千问 qwen7B 微调后的模型推理代码

 watch -n 1 nvidia-smi

1、数据处理代码

import ast

train_dataset = []

# 定义合法的字段列表
valid_fields = ["id", "conversations"]

with open('train.json', 'r', encoding="utf-8") as f:
    train_data = f.readlines()
for i, item in enumerate(train_data):
    json_data = ast.literal_eval(item)
    hh = {"id": i, "conversations": [{"from": "user", "value": json_data['instruction'].replace("'", "").replace('"', "") },
                                    {"from": "assistant", "value": json_data['output'].replace("'", "").replace('"', "") }]}

    train_dataset.append(hh)



with open("data.json", "w", encoding='utf-8') as f:
    f.write(",\n".join(str(x) for x in train_dataset))


# nohup bash finetune/finetune_lora_single_gpu.sh > /home/wangyp/Big_Model/train.log 2>&1 &

2、微调代码

https://github.com/QwenLM/

3、调用微调模型进行推理代码 

 CUDA_VISIBLE_DEVICES=0 python call_finetune.py

"""
调用微调以后的模型代码
"""


from peft import AutoPeftModelForCausalLM
from transformers import AutoTokenizer, AutoModelForCausalLM, GenerationConfig

model = AutoPeftModelForCausalLM.from_pretrained(
    '/home/wangyp/Big_Model/qwen/output_qwen', # path to the output directory
    device_map="auto",
    trust_remote_code=True
).eval()
merged_model = model.merge_and_unload()
merged_model.save_pretrained('/home/wangyp/Big_Model/qwen/qwen72b_sft', max_shard_size="2048MB", safe_serialization=True)


tokenizer = AutoTokenizer.from_pretrained("/home/wangyp/Big_Model/qwen/output_qwen", trust_remote_code=True)
tokenizer.save_pretrained("/home/wangyp/Big_Model/qwen/qwen72b_sft")
print("tikenizer 保存成功！")


query = "一名年龄在70岁的女性，出现了晕厥、不自主颤抖、情绪不稳等症状，请详细说明其手术治疗和术前准备。"

tokenizer = AutoTokenizer.from_pretrained("/home/wangyp/Big_Model/qwen/qwen72b_sft", trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained("/home/wangyp/Big_Model/qwen/qwen72b_sft", device_map="auto",trust_remote_code=True).eval()
response, history = model.chat(tokenizer, query, history=None)
print("微调后qwen答：" + response)

tokenizer = AutoTokenizer.from_pretrained("/mnt/sdd/Qwen-7B-Chat", trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained("/mnt/sdd/Qwen-7B-Chat", device_map="auto",trust_remote_code=True).eval()
response, history = model.chat(tokenizer, query, history=None)
print("官方qwen答：" + response)