Qwen大模型微调
在data/dataset_info.json添加自己的数据新增自己的数据，在webui界面可选训练数据集。微调后测试，可使用基座大模型和微调lora模型。模型合并同时也可运行合并后的模型测试。根据自己要微调的模型和数据进行选择。1.Qwen基本问答demo2。也可直接运行以下命令训练。编译llama.cpp。
·
1.Qwen基本问答demo2
from transformers import AutoModelForCausalLM, AutoTokenizer from peft import PeftModel model_name = "Qwen/Qwen2-7B-Instruct" device = "cuda" # the device to load the model onto model = AutoModelForCausalLM.from_pretrained( model_name, torch_dtype="auto", device_map="auto" ) tokenizer = AutoTokenizer.from_pretrained(model_name) prompt = "是否需要预约才能拜访楼上的公司？" messages = [ {"role": "system", "content": "You are a helpful assistant."}, {"role": "user", "content": prompt} ] text = tokenizer.apply_chat_template( messages, tokenize=False, add_generation_prompt=True ) model_inputs = tokenizer([text], return_tensors="pt").to(device) generated_ids = model.generate( **model_inputs, max_new_tokens=512 ) generated_ids = [ output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids) ] response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0] print("response:", response)
2.微调
python 3.11 CUDA 12.1
cd LLaMA-Factory
pip install -e ".[torch,metrics]"Use pip install --no-deps -e . to resolve package conflicts.
python src/webui.py根据自己要微调的模型和数据进行选择
微调数据格式
[
{
"instruction": "user instruction (required)",
"input": "user input (optional)",
"output": "model response (required)",
"system": "system prompt (optional)",
"history": [
["user instruction in the first round (optional)", "model response in the first round (optional)"],
["user instruction in the second round (optional)", "model response in the second round (optional)"]
]
}
]数据格式及代码
import pandas as pd
import json
# 读取 Excel 文件
excel_file_path = 'C:\\Users\\Administrator\\Desktop\\知识库V0.1-英文.xlsx'
df = pd.read_excel(excel_file_path)
# 假设 Excel 文件有两列：'Question' 和 'Answer'
# 如果列名不同，请相应地修改
questions = df['Question']
answers = df['Answer']
# 转换为 alpaca 格式
alpaca_data = []
for question, answer in zip(questions, answers):
alpaca_item = {
"instruction": question,
"input": "",
"output": answer,
"system": "",
"history": []
}
alpaca_data.append(alpaca_item)
# 将结果写入 JSON 文件
json_file_path = 'C:\\Users\\Administrator\\Desktop\\data.json'
with open(json_file_path, 'w', encoding='utf-8') as f:
json.dump(alpaca_data, f, ensure_ascii=False, indent=4)
print(f"转换完成，结果已保存到 {json_file_path}")在data/dataset_info.json添加自己的数据新增自己的数据，在webui界面可选训练数据集
“our_data": {
"file_name": "data.json"
},也可直接运行以下命令训练
llamafactory-cli train --stage sft --do_train True --model_name_or_path Qwen/Qwen1.5-7B-Chat --preprocessing_num_workers 16 --finetuning_type lora --template qwen --flash_attn auto --dataset_dir data --dataset our_data --cutoff_len 1024 --learning_rate 5e-05 --num_train_epochs 1000.0 --max_samples 100000 --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/Qwen1.5-7B-Chat/lora/train_2024-08-16-17-21-38 --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微调后测试，可使用基座大模型和微调lora模型
from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel
model_name = "Qwen/Qwen2-7B-Instruct"
#model_name = "/media/dgh/LLaMA-Factory/saves/Qwen2-7B-Chat/lora/train_2024-08-10-14-15-57/checkpoint-4400"
device = "cuda" # the device to load the model onto
model = AutoModelForCausalLM.from_pretrained(
model_name,
torch_dtype="auto",
device_map="auto"
)
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = PeftModel.from_pretrained(model, model_id = "/media/dgh/LLaMA-Factory/saves/Qwen2-7B-Chat/lora/train_2024-08-10-14-15-57/checkpoint-4400")
prompt = "是否需要预约才能拜访楼上的公司？"
messages = [
{"role": "system", "content": "You are a helpful assistant."},
{"role": "user", "content": prompt}
]
text = tokenizer.apply_chat_template(
messages,
tokenize=False,
add_generation_prompt=True
)
model_inputs = tokenizer([text], return_tensors="pt").to(device)
generated_ids = model.generate(
**model_inputs,
max_new_tokens=512
)
generated_ids = [
output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
]
response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
print("response:", response)
模型合并同时也可运行合并后的模型测试
CUDA_VISIBLE_DEVICES=0 llamafactory-cli export \
--model_name_or_path Qwen/Qwen1.5-7B-Chat\
--adapter_name_or_path /media/dgh/LLaMA-Factory/saves/Qwen1.5-7B-Chat/lora/train_2024-08-16-17-21-38/checkpoint-7300 \
--template qwen \
--finetuning_type lora \
--export_dir /media/dgh/Qwen2-main/save \
--export_size 2 \
--export_legacy_format False编译llama.cpp
git clone https://github.com/ggerganov/llama.cpp
cd llama.cpp
makeguuf格式生成
python convert_hf_to_gguf.py /media/dgh/Qwen2-main/save --outfile /media/dgh/Qwen2-main/7b_guuf/qwen2-7b-instruct-fp16.gguf量化GGUF模型： q2_k 、 q3_k_m 、 q4_0 、 q4_k_m 、 q5_0 、 q5_k_m 、 q6_k 和 q8_0 。 了解更多信息，请访问 llama.cpp 。
q4_0
./llama-quantize /home/dgh/LLaMA-Factory/saves/Qwen1.5-14b_guuf/Qwen1.5-14B-Chat-F16.gguf /home/dgh/LLaMA-Factory/saves/qwen1.5-14b-q4_0_gguf/qwen1.5-14b-q4_0.gguf q4_0 q5_k_m
./llama-quantize /home/dgh/LLaMA-Factory/saves/Qwen1.5-14b_guuf/Qwen1.5-14B-Chat-F16.gguf /home/dgh/LLaMA-Factory/saves/qwen1.5-14b-q5_k_m_gguf/qwen1.5-14b-q5_k_m.gguf q5_k_mrun
./llama-cli -m /home/dgh/LLaMA-Factory/saves/qwen1.5-14b-q5_k_m_gguf/qwen1.5-14b-q5_k_m.gguf \
-n 512 -co -i -if -f prompts/chat-with-qwen.txt \
--in-prefix "<|im_start|>user\n" \
--in-suffix "<|im_end|>\n<|im_start|>assistant\n" \
-ngl 80 -fa
阿里百炼在线大模型调用大模型服务平台百炼
在阿里百炼可以零代码创建自己的智能体，其中调用API如下：
import os
from openai import OpenAI
client = OpenAI(
api_key=os.getenv("DASHSCOPE_API_KEY"),
base_url="https://dashscope.aliyuncs.com/compatible-mode/v1",
)
completion = client.chat.completions.create(
model="qwen-turbo",
messages=[
{'role': 'system', 'content': 'You are a helpful assistant.'},
{'role': 'user', 'content': '你是谁？'}
],
temperature=0.8
)
print(completion.choices[0].message.content)创建自己的智能体也可根据给定API调用，具体参考官方使用指导教程
下载最新的LLaMA-Factory 支持微调Qwen-VL
遇到报错：
ValueError: The checkpoint you are trying to load has model type `qwen2_vl` but Transformers does not recognize this architecture. This could be because of an issue with the checkpoint, or because your version of Transformers is out of date
原因：pip中transformers还没有更新，用github上的版本更新
解决方法：pip install git+https://github.com/huggingface/transformers
采用swift框架微调Qwen2-VL
使用ModelScope CLI下载模型
modelscope download --model=qwen/Qwen2-VL-7B-Instruct --local_dir ./Qwen2-VL-7B-Instruct采用的是python 3.8。CUDA 12.1
git clone https://github.com/modelscope/swift.git
cd swift
pip install -e .[llm]
pip install pyav qwen_vl_utils采用官方指定数据集训练
CUDA_VISIBLE_DEVICES=0,1,2,3 NPROC_PER_NODE=4 swift sft \
--model_type qwen2-vl-7b-instruct \
--model_id_or_path qwen/Qwen2-VL-7B-Instruct \
--sft_type lora \
--dataset coco-en-mini#20000 \
--deepspeed default-zero2采用自己的数据集训练
--dataset train.jsonl \
--val_dataset val.jsonl \数据集格式
{"query": "<image>55555", "response": "66666", "images": ["image_path"]}
{"query": "eeeee<image>eeeee<image>eeeee", "response": "fffff", "history": [], "images": ["image_path1", "image_path2"]}
{"query": "EEEEE", "response": "FFFFF", "history": [["query1", "response2"], ["query2", "response2"]], "images": []}微调后推理，并合并模型
CUDA_VISIBLE_DEVICES=0 swift infer \
--ckpt_dir output/qwen2-vl-7b-instruct/vx-xxx/checkpoint-xxx \
--load_dataset_config true --merge_lora true合并之后的模型可直接调用
from PIL import Image
import torch
from transformers import Qwen2VLForConditionalGeneration, AutoTokenizer, AutoProcessor
from qwen_vl_utils import process_vision_info
from modelscope import snapshot_download
model_dir = "/Qwen2-VL-2B-Instruct/output/v4-20240923/checkpoint-1000-merged"
# Load the model in half-precision on the available device(s)
model = Qwen2VLForConditionalGeneration.from_pretrained(model_dir, device_map="auto", torch_dtype = torch.float16)
min_pixels = 256*28*28
max_pixels = 1280*28*28
processor = AutoProcessor.from_pretrained(model_dir, min_pixels=min_pixels, max_pixels=max_pixels)
messages = [{"role": "user", "content": [{"type": "image", "image": "https://qianwen-res.oss-cn-beijing.aliyuncs.com/Qwen-VL/assets/demo.jpeg"}, {"type": "text", "text": "Describe this image."}]}]
# Preparation for inference
text = processor.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
image_inputs, video_inputs = process_vision_info(messages)
inputs = processor(text=[text], images=image_inputs, videos=video_inputs, padding=True, return_tensors="pt")
inputs = inputs.to('cuda')
# Inference: Generation of the output
generated_ids = model.generate(**inputs, max_new_tokens=128)
generated_ids_trimmed = [out_ids[len(in_ids):] for in_ids, out_ids in zip(inputs.input_ids, generated_ids)]
output_text = processor.batch_decode(generated_ids_trimmed, skip_special_tokens=True, clean_up_tokenization_spaces=False)
print(output_text)多图推理
# Messages containing multiple images and a text query
messages = [{"role": "user", "content": [{"type": "image", "image": "file:///path/to/image1.jpg"}, {"type": "image", "image": "file:///path/to/image2.jpg"}, {"type": "text", "text": "Identify the similarities between these images."}]}]
# Preparation for inference
text = processor.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
image_inputs, video_inputs = process_vision_info(messages)
inputs = processor(text=[text], images=image_inputs, videos=video_inputs, padding=True, return_tensors="pt")
inputs = inputs.to('cuda')
# Inference
generated_ids = model.generate(**inputs, max_new_tokens=128)
generated_ids_trimmed = [out_ids[len(in_ids):] for in_ids, out_ids in zip(inputs.input_ids, generated_ids)]
output_text = processor.batch_decode(generated_ids_trimmed, skip_special_tokens=True, clean_up_tokenization_spaces=False)
print(output_text)视频理解
# Messages containing a video and a text query
messages = [{"role": "user", "content": [{"type": "video", "video": "file:///path/to/video1.mp4", 'max_pixels': 360*420, 'fps': 1.0}, {"type": "text", "text": "Describe this video."}]}]
# Preparation for inference
text = processor.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
image_inputs, video_inputs = process_vision_info(messages)
inputs = processor(text=[text], images=image_inputs, videos=video_inputs, padding=True, return_tensors="pt")
inputs = inputs.to('cuda')
# Inference
generated_ids = model.generate(**inputs, max_new_tokens=128)
generated_ids_trimmed = [out_ids[len(in_ids):] for in_ids, out_ids in zip(inputs.input_ids, generated_ids)]
output_text = processor.batch_decode(generated_ids_trimmed, skip_special_tokens=True, clean_up_tokenization_spaces=False)
print(output_text)在线调用视觉大模型API
from openai import OpenAI
import os
import base64
# base 64 编码格式
def encode_image(image_path):
with open(image_path, "rb") as image_file:
return base64.b64encode(image_file.read()).decode('utf-8')
def get_response(image_path):
base64_image = encode_image(image_path)
client = OpenAI(
api_key=os.getenv("DASHSCOPE_API_KEY"),
base_url="https://dashscope.aliyuncs.com/compatible-mode/v1",
)
completion = client.chat.completions.create(
model="qwen-vl-max",
messages=[
{
"role": "user",
"content": [
{
"type": "image_url",
"image_url": {
"url": f"data:image/jpeg;base64,{base64_image}"
}
},
{
"type": "text",
"text": "描述一下。"
}
]
}
]
)
print(completion.model_dump_json())
if __name__=='__main__':
import time
start = time.time()
get_response("./26.jpg")
end = time.time()
print("total-times:", end - start)本地多卡调用修改代码如下
from PIL import Image
import torch
from transformers import Qwen2VLForConditionalGeneration, AutoTokenizer, AutoProcessor
from qwen_vl_utils import process_vision_info
from modelscope import snapshot_download
from torch.nn import DataParallel
model_dir = "/media/checkpoint-1000-merged"
# Load the model in half-precision on the available device(s)
model = Qwen2VLForConditionalGeneration.from_pretrained(model_dir, device_map="auto", torch_dtype = torch.float16)
# 确保模型在主GPU上
# model.to('cuda:0')
# 使用DataParallel包装模型
model = DataParallel(model) #多卡
min_pixels = 256*28*28
max_pixels = 1280*28*28
processor = AutoProcessor.from_pretrained(model_dir, min_pixels=min_pixels, max_pixels=max_pixels)
messages = [{"role": "user", "content": [{"type": "image", "image": "/media/normal/1.jpg"}, {"type": "text", "text": "描述一下"}]}]
# messages = [{"role": "user", "content": [{"type": "video", "video": "/Qwen2-VL/1.mp4", 'max_pixels': 360*420, 'fps': 1.0}, {"type": "text", "text": "描述一下"}]}]
# Preparation for inference
text = processor.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
image_inputs, video_inputs = process_vision_info(messages)
inputs = processor(text=[text], images=image_inputs, videos=video_inputs, padding=True, return_tensors="pt")
inputs = inputs.to('cuda')
# Inference: Generation of the output
# generated_ids = model.generate(**inputs, max_new_tokens=128) #多卡修改如下
generated_ids = model.module.generate(**inputs, max_new_tokens=128)
generated_ids_trimmed = [out_ids[len(in_ids):] for in_ids, out_ids in zip(inputs.input_ids, generated_ids)]
output_text = processor.batch_decode(generated_ids_trimmed, skip_special_tokens=True, clean_up_tokenization_spaces=False)
print(output_text)指定epoch微调
CUDA_VISIBLE_DEVICES=0,1,2 NPROC_PER_NODE=3 swift sft --model_type qwen2-vl-7b-instruct --model_id_or_path /media/dgh/dgh/qwen2-vl-server/Qwen2-VL-2B-Instruct --sft_type lora --dataset coco-en-mini#20000 --deepspeed default-zero2 --num_train_epochs 2在archlinux系统，cuda12.4，python=3.10的环境
absl-py 2.1.0
accelerate 1.1.1
addict 2.4.0
aiofiles 23.2.1
aiohappyeyeballs 2.4.4
aiohttp 3.11.9
aiosignal 1.3.1
aliyun-python-sdk-core 2.16.0
aliyun-python-sdk-kms 2.16.5
annotated-types 0.7.0
anyio 4.6.2.post1
async-timeout 5.0.1
attrdict 2.0.1
attrs 24.2.0
auto_gptq 0.7.1
av 14.0.0
binpacking 1.5.2
certifi 2024.8.30
cffi 1.17.1
charset-normalizer 3.4.0
click 8.1.7
coloredlogs 15.0.1
contourpy 1.3.1
cpm-kernels 1.0.11
crcmod 1.7
cryptography 44.0.0
cycler 0.12.1
dacite 1.8.1
datasets 3.0.1
deepspeed 0.15.0
dill 0.3.8
distro 1.9.0
docstring_parser 0.16
einops 0.8.0
exceptiongroup 1.2.2
fastapi 0.115.5
ffmpy 0.4.0
filelock 3.16.1
fonttools 4.55.1
frozenlist 1.5.0
fsspec 2024.6.1
future 1.0.0
gekko 1.2.1
gradio 5.7.1
gradio_client 1.5.0
grpcio 1.68.1
h11 0.14.0
hjson 3.1.0
httpcore 1.0.7
httpx 0.28.0
huggingface-hub 0.26.3
humanfriendly 10.0
idna 3.10
importlib_metadata 8.5.0
jieba 0.42.1
Jinja2 3.1.4
jiter 0.8.0
jmespath 0.10.0
joblib 1.4.2
kiwisolver 1.4.7
Markdown 3.7
markdown-it-py 3.0.0
MarkupSafe 2.1.5
matplotlib 3.9.3
mdurl 0.1.2
modelscope 1.21.0
mpmath 1.3.0
ms-swift 2.6.1
msgpack 1.1.0
multidict 6.1.0
multiprocess 0.70.16
networkx 3.4.2
ninja 1.11.1.2
nltk 3.9.1
numpy 1.26.4
nvidia-cublas-cu12 12.1.3.1
nvidia-cuda-cupti-cu12 12.1.105
nvidia-cuda-nvrtc-cu12 12.1.105
nvidia-cuda-runtime-cu12 12.1.105
nvidia-cudnn-cu12 9.1.0.70
nvidia-cufft-cu12 11.0.2.54
nvidia-curand-cu12 10.3.2.106
nvidia-cusolver-cu12 11.4.5.107
nvidia-cusparse-cu12 12.1.0.106
nvidia-ml-py 12.560.30
nvidia-nccl-cu12 2.20.5
nvidia-nvjitlink-cu12 12.4.127
nvidia-nvtx-cu12 12.1.105
openai 1.56.0
optimum 1.23.3
orjson 3.10.12
oss2 2.19.1
packaging 24.2
pandas 2.2.3
peft 0.12.0
pillow 11.0.0
pip 24.2
propcache 0.2.1
protobuf 5.29.0
psutil 6.1.0
py-cpuinfo 9.0.0
pyarrow 18.1.0
pyav 14.0.0
pycparser 2.22
pycryptodome 3.21.0
pydantic 2.10.2
pydantic_core 2.27.1
pydub 0.25.1
Pygments 2.18.0
pyparsing 3.2.0
python-dateutil 2.9.0.post0
python-multipart 0.0.12
pytz 2024.2
PyYAML 6.0.2
qwen-vl-utils 0.0.8
regex 2024.11.6
requests 2.32.3
rich 13.9.4
rouge 1.0.1
ruff 0.8.1
safehttpx 0.1.6
safetensors 0.4.5
scipy 1.14.1
semantic-version 2.10.0
sentencepiece 0.2.0
setuptools 69.5.1
shellingham 1.5.4
shtab 1.7.1
simplejson 3.19.3
six 1.16.0
sniffio 1.3.1
sortedcontainers 2.4.0
starlette 0.41.3
sympy 1.13.1
tensorboard 2.18.0
tensorboard-data-server 0.7.2
tiktoken 0.8.0
tokenizers 0.20.3
tomlkit 0.12.0
torch 2.4.0
torchaudio 2.5.1+cu124
torchvision 0.19.0
tqdm 4.67.1
transformers 4.45.2
transformers-stream-generator 0.0.5
triton 3.0.0
trl 0.11.4
typeguard 4.4.1
typer 0.14.0
typing_extensions 4.12.2
tyro 0.9.2
tzdata 2024.2
urllib3 2.2.3
uvicorn 0.32.1
websockets 12.0
Werkzeug 3.1.3
wheel 0.44.0
xxhash 3.5.0
yarl 1.18.3
zipp 3.21.0VL模型量化
pip install auto-gptq
CUDA_VISIBLE_DEVICES=0,1,2 swift export --ckpt_dir /media/swift/output/qwen2-vl-2b-instruct/v0-20241213-145017/checkpoint-400/ --merge_lora true --quant_bits 8 --load_dataset_config true --quant_method gptq若出现报错
File "/home/dengguanghong/.conda/envs/swift2/lib/python3.10/site-packages/transformers/models/qwen2_vl/modeling_qwen2_vl.py", line 183, in forward
freqs = (inv_freq_expanded.float() @ position_ids_expanded.float()).transpose(2, 3)
RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cpu and cuda:0! (when checking argument for argument mat2 in method wrapper_CUDA_bmm)运行指令
pip install ms-swift -U
ModelScope旨在打造下一代开源的模型即服务共享平台，为泛AI开发者提供灵活、易用、低成本的一站式模型服务产品，让模型应用更简单！
更多推荐
8
0- 0
已为社区贡献1条内容
所有评论(0)