From e93e9641f52bdc88e2a7f81202ae6edd9ccc1795 Mon Sep 17 00:00:00 2001 From: hiyouga Date: Fri, 18 Aug 2023 01:51:55 +0800 Subject: [PATCH] update readme Former-commit-id: e4eec9ddfd3a9688733e018a96274dff0d5d9962 --- README.md | 77 +++++++++++++++++++++++++++------------------------- README_zh.md | 74 ++++++++++++++++++++++++++------------------------ 2 files changed, 78 insertions(+), 73 deletions(-) diff --git a/README.md b/README.md index a93a82f0..ce188b20 100644 --- a/README.md +++ b/README.md @@ -164,7 +164,9 @@ We strongly recommend using the all-in-one Web UI for newcomers since it can als Currently the web UI only supports training on **a single GPU**. -### Pre-Training +### Train on a single GPU + +#### Pre-Training ```bash CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \ @@ -187,7 +189,7 @@ CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \ --fp16 ``` -### Supervised Fine-Tuning +#### Supervised Fine-Tuning ```bash CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \ @@ -210,7 +212,7 @@ CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \ --fp16 ``` -### Reward Modeling +#### Reward Modeling ```bash CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \ @@ -234,7 +236,7 @@ CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \ --fp16 ``` -### PPO Training +#### PPO Training ```bash CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \ @@ -255,10 +257,11 @@ CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \ --save_steps 1000 \ --learning_rate 1e-5 \ --num_train_epochs 1.0 \ - --plot_loss + --plot_loss \ + --fp16 ``` -### DPO Training +#### DPO Training ```bash CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \ @@ -357,40 +360,15 @@ deepspeed --num_gpus 8 --master_port=9901 src/train_bash.py \ -### Evaluation (BLEU and ROUGE_CHINESE) +### Export model ```bash -CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \ - --stage sft \ +python src/export_model.py \ --model_name_or_path path_to_your_model \ - --do_eval \ - --dataset alpaca_gpt4_en \ --template default \ --finetuning_type lora \ --checkpoint_dir path_to_checkpoint \ - --output_dir path_to_eval_result \ - --per_device_eval_batch_size 8 \ - --max_samples 100 \ - --predict_with_generate -``` - -We recommend using `--per_device_eval_batch_size=1` and `--max_target_length 128` at 4/8-bit evaluation. - -### Predict - -```bash -CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \ - --stage sft \ - --model_name_or_path path_to_your_model \ - --do_predict \ - --dataset alpaca_gpt4_en \ - --template default \ - --finetuning_type lora \ - --checkpoint_dir path_to_checkpoint \ - --output_dir path_to_predict_result \ - --per_device_eval_batch_size 8 \ - --max_samples 100 \ - --predict_with_generate + --output_dir path_to_export ``` ### API Demo @@ -425,15 +403,40 @@ python src/web_demo.py \ --checkpoint_dir path_to_checkpoint ``` -### Export model +### Evaluation (BLEU and ROUGE_CHINESE) ```bash -python src/export_model.py \ +CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \ + --stage sft \ --model_name_or_path path_to_your_model \ + --do_eval \ + --dataset alpaca_gpt4_en \ --template default \ --finetuning_type lora \ --checkpoint_dir path_to_checkpoint \ - --output_dir path_to_export + --output_dir path_to_eval_result \ + --per_device_eval_batch_size 8 \ + --max_samples 100 \ + --predict_with_generate +``` + +We recommend using `--per_device_eval_batch_size=1` and `--max_target_length 128` at 4/8-bit evaluation. + +### Predict + +```bash +CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \ + --stage sft \ + --model_name_or_path path_to_your_model \ + --do_predict \ + --dataset alpaca_gpt4_en \ + --template default \ + --finetuning_type lora \ + --checkpoint_dir path_to_checkpoint \ + --output_dir path_to_predict_result \ + --per_device_eval_batch_size 8 \ + --max_samples 100 \ + --predict_with_generate ``` ## TODO diff --git a/README_zh.md b/README_zh.md index fe96c55a..08cd4504 100644 --- a/README_zh.md +++ b/README_zh.md @@ -164,7 +164,9 @@ CUDA_VISIBLE_DEVICES=0 python src/train_web.py 目前网页 UI 仅支持**单卡训练**。 -### 预训练 +### 单 GPU 训练 + +#### 预训练 ```bash CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \ @@ -187,7 +189,7 @@ CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \ --fp16 ``` -### 指令监督微调 +#### 指令监督微调 ```bash CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \ @@ -210,7 +212,7 @@ CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \ --fp16 ``` -### 奖励模型训练 +#### 奖励模型训练 ```bash CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \ @@ -234,7 +236,7 @@ CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \ --fp16 ``` -### PPO 训练 +#### PPO 训练 ```bash CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \ @@ -258,7 +260,7 @@ CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \ --plot_loss ``` -### DPO 训练 +#### DPO 训练 ```bash CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \ @@ -357,40 +359,15 @@ deepspeed --num_gpus 8 --master_port=9901 src/train_bash.py \ -### 指标评估（BLEU 分数和汉语 ROUGE 分数） +### 导出微调后的模型 ```bash -CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \ - --stage sft \ +python src/export_model.py \ --model_name_or_path path_to_your_model \ - --do_eval \ - --dataset alpaca_gpt4_zh \ --template default \ --finetuning_type lora \ --checkpoint_dir path_to_checkpoint \ - --output_dir path_to_eval_result \ - --per_device_eval_batch_size 8 \ - --max_samples 100 \ - --predict_with_generate -``` - -我们建议在量化模型的评估中使用 `--per_device_eval_batch_size=1` 和 `--max_target_length 128`。 - -### 模型预测 - -```bash -CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \ - --stage sft \ - --model_name_or_path path_to_your_model \ - --do_predict \ - --dataset alpaca_gpt4_zh \ - --template default \ - --finetuning_type lora \ - --checkpoint_dir path_to_checkpoint \ - --output_dir path_to_predict_result \ - --per_device_eval_batch_size 8 \ - --max_samples 100 \ - --predict_with_generate + --output_dir path_to_export ``` ### API 服务 @@ -425,15 +402,40 @@ python src/web_demo.py \ --checkpoint_dir path_to_checkpoint ``` -### 导出微调模型 +### 指标评估（BLEU 分数和汉语 ROUGE 分数） ```bash -python src/export_model.py \ +CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \ + --stage sft \ --model_name_or_path path_to_your_model \ + --do_eval \ + --dataset alpaca_gpt4_zh \ --template default \ --finetuning_type lora \ --checkpoint_dir path_to_checkpoint \ - --output_dir path_to_export + --output_dir path_to_eval_result \ + --per_device_eval_batch_size 8 \ + --max_samples 100 \ + --predict_with_generate +``` + +我们建议在量化模型的评估中使用 `--per_device_eval_batch_size=1` 和 `--max_target_length 128`。 + +### 模型预测 + +```bash +CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \ + --stage sft \ + --model_name_or_path path_to_your_model \ + --do_predict \ + --dataset alpaca_gpt4_zh \ + --template default \ + --finetuning_type lora \ + --checkpoint_dir path_to_checkpoint \ + --output_dir path_to_predict_result \ + --per_device_eval_batch_size 8 \ + --max_samples 100 \ + --predict_with_generate ``` ## TODO