[str, str], support_set: Sequence[Dict[str, str]], subject_name: str
     ) -> List[Dict[str, str]]:
+        r"""
+        Converts dataset examples to messages.
+        """
         messages = []
         for k in range(len(support_set)):
-            prompt, response = self.parse_example(support_set[k])
+            prompt, response = self._parse_example(support_set[k])
-            messages.append({"role": Role.USER, "content": prompt})
+            messages.append({"role": Role.USER.value, "content": prompt})
-            messages.append({"role": Role.ASSISTANT, "content": response})
+            messages.append({"role": Role.ASSISTANT.value, "content": response})
 
-        prompt, response = self.parse_example(target_data)
+        prompt, response = self._parse_example(target_data)
-        messages.append({"role": Role.USER, "content": prompt})
+        messages.append({"role": Role.USER.value, "content": prompt})
-        messages.append({"role": Role.ASSISTANT, "content": response})
+        messages.append({"role": Role.ASSISTANT.value, "content": response})
         messages[0]["content"] = self.system.format(subject=subject_name) + messages[0]["content"]
         return messages
 
@@ -39,7 +42,7 @@ class EvalTemplate:
 eval_templates: Dict[str, "EvalTemplate"] = {}
 
 
-def register_eval_template(name: str, system: str, choice: str, answer: str, prefix: str) -> None:
+def _register_eval_template(name: str, system: str, choice: str, answer: str, prefix: str) -> None:
     eval_templates[name] = EvalTemplate(system=system, choice=choice, answer=answer, prefix=prefix)
 
 
@@ -49,7 +52,7 @@ def get_eval_template(name: str) -> "EvalTemplate":
     return eval_template
 
 
-register_eval_template(
+_register_eval_template(
     name="en",
     system="The following are multiple choice questions (with answers) about {subject}.\n\n",
     choice="\n{choice}. {content}",
@@ -58,10 +61,10 @@ register_eval_template(
 )
 
 
-register_eval_template(
+_register_eval_template(
     name="zh",
     system="以下是中国关于{subject}考试的单项选择题，请选出其中的正确答案。\n\n",
     choice="\n{choice}. {content}",
     answer="\n答案：",
-    prefix="\n",
+    prefix=" ",
 )

src/llmtuner/extras/callbacks.py

@@ -58,9 +58,17 @@ class LogCallback(TrainerCallback):
             self.in_training = True
             self.start_time = time.time()
             self.max_steps = state.max_steps
-            if os.path.exists(os.path.join(args.output_dir, LOG_FILE_NAME)) and args.overwrite_output_dir:
+
-                logger.warning("Previous log file in this folder will be deleted.")
+        if args.save_on_each_node:
-                os.remove(os.path.join(args.output_dir, LOG_FILE_NAME))
+            if not state.is_local_process_zero:
+                return
+        else:
+            if not state.is_world_process_zero:
+                return
+
+        if os.path.exists(os.path.join(args.output_dir, LOG_FILE_NAME)) and args.overwrite_output_dir:
+            logger.warning("Previous log file in this folder will be deleted.")
+            os.remove(os.path.join(args.output_dir, LOG_FILE_NAME))
 
     def on_train_end(self, args: "TrainingArguments", state: "TrainerState", control: "TrainerControl", **kwargs):
         r"""
@@ -112,8 +120,12 @@ class LogCallback(TrainerCallback):
         r"""
         Event called after logging the last logs.
         """
-        if not state.is_local_process_zero:
+        if args.save_on_each_node:
-            return
+            if not state.is_local_process_zero:
+                return
+        else:
+            if not state.is_world_process_zero:
+                return
 
         logs = dict(
             current_steps=self.cur_steps,

src/llmtuner/extras/misc.py

@@ -64,7 +64,7 @@ def check_dependencies() -> None:
         require_version("transformers>=4.37.2", "To fix: pip install transformers>=4.37.2")
         require_version("datasets>=2.14.3", "To fix: pip install datasets>=2.14.3")
         require_version("accelerate>=0.27.2", "To fix: pip install accelerate>=0.27.2")
-        require_version("peft>=0.9.0", "To fix: pip install peft>=0.9.0")
+        require_version("peft>=0.10.0", "To fix: pip install peft>=0.10.0")
         require_version("trl>=0.8.1", "To fix: pip install trl>=0.8.1")
 
 

src/llmtuner/hparams/finetuning_args.py

@@ -102,6 +102,10 @@ class RLHFArguments:
         default="sigmoid",
         metadata={"help": "The type of DPO loss to use."},
     )
+    dpo_label_smoothing: float = field(
+        default=0.0,
+        metadata={"help": "The robust DPO label smoothing parameter in cDPO that should be between 0 and 0.5."},
+    )
     dpo_ftx: float = field(
         default=0.0,
         metadata={"help": "The supervised fine-tuning loss coefficient in DPO training."},
@@ -114,10 +118,6 @@ class RLHFArguments:
         default=4,
         metadata={"help": "The number of epochs to perform in a PPO optimization step."},
     )
-    ppo_logger: Optional[str] = field(
-        default=None,
-        metadata={"help": 'Log with either "wandb" or "tensorboard" in PPO training.'},
-    )
     ppo_score_norm: bool = field(
         default=False,
         metadata={"help": "Use score normalization in PPO training."},
@@ -248,6 +248,9 @@ class FinetuningArguments(FreezeArguments, LoraArguments, RLHFArguments, GaloreA
         if self.stage == "ppo" and self.reward_model_type == "lora" and self.finetuning_type != "lora":
             raise ValueError("`reward_model_type` cannot be lora for Freeze/Full PPO training.")
 
+        if self.stage == "dpo" and self.dpo_loss != "sigmoid" and self.dpo_label_smoothing > 1e-6:
+            raise ValueError("`dpo_label_smoothing` is only valid for sigmoid loss function.")
+
         if self.use_llama_pro and self.finetuning_type == "full":
             raise ValueError("`use_llama_pro` is only valid for the Freeze or LoRA method.")
 

src/llmtuner/hparams/parser.py

@@ -8,7 +8,6 @@ import transformers
 from transformers import HfArgumentParser, Seq2SeqTrainingArguments
 from transformers.trainer_utils import get_last_checkpoint
 from transformers.utils import is_torch_bf16_gpu_available
-from transformers.utils.versions import require_version
 
 from ..extras.logging import get_logger
 from ..extras.misc import check_dependencies
@@ -119,6 +118,13 @@ def get_train_args(args: Optional[Dict[str, Any]] = None) -> _TRAIN_CLS:
     if finetuning_args.stage == "ppo" and finetuning_args.reward_model_type == "lora" and model_args.use_unsloth:
         raise ValueError("Unsloth does not support lora reward model.")
 
+    if (
+        finetuning_args.stage == "ppo"
+        and training_args.report_to is not None
+        and training_args.report_to[0] not in ["wandb", "tensorboard"]
+    ):
+        raise ValueError("PPO only accepts wandb or tensorboard logger.")
+
     if training_args.max_steps == -1 and data_args.streaming:
         raise ValueError("Please specify `max_steps` in streaming mode.")
 
@@ -128,12 +134,8 @@ def get_train_args(args: Optional[Dict[str, Any]] = None) -> _TRAIN_CLS:
     if training_args.do_train and model_args.use_unsloth and not is_unsloth_available():
         raise ValueError("Unsloth was not installed: https://github.com/unslothai/unsloth")
 
-    if finetuning_args.use_dora:
+    if finetuning_args.use_dora and model_args.use_unsloth:
-        if model_args.quantization_bit is not None:
+        raise ValueError("Unsloth does not support DoRA.")
-            require_version("peft>=0.10.0", "To fix: pip install peft>=0.10.0")
-
-        if model_args.use_unsloth:
-            raise ValueError("Unsloth does not support DoRA.")
 
     if finetuning_args.pure_bf16:
         if not is_torch_bf16_gpu_available():

src/llmtuner/model/patcher.py

@@ -173,7 +173,7 @@ def _configure_quantization(
     """
     if getattr(config, "quantization_config", None):  # ptq
         if is_deepspeed_zero3_enabled():
-            raise ValueError("DeepSpeed ZeRO-3 is incompatible with quantization.")
+            raise ValueError("DeepSpeed ZeRO-3 is incompatible with quantized models.")
 
         init_kwargs["device_map"] = {"": get_current_device()}
         quantization_config: Dict[str, Any] = getattr(config, "quantization_config", None)

src/llmtuner/model/utils.py

@@ -47,6 +47,8 @@ def find_all_linear_modules(model: "PreTrainedModel") -> List[str]:
     output_layer_names = ["lm_head"]
     if model.config.model_type == "chatglm":
         output_layer_names.append("output_layer")
+    elif model.config.model_type == "internlm2":
+        output_layer_names.append("output")
 
     module_names = set()
     for name, module in model.named_modules():

src/llmtuner/train/dpo/trainer.py

@@ -8,7 +8,7 @@ from trl import DPOTrainer
 from trl.trainer.utils import disable_dropout_in_model
 
 from ...extras.constants import IGNORE_INDEX
-from ..utils import create_custom_optimzer
+from ..utils import create_custom_optimzer, create_custom_scheduler
 
 
 if TYPE_CHECKING:
@@ -20,12 +20,9 @@ if TYPE_CHECKING:
 class CustomDPOTrainer(DPOTrainer):
     def __init__(
         self,
-        beta: float,
-        loss_type: Literal["sigmoid", "hinge", "ipo", "kto_pair"],
-        ftx_gamma: float,
         model: Union["PreTrainedModel", torch.nn.Module],
+        ref_model: Optional[Union["PreTrainedModel", torch.nn.Module]],
         finetuning_args: "FinetuningArguments",
-        ref_model: Optional[Union["PreTrainedModel", torch.nn.Module]] = None,
         disable_dropout: bool = True,
         **kwargs,
     ):
@@ -47,10 +44,10 @@ class CustomDPOTrainer(DPOTrainer):
         self._peft_has_been_casted_to_bf16 = False
 
         self.ref_model = ref_model
-        self.beta = beta
+        self.beta = finetuning_args.dpo_beta
-        self.label_smoothing = 0
+        self.label_smoothing = finetuning_args.dpo_label_smoothing
-        self.loss_type = loss_type
+        self.loss_type = finetuning_args.dpo_loss
-        self.ftx_gamma = ftx_gamma
+        self.ftx_gamma = finetuning_args.dpo_ftx
         self._stored_metrics = defaultdict(lambda: defaultdict(list))
 
         Trainer.__init__(self, model=model, **kwargs)
@@ -66,12 +63,16 @@ class CustomDPOTrainer(DPOTrainer):
             else:
                 self.ref_model = self.accelerator.prepare_model(self.ref_model, evaluation_mode=True)
 
-    def create_optimizer_and_scheduler(self, num_training_steps: int) -> None:
+    def create_optimizer(self) -> "torch.optim.Optimizer":
-        self.optimizer = create_custom_optimzer(self.model, self.args, self.finetuning_args, num_training_steps)
         if self.optimizer is None:
-            self.create_optimizer()
+            self.optimizer = create_custom_optimzer(self.model, self.args, self.finetuning_args)
+        return super().create_optimizer()
 
-        self.create_scheduler(num_training_steps=num_training_steps, optimizer=self.optimizer)
+    def create_scheduler(
+        self, num_training_steps: int, optimizer: Optional["torch.optim.Optimizer"] = None
+    ) -> "torch.optim.lr_scheduler.LRScheduler":
+        create_custom_scheduler(self.args, num_training_steps, optimizer)
+        return super().create_scheduler(num_training_steps, optimizer)
 
     def sft_loss(self, chosen_logits: torch.FloatTensor, chosen_labels: torch.LongTensor) -> torch.Tensor:
         r"""

src/llmtuner/train/dpo/workflow.py

@@ -28,6 +28,7 @@ def run_dpo(
     tokenizer = load_tokenizer(model_args)
     dataset = get_dataset(tokenizer, model_args, data_args, training_args, stage="rm")
     model = load_model(tokenizer, model_args, finetuning_args, training_args.do_train)
+
     data_collator = DPODataCollatorWithPadding(
         tokenizer=tokenizer,
         pad_to_multiple_of=8,
@@ -45,13 +46,10 @@ def run_dpo(
 
     # Initialize our Trainer
     trainer = CustomDPOTrainer(
-        beta=finetuning_args.dpo_beta,
-        loss_type=finetuning_args.dpo_loss,
-        ftx_gamma=finetuning_args.dpo_ftx,
-        finetuning_args=finetuning_args,
         model=model,
         ref_model=ref_model,
         args=training_args,
+        finetuning_args=finetuning_args,
         tokenizer=tokenizer,
         data_collator=data_collator,
         callbacks=callbacks,

src/llmtuner/train/ppo/trainer.py

@@ -6,20 +6,23 @@ from typing import TYPE_CHECKING, Dict, List, Optional, Tuple
 import torch
 from tqdm import tqdm
 from transformers import GenerationConfig, Trainer, TrainerControl, TrainerState
+from transformers.optimization import get_scheduler
 from transformers.trainer_pt_utils import remove_dummy_checkpoint
 from transformers.trainer_utils import PREFIX_CHECKPOINT_DIR
 from transformers.utils import SAFE_WEIGHTS_NAME, WEIGHTS_NAME
-from trl import PPOTrainer
+from trl import PPOConfig, PPOTrainer
 from trl.core import PPODecorators, logprobs_from_logits
 
 from ...extras.callbacks import FixValueHeadModelCallback, LogCallback
 from ...extras.logging import get_logger
 from ...extras.misc import AverageMeter, count_parameters, get_current_device, get_logits_processor
+from ..utils import create_custom_optimzer, create_custom_scheduler
 from .utils import dump_layernorm, get_rewards_from_server, replace_model, restore_layernorm
 
 
 if TYPE_CHECKING:
-    from transformers import Seq2SeqTrainingArguments, TrainerCallback
+    from datasets import Dataset
+    from transformers import DataCollatorWithPadding, PreTrainedTokenizer, Seq2SeqTrainingArguments, TrainerCallback
     from trl import AutoModelForCausalLMWithValueHead
 
     from ...hparams import FinetuningArguments, GeneratingArguments, ModelArguments
@@ -40,10 +43,53 @@ class CustomPPOTrainer(PPOTrainer, Trainer):
         finetuning_args: "FinetuningArguments",
         generating_args: "GeneratingArguments",
         callbacks: List["TrainerCallback"],
-        reward_model: "AutoModelForCausalLMWithValueHead",
+        model: "AutoModelForCausalLMWithValueHead",
-        **kwargs,
+        reward_model: Optional["AutoModelForCausalLMWithValueHead"],
+        ref_model: Optional["AutoModelForCausalLMWithValueHead"],
+        tokenizer: "PreTrainedTokenizer",
+        dataset: "Dataset",
+        data_collator: "DataCollatorWithPadding",
     ):
-        PPOTrainer.__init__(self, **kwargs)
+        backward_batch_size = training_args.per_device_train_batch_size * training_args.gradient_accumulation_steps
+        ppo_config = PPOConfig(
+            model_name=model_args.model_name_or_path,
+            learning_rate=training_args.learning_rate,
+            mini_batch_size=training_args.per_device_train_batch_size,
+            batch_size=backward_batch_size * finetuning_args.ppo_buffer_size,
+            gradient_accumulation_steps=training_args.gradient_accumulation_steps,
+            ppo_epochs=finetuning_args.ppo_epochs,
+            max_grad_norm=training_args.max_grad_norm,
+            seed=training_args.seed,
+            optimize_device_cache=True,
+            target=finetuning_args.ppo_target,
+            use_score_scaling=finetuning_args.ppo_score_norm,
+            use_score_norm=finetuning_args.ppo_score_norm,
+            whiten_rewards=finetuning_args.ppo_whiten_rewards,
+            accelerator_kwargs={"step_scheduler_with_optimizer": False},
+            log_with=training_args.report_to[0] if training_args.report_to is not None else None,
+            project_kwargs={"logging_dir": training_args.logging_dir},
+        )
+
+        # Create optimizer and scheduler
+        if training_args.max_steps > 0:
+            num_training_steps = training_args.max_steps
+        else:
+            total_train_batch_size = backward_batch_size * finetuning_args.ppo_buffer_size * training_args.world_size
+            num_training_steps = training_args.num_train_epochs * math.ceil(len(dataset) / total_train_batch_size)
+
+        optimizer = self.create_optimizer(model, training_args, finetuning_args)
+        scheduler = self.create_scheduler(training_args, num_training_steps, optimizer)
+
+        PPOTrainer.__init__(
+            self,
+            config=ppo_config,
+            model=model,
+            ref_model=ref_model,
+            tokenizer=tokenizer,
+            dataset=dataset,
+            data_collator=data_collator,
+            lr_scheduler=scheduler,
+        )
 
         self.args = training_args
         self.model_args = model_args
@@ -205,6 +251,44 @@ class CustomPPOTrainer(PPOTrainer, Trainer):
             self.args, self.state, self.control, model=self.accelerator.unwrap_model(self.model)
         )
 
+    def create_optimizer(
+        self,
+        model: "AutoModelForCausalLMWithValueHead",
+        training_args: "Seq2SeqTrainingArguments",
+        finetuning_args: "FinetuningArguments",
+    ) -> "torch.optim.Optimizer":
+        optimizer = create_custom_optimzer(model, training_args, finetuning_args)
+        if optimizer is None:
+            decay_params, nodecay_params = [], []
+            decay_param_names = self.get_decay_parameter_names(model)
+            for name, param in model.named_parameters():
+                if param.requires_grad:
+                    if name in decay_param_names:
+                        decay_params.append(param)
+                    else:
+                        nodecay_params.append(param)
+
+            optim_class, optim_kwargs = Trainer.get_optimizer_cls_and_kwargs(training_args)
+            param_groups = [
+                dict(params=nodecay_params),
+                dict(params=decay_params, weight_decay=training_args.weight_decay),
+            ]
+            optimizer = optim_class(param_groups, **optim_kwargs)
+
+        return optimizer
+
+    def create_scheduler(
+        self, training_args: "Seq2SeqTrainingArguments", num_training_steps: int, optimizer: "torch.optim.Optimizer"
+    ) -> "torch.optim.lr_scheduler.LRScheduler":
+        create_custom_scheduler(training_args, num_training_steps, optimizer)
+        lr_scheduler = get_scheduler(
+            training_args.lr_scheduler_type,
+            optimizer=optimizer,
+            num_warmup_steps=training_args.get_warmup_steps(num_training_steps),
+            num_training_steps=num_training_steps,
+        )
+        return lr_scheduler
+
     @torch.no_grad()
     def get_inputs(self, batch: Dict[str, torch.Tensor]) -> Tuple[List[torch.Tensor], List[torch.Tensor]]:
         r"""

src/llmtuner/train/ppo/workflow.py

@@ -1,19 +1,15 @@
 # Inspired by: https://github.com/lvwerra/trl/blob/main/examples/research_projects/stack_llama/scripts/rl_training.py
 
-import math
 from typing import TYPE_CHECKING, List, Optional
 
-from torch.optim import AdamW
 from transformers import DataCollatorWithPadding
-from transformers.optimization import get_scheduler
-from trl import PPOConfig
 
 from ...data import get_dataset
 from ...extras.callbacks import FixValueHeadModelCallback
 from ...extras.misc import fix_valuehead_checkpoint
 from ...extras.ploting import plot_loss
 from ...model import load_model, load_tokenizer
-from ..utils import create_custom_optimzer, create_ref_model, create_reward_model
+from ..utils import create_ref_model, create_reward_model
 from .trainer import CustomPPOTrainer
 
 
@@ -42,45 +38,6 @@ def run_ppo(
     ref_model = create_ref_model(model_args, finetuning_args, add_valuehead=True)
     reward_model = create_reward_model(model, model_args, finetuning_args)
 
-    # Create ppo config
-    backward_batch_size = training_args.per_device_train_batch_size * training_args.gradient_accumulation_steps
-    ppo_config = PPOConfig(
-        model_name=model_args.model_name_or_path,
-        learning_rate=training_args.learning_rate,
-        mini_batch_size=training_args.per_device_train_batch_size,
-        batch_size=backward_batch_size * finetuning_args.ppo_buffer_size,
-        gradient_accumulation_steps=training_args.gradient_accumulation_steps,
-        ppo_epochs=finetuning_args.ppo_epochs,
-        max_grad_norm=training_args.max_grad_norm,
-        seed=training_args.seed,
-        optimize_device_cache=True,
-        target=finetuning_args.ppo_target,
-        log_with=finetuning_args.ppo_logger,
-        use_score_scaling=finetuning_args.ppo_score_norm,
-        use_score_norm=finetuning_args.ppo_score_norm,
-        whiten_rewards=finetuning_args.ppo_whiten_rewards,
-        accelerator_kwargs={"step_scheduler_with_optimizer": False},
-        project_kwargs={"logging_dir": training_args.logging_dir},
-    )
-
-    # Create optimizer and scheduler
-    if training_args.max_steps > 0:
-        num_training_steps = training_args.max_steps
-    else:
-        total_train_batch_size = backward_batch_size * finetuning_args.ppo_buffer_size * training_args.world_size
-        num_training_steps = training_args.num_train_epochs * math.ceil(len(dataset) / total_train_batch_size)
-
-    optimizer = create_custom_optimzer(model, training_args, finetuning_args, num_training_steps)
-    if optimizer is None:
-        optimizer = AdamW(filter(lambda p: p.requires_grad, model.parameters()), lr=training_args.learning_rate)
-
-    lr_scheduler = get_scheduler(
-        training_args.lr_scheduler_type,
-        optimizer=optimizer,
-        num_warmup_steps=training_args.get_warmup_steps(num_training_steps),
-        num_training_steps=num_training_steps,
-    )
-
     # Initialize our Trainer
     ppo_trainer = CustomPPOTrainer(
         model_args=model_args,
@@ -88,15 +45,12 @@ def run_ppo(
         finetuning_args=finetuning_args,
         generating_args=generating_args,
         callbacks=callbacks + [FixValueHeadModelCallback()],
-        reward_model=reward_model,
-        config=ppo_config,
         model=model,
+        reward_model=reward_model,
         ref_model=ref_model,
         tokenizer=tokenizer,
         dataset=dataset,
         data_collator=data_collator,
-        optimizer=optimizer,
-        lr_scheduler=lr_scheduler,
     )
 
     # Training

src/llmtuner/train/pt/trainer.py

@@ -1,12 +1,14 @@
-from typing import TYPE_CHECKING
+from typing import TYPE_CHECKING, Optional
 
 from transformers import Trainer
 
 from ...extras.logging import get_logger
-from ..utils import create_custom_optimzer
+from ..utils import create_custom_optimzer, create_custom_scheduler
 
 
 if TYPE_CHECKING:
+    import torch
+
     from ...hparams import FinetuningArguments
 
 
@@ -22,9 +24,13 @@ class CustomTrainer(Trainer):
         super().__init__(**kwargs)
         self.finetuning_args = finetuning_args
 
-    def create_optimizer_and_scheduler(self, num_training_steps: int) -> None:
+    def create_optimizer(self) -> "torch.optim.Optimizer":
-        self.optimizer = create_custom_optimzer(self.model, self.args, self.finetuning_args, num_training_steps)
         if self.optimizer is None:
-            self.create_optimizer()
+            self.optimizer = create_custom_optimzer(self.model, self.args, self.finetuning_args)
+        return super().create_optimizer()
 
-        self.create_scheduler(num_training_steps=num_training_steps, optimizer=self.optimizer)
+    def create_scheduler(
+        self, num_training_steps: int, optimizer: Optional["torch.optim.Optimizer"] = None
+    ) -> "torch.optim.lr_scheduler.LRScheduler":
+        create_custom_scheduler(self.args, num_training_steps, optimizer)
+        return super().create_scheduler(num_training_steps, optimizer)

src/llmtuner/train/rm/trainer.py

@@ -1,12 +1,12 @@
 import json
 import os
-from typing import TYPE_CHECKING, Dict, List, Tuple, Union
+from typing import TYPE_CHECKING, Dict, List, Optional, Tuple, Union
 
 import torch
 from transformers import Trainer
 
 from ...extras.logging import get_logger
-from ..utils import create_custom_optimzer
+from ..utils import create_custom_optimzer, create_custom_scheduler
 
 
 if TYPE_CHECKING:
@@ -29,12 +29,16 @@ class PairwiseTrainer(Trainer):
         self.finetuning_args = finetuning_args
         self.can_return_loss = True  # override property to return eval_loss
 
-    def create_optimizer_and_scheduler(self, num_training_steps: int) -> None:
+    def create_optimizer(self) -> "torch.optim.Optimizer":
-        self.optimizer = create_custom_optimzer(self.model, self.args, self.finetuning_args, num_training_steps)
         if self.optimizer is None:
-            self.create_optimizer()
+            self.optimizer = create_custom_optimzer(self.model, self.args, self.finetuning_args)
+        return super().create_optimizer()
 
-        self.create_scheduler(num_training_steps=num_training_steps, optimizer=self.optimizer)
+    def create_scheduler(
+        self, num_training_steps: int, optimizer: Optional["torch.optim.Optimizer"] = None
+    ) -> "torch.optim.lr_scheduler.LRScheduler":
+        create_custom_scheduler(self.args, num_training_steps, optimizer)
+        return super().create_scheduler(num_training_steps, optimizer)
 
     def compute_loss(
         self, model: "PreTrainedModel", inputs: Dict[str, torch.Tensor], return_outputs: bool = False

src/llmtuner/train/sft/trainer.py

@@ -8,7 +8,7 @@ from transformers import Seq2SeqTrainer
 
 from ...extras.constants import IGNORE_INDEX
 from ...extras.logging import get_logger
-from ..utils import create_custom_optimzer
+from ..utils import create_custom_optimzer, create_custom_scheduler
 
 
 if TYPE_CHECKING:
@@ -29,12 +29,16 @@ class CustomSeq2SeqTrainer(Seq2SeqTrainer):
         super().__init__(**kwargs)
         self.finetuning_args = finetuning_args
 
-    def create_optimizer_and_scheduler(self, num_training_steps: int) -> None:
+    def create_optimizer(self) -> "torch.optim.Optimizer":
-        self.optimizer = create_custom_optimzer(self.model, self.args, self.finetuning_args, num_training_steps)
         if self.optimizer is None:
-            self.create_optimizer()
+            self.optimizer = create_custom_optimzer(self.model, self.args, self.finetuning_args)
+        return super().create_optimizer()
 
-        self.create_scheduler(num_training_steps=num_training_steps, optimizer=self.optimizer)
+    def create_scheduler(
+        self, num_training_steps: int, optimizer: Optional["torch.optim.Optimizer"] = None
+    ) -> "torch.optim.lr_scheduler.LRScheduler":
+        create_custom_scheduler(self.args, num_training_steps, optimizer)
+        return super().create_scheduler(num_training_steps, optimizer)
 
     def prediction_step(
         self,

src/llmtuner/train/utils.py

@@ -29,7 +29,13 @@ logger = get_logger(__name__)
 
 
 class DummyOptimizer(torch.optim.Optimizer):
-    def __init__(self, lr: float = 1e-3, optimizer_dict: Optional[dict] = None, *args, **kwargs) -> None:
+    r"""
+    A dummy optimizer used for the GaLore algorithm.
+    """
+
+    def __init__(
+        self, lr: float = 1e-3, optimizer_dict: Optional[Dict["torch.nn.Parameter", "torch.optim.Optimizer"]] = None
+    ) -> None:
         dummy_tensor = torch.randn(1, 1)
         self.optimizer_dict = optimizer_dict
         super().__init__([dummy_tensor], {"lr": lr})
@@ -64,7 +70,7 @@ def create_modelcard_and_push(
 
 def create_ref_model(
     model_args: "ModelArguments", finetuning_args: "FinetuningArguments", add_valuehead: bool = False
-) -> Union["PreTrainedModel", "AutoModelForCausalLMWithValueHead"]:
+) -> Optional[Union["PreTrainedModel", "AutoModelForCausalLMWithValueHead"]]:
     r"""
     Creates reference model for PPO/DPO training. Evaluation mode is not supported.
 
@@ -99,7 +105,7 @@ def create_ref_model(
 
 def create_reward_model(
     model: "AutoModelForCausalLMWithValueHead", model_args: "ModelArguments", finetuning_args: "FinetuningArguments"
-) -> "AutoModelForCausalLMWithValueHead":
+) -> Optional["AutoModelForCausalLMWithValueHead"]:
     r"""
     Creates reward model for PPO training.
     """
@@ -156,8 +162,9 @@ def _create_galore_optimizer(
     model: "PreTrainedModel",
     training_args: "Seq2SeqTrainingArguments",
     finetuning_args: "FinetuningArguments",
-    max_steps: int,
 ) -> "torch.optim.Optimizer":
+    require_version("galore_torch", "To fix: pip install galore_torch")
+
     if len(finetuning_args.galore_target) == 1 and finetuning_args.galore_target[0] == "all":
         galore_targets = find_all_linear_modules(model)
     else:
@@ -212,29 +219,19 @@ def _create_galore_optimizer(
         for param in decay_params:
             param_groups = [dict(params=[param], weight_decay=training_args.weight_decay)]
             optimizer_dict[param] = optim_class(param_groups, **optim_kwargs)
-        for param in galore_params:
+        for param in galore_params:  # galore params have weight decay
             param_groups = [dict(params=[param], weight_decay=training_args.weight_decay, **galore_kwargs)]
             optimizer_dict[param] = optim_class(param_groups, **optim_kwargs)
 
-        scheduler_dict: Dict["torch.Tensor", "torch.optim.lr_scheduler.LRScheduler"] = {}
+        def optimizer_hook(param: "torch.nn.Parameter"):
-        for param in trainable_params:
-            scheduler_dict[param] = get_scheduler(
-                training_args.lr_scheduler_type,
-                optimizer=optimizer_dict[param],
-                num_warmup_steps=training_args.get_warmup_steps(max_steps) * 2,
-                num_training_steps=max_steps * 2,
-            )
-
-        def optimizer_hook(param: "torch.Tensor"):
             if param.grad is not None:
                 optimizer_dict[param].step()
                 optimizer_dict[param].zero_grad()
-                scheduler_dict[param].step()
 
         for param in trainable_params:
             param.register_post_accumulate_grad_hook(optimizer_hook)
 
-        optimizer = DummyOptimizer(lr=training_args.learning_rate)  # display scheduler result
+        optimizer = DummyOptimizer(lr=training_args.learning_rate, optimizer_dict=optimizer_dict)
     else:
         param_groups = [
             dict(params=nodecay_params),
@@ -293,10 +290,34 @@ def create_custom_optimzer(
     model: "PreTrainedModel",
     training_args: "Seq2SeqTrainingArguments",
     finetuning_args: "FinetuningArguments",
-    max_steps: int,
 ) -> Optional["torch.optim.Optimizer"]:
     if finetuning_args.use_galore:
-        return _create_galore_optimizer(model, training_args, finetuning_args, max_steps)
+        return _create_galore_optimizer(model, training_args, finetuning_args)
 
     if finetuning_args.loraplus_lr_ratio is not None:
         return _create_loraplus_optimizer(model, training_args, finetuning_args)
+
+
+def create_custom_scheduler(
+    training_args: "Seq2SeqTrainingArguments",
+    num_training_steps: int,
+    optimizer: Optional["torch.optim.Optimizer"] = None,
+) -> None:
+    if optimizer is not None and isinstance(optimizer, DummyOptimizer):
+        optimizer_dict = optimizer.optimizer_dict
+        scheduler_dict: Dict["torch.nn.Parameter", "torch.optim.lr_scheduler.LRScheduler"] = {}
+
+        for param in optimizer_dict.keys():
+            scheduler_dict[param] = get_scheduler(
+                training_args.lr_scheduler_type,
+                optimizer=optimizer_dict[param],
+                num_warmup_steps=training_args.get_warmup_steps(num_training_steps) * 2,
+                num_training_steps=num_training_steps * 2,
+            )
+
+        def scheduler_hook(param: "torch.nn.Parameter"):
+            if param.grad is not None:
+                scheduler_dict[param].step()
+
+        for param in optimizer_dict.keys():
+            param.register_post_accumulate_grad_hook(scheduler_hook)