cloud studio 使用 unsloth 微调 qwen2.5 模型

本教程使用到 conda 安装虚拟环境，而 cloud studio 容器中默认已经安装好了，并且使用到 jupyter notebook 进行演示，如果不知道如何安装环境，可以参考 Cloud Studio 搭建 anaconda 环境 安装 jupyter notebook

安装运行环境

# 安装软件包
conda create --name unsloth_env \
    python=3.11 \
    pytorch-cuda=12.1 \
    pytorch cudatoolkit xformers -c pytorch -c nvidia -c xformers \
    -y

# 激活虚拟环境
conda activate unsloth_env

# 安装 unsloth
pip install unsloth

下载 qwen2.5 模型

这里介绍通过离线下载方式，先将 qwen2.5 7b 模型先下载到本地。使用的是 Huggingface-cli, 它是 Hugging Face 官方提供的命令行工具，支持下载功能。

使用的是 unsloth 官方提供的模型
https://huggingface.co/unsloth/Qwen2.5-7B

# 安装依赖
pip install -U huggingface_hub

# 设置环境变量
export HF_ENDPOINT=https://hf-mirror.com

# 下载模型
huggingface-cli download --resume-download unsloth/Qwen2.5-7B --local-dir Qwen2.5-7B

加载训练模型

from unsloth import FastLanguageModel
import torch
max_seq_length = 2048 # Choose any! We auto support RoPE Scaling internally!
dtype = None # None for auto detection. Float16 for Tesla T4, V100, Bfloat16 for Ampere+
load_in_4bit = True # Use 4bit quantization to reduce memory usage. Can be False.

model, tokenizer = FastLanguageModel.from_pretrained(
    # Can select any from the below:
    # "unsloth/Qwen2.5-0.5B", "unsloth/Qwen2.5-1.5B", "unsloth/Qwen2.5-3B"
    # "unsloth/Qwen2.5-14B",  "unsloth/Qwen2.5-32B",  "unsloth/Qwen2.5-72B",
    # And also all Instruct versions and Math. Coding verisons!
    # 这里是远程自动下载方式
    #model_name = "unsloth/Qwen2.5-7B",
    # 这里是离线下载的本地路径
    model_name = "/workspace/model/Qwen2.5-7B",
    max_seq_length = max_seq_length,
    dtype = dtype,
    load_in_4bit = load_in_4bit,
    # token = "hf_...", # use one if using gated models like meta-llama/Llama-2-7b-hf
)

添加 LoRA 适配器

model = FastLanguageModel.get_peft_model(
    model,
    r = 16, # Choose any number > 0 ! Suggested 8, 16, 32, 64, 128
    target_modules = ["q_proj", "k_proj", "v_proj", "o_proj",
                      "gate_proj", "up_proj", "down_proj",],
    lora_alpha = 16,
    lora_dropout = 0, # Supports any, but = 0 is optimized
    bias = "none",    # Supports any, but = "none" is optimized
    # [NEW] "unsloth" uses 30% less VRAM, fits 2x larger batch sizes!
    use_gradient_checkpointing = "unsloth", # True or "unsloth" for very long context
    random_state = 3407,
    use_rslora = False,  # We support rank stabilized LoRA
    loftq_config = None, # And LoftQ
)

加载训练数据

使用 alpaca 格式模板，也可以到 huggingface 上下载数据集

alpaca_prompt = """Below is an instruction that describes a task, paired with an input that provides further context. Write a response that appropriately completes the request.

### Instruction:
{}

### Input:
{}

### Response:
{}"""

EOS_TOKEN = tokenizer.eos_token # Must add EOS_TOKEN
def formatting_prompts_func(examples):
    instructions = examples["instruction"]
    inputs       = examples["input"]
    outputs      = examples["output"]
    texts = []
    for instruction, input, output in zip(instructions, inputs, outputs):
        # Must add EOS_TOKEN, otherwise your generation will go on forever!
        text = alpaca_prompt.format(instruction, input, output) + EOS_TOKEN
        texts.append(text)
    return { "text" : texts, }
pass

from datasets import load_dataset
# 这里使用离线的方式下载好数据集
dataset = load_dataset("/workspace/sft/dataset/alpaca-cleaned", split = "train")
dataset = dataset.map(formatting_prompts_func, batched = True,)

模型训练

from trl import SFTTrainer
from transformers import TrainingArguments
from unsloth import is_bfloat16_supported

trainer = SFTTrainer(
    model = model,
    tokenizer = tokenizer,
    train_dataset = dataset,
    dataset_text_field = "text",
    max_seq_length = max_seq_length,
    dataset_num_proc = 2,
    packing = False, # Can make training 5x faster for short sequences.
    args = TrainingArguments(
        per_device_train_batch_size = 2,
        gradient_accumulation_steps = 4,
        warmup_steps = 5,
        # num_train_epochs = 1, # Set this for 1 full training run.
        max_steps = 60,
        learning_rate = 2e-4,
        fp16 = not is_bfloat16_supported(),
        bf16 = is_bfloat16_supported(),
        logging_steps = 1,
        optim = "adamw_8bit",
        weight_decay = 0.01,
        lr_scheduler_type = "linear",
        seed = 3407,
        output_dir = "outputs",
        report_to = "none", # Use this for WandB etc
    ),
)

trainer_stats = trainer.train()

推理

训练完成后，我们就可以针对模型进行推理测试

# alpaca_prompt = Copied from above
FastLanguageModel.for_inference(model) # Enable native 2x faster inference
inputs = tokenizer(
[
    alpaca_prompt.format(
        "Continue the fibonnaci sequence.", # instruction
        "1, 1, 2, 3, 5, 8", # input
        "", # output - leave this blank for generation!
    )
], return_tensors = "pt").to("cuda")

outputs = model.generate(**inputs, max_new_tokens = 64, use_cache = True)
tokenizer.batch_decode(outputs)

保存训练后的模型

将上面训练后的模型保存到本地为 lora_model 的目录里

model.save_pretrained("lora_model")  # Local saving
tokenizer.save_pretrained("lora_model")
# model.push_to_hub("your_name/lora_model", token = "...") # Online saving
# tokenizer.push_to_hub("your_name/lora_model", token = "...") # Online saving

导出至 Ollama

可以将经过微调的模型导出为 GGUF 格式，这样就可以在一些 UI 系统中使用如 open webui，或者导出至 Ollama 中直接使用

需要先安装 cmake 环境，可查看 Cloud Studio 软件环境安装 关于 cmake 安装的部分

使用 q4_k_m 格式

选择的量化方法是 q4_k_m 格式，即打开以下为 True 的方法

如果提示 Could NOT find CURL (missing: CURL_LIBRARY CURL_INCLUDE_DIR)，需要额外安装 apt install libcurl4-openssl-dev

# Save to 8bit Q8_0
if False: model.save_pretrained_gguf("model", tokenizer,)
# Remember to go to https://huggingface.co/settings/tokens for a token!
# And change hf to your username!
if False: model.push_to_hub_gguf("hf/model", tokenizer, token = "")

# Save to 16bit GGUF
if False: model.save_pretrained_gguf("model", tokenizer, quantization_method = "f16")
if False: model.push_to_hub_gguf("hf/model", tokenizer, quantization_method = "f16", token = "")

# Save to q4_k_m GGUF
if True: model.save_pretrained_gguf("model", tokenizer, quantization_method = "q4_k_m")
if False: model.push_to_hub_gguf("hf/model", tokenizer, quantization_method = "q4_k_m", token = "")

# Save to multiple GGUF options - much faster if you want multiple!
if False:
    model.push_to_hub_gguf(
        "hf/model", # Change hf to your username!
        tokenizer,
        quantization_method = ["q4_k_m", "q8_0", "q5_k_m",],
        token = "", # Get a token at https://huggingface.co/settings/tokens
    )

自动创建Modelfile

Unsloth 在转化模型为GGUF格式的时候，自动生成Ollama所需的Modelfile文件，其中包括模型的路径和我们用于微调过程的聊天模板。

# 打印Modelfile生成的模板
print(tokenizer._ollama_modelfile)

创建自定义模型

需要先本地安装 ollama 服务，参考 Cloud Studio 软件环境安装 的 Ollama 安装部分的内容

使用ollama create命令创建自定义模型

!ollama create unsloth_qwen2 -f /mnt/workspace/model/Modelfile

打开终端运行模型

!ollama run unsloth_qwen2