AI_Coder 模型
简介：本模型是基于swift框架，对qwen-7b-chat模型进行微调获得的。模型通过自我认知微调，能够正确输出自己的名字和开发者。同时，模型通过微调，能够输出格式正确且可以运行的、多语言的代码。
1.实验环境
swift框架 ubuntu 22.04 cuda 12.0.1 python 3.10 pytorch 2.1.2
2.训练方法
2.1数据集：
2.2微调的模型：
3.训练的步骤
3.1使用modelscope上的免费的GPU算力
3.2启动后进入命令行界面
3.3安装swift
通过以下命令安装swift
git clone https://github.com/modelscope/swift.git
cd swift
pip install -e .[llm]
4.微调代码
import os
os.environ['CUDA_VISIBLE_DEVICES'] = '0'
import torch
from swift.llm import (
DatasetName, InferArguments, ModelType, SftArguments,
infer_main, sft_main, app_ui_main, merge_lora_main
)
model_type = ModelType.qwen_1_8b_chat
sft_args = SftArguments(
model_type=model_type,
train_dataset_sample=2000,
dataset=[DatasetName.code_alpaca_en,DatasetName.leetcode_python_en],
output_dir='output',
model_name=['菜菜', 'caicai'],
model_author=['用AI发电', 'YAIE']
)
sft_args.model_name=['菜菜', 'caicai']
sft_args.model_author=['用AI发电', 'YAIE']
sft_args.self_cognition_sample=500
sft_args.lora_target_modules[0]='ALL'
print(f"sft_args.model_name-->{sft_args.model_name}")
print(f"sft_args.model_author-->{sft_args.model_author}")
result = sft_main(sft_args)
best_model_checkpoint = result['best_model_checkpoint']
print(f'best_model_checkpoint: {best_model_checkpoint}')
torch.cuda.empty_cache()
infer_args = InferArguments(
ckpt_dir=best_model_checkpoint,
show_dataset_sample=10)
# merge_lora_main(infer_args)
result = infer_main(infer_args)
5.示例代码
import os
os.environ['CUDA_VISIBLE_DEVICES'] = '0'
import torch
from swift.llm import (
DatasetName, InferArguments, ModelType, SftArguments,
infer_main, sft_main, app_ui_main, merge_lora_main
)
#best_model_checkpoint = "/mnt/workspace/kaka/output/qwen-7b-chat/v0-20240123-102915/checkpoint-100"
#best_model_checkpoint="/mnt/workspace/output/qwen-1_8b-chat/v0-20240124-165543/checkpoint-156"
best_model_checkpoint="/mnt/workspace/output/qwen-7b-chat/v0-20240124-163349/checkpoint-156"
print(f'best_model_checkpoint: {best_model_checkpoint}')
torch.cuda.empty_cache()
infer_args = InferArguments(
ckpt_dir=best_model_checkpoint,
show_dataset_sample=10)
# merge_lora_main(infer_args)
infer_args.system="You are a programming expert. The user will give some Chinese or English programming questions, and you need to carefully write code that can run, and the code you write needs to be included in ``` ```. For example, the user asks 'Write quicksort using Python.', and your answer is {```quicksort python code```}. In addition to writing code, you must explain your code in detail afterward, and try to provide comments in the code."
print(f"infer_args-->{infer_args.system}")
result = infer_main(infer_args)
torch.cuda.empty_cache()
6.结果展示
评论