Qwen2 大模型

LLaMA-Factory微调 lora 模型

工具依赖：llamafactory-0.8.2

llamafactory-cli webui 可视化微调、训练，预测启动命令0

Running on local URL: http://0.0.0.0:7860

定义名字

import hashlib
 
def calculate_sha1(file_path):
    sha1 = hashlib.sha1()
    try:
        with open(file_path, 'rb') as file:
            while True:
                data = file.read(8192)  # Read in chunks to handle large files
                if not data:
                    break
                sha1.update(data)
        return sha1.hexdigest()
    except FileNotFoundError:
        return "File not found."
 
# 使用示例
file_path = r'/home/jm/data/LLaMA-Factory-main/data/train_data.json'  # 替换为您的文件路径
sha1_hash = calculate_sha1(file_path)
print("SHA-1 Hash:", sha1_hash)

llamafactory-cli train \
    --stage sft \
    --do_train True \
    --model_name_or_path /home/jm/data/model/qwen/Qwen2-7B-Instruct-GPTQ-Int4 \
    --preprocessing_num_workers 16 \
    --finetuning_type lora \
    --quantization_method bitsandbytes \ # 注意这个参数，有些服务器上不支持
    --template qwen \
    --flash_attn auto \
    --dataset_dir data \
    --dataset identity \
    --cutoff_len 1024 \
    --learning_rate 1e-05 \
    --num_train_epochs 3.0 \
    --max_samples 10000 \
    --per_device_train_batch_size 2 \
    --gradient_accumulation_steps 8 \
    --lr_scheduler_type cosine \
    --max_grad_norm 1.0 \
    --logging_steps 5 \
    --save_steps 100 \
    --warmup_steps 0 \
    --optim adamw_torch \
    --packing False \
    --report_to none \
    --output_dir saves/Qwen2-7B-int4-Chat/lora/train_2024-07-08-08-48-52 \
    --bf16 True \
    --plot_loss True \
    --ddp_timeout 180000000 \
    --include_num_input_tokens_seen True \
    --lora_rank 8 \
    --lora_alpha 16 \
    --lora_dropout 0 \
    --lora_target all \
    --deepspeed cache/ds_z3_config.json

llamafactory-cli train \
    --stage sft \
    --model_name_or_path /home/jm/data/model/qwen/Qwen2-7B-Instruct-GPTQ-Int4 \
    --preprocessing_num_workers 16 \
    --finetuning_type lora \
    --quantization_method bitsandbytes \ # 注意这个参数，有些服务器上不支持
    --template qwen \
    --flash_attn fa2 \
    --dataset_dir data \
    --dataset identity \
    --cutoff_len 1024 \
    --max_samples 100000 \
    --per_device_eval_batch_size 10 \
    --predict_with_generate True \
    --max_new_tokens 512 \
    --top_p 0.7 \
    --temperature 0.1 \
    --output_dir saves/Qwen2-7B-int4-Chat/lora/eval_2024-07-08-10-48-52 \
    --do_predict True \
    --adapter_name_or_path saves/Qwen2-7B-int4-Chat/lora/train_2024-07-08-08-48-52

python -m vllm.entrypoints.openai.api_server \
--model /home/jm/data/model/qwen/Qwen2-7B-Instruct-GPTQ-Int4/ \
--enable-lora \
--lora-modules lora1=/home/jm/data/LLaMA-Factory-main/saves/Qwen2-7B-int4-Chat/lora/train_2024-07-09-14-50-10/ \
-tp 2 \
--enforce-eager

from openai import OpenAI
from openai import OpenAI
# Set OpenAI's API key and API base to use vLLM's API server.
openai_api_key = "EMPTY"
openai_api_base = "http://127.0.0.1:8000/v1"
client = OpenAI(
    api_key=openai_api_key,
    base_url=openai_api_base,
)
chat_response = client.chat.completions.create(
    model="lora1",
    messages=[
        {"role": "system", "content": "你是一个文本分类领域的专家,你会收到一段文本和几个潜在的分类选项,请选择问题最接近的一个分类选项,潜在的分类选项:农业/租赁法律/交通运输仓储邮政/水利环境/采矿/信息软件/政府组织/国际组织/房地产/制造业/文体娱乐/金融/电力燃气水生产/教育/住宿餐饮/科学技术/卫生医疗/建筑。只输出结果不做任何的解释说明："},
        {"role": "user", "content": "白条乌鸡价格 北京新发地农副产品批发市场信息中心 ￥13.20 ￥13.40 ￥13.30 元/公斤 查找"},
    ]
)
print("Chat response:", chat_response.choices[0].message.content.strip())

llamafactory-cli