使用 Google Colab 免费算力微调 Llama-3 8b 大模型教程

使用 Google Colab 免费算力微调 Llama-3 8b 大模型

微调大语言模型可以让通用模型更擅长特定任务。本教程将演示如何在 Google Colab 环境下，利用 Unsloth 库对 Llama-3 8b 模型进行微调。

一、准备工作

环境要求：需要拥有 Google 账号并访问 Google Colab。
打开 Colab：登录谷歌 Colab 并创建新笔记本。
选择 GPU：在菜单栏点击 运行时 (Runtime) -> 更改运行时类型 (Change runtime type)，选择 GPU 作为硬件加速器。

二、安装依赖

运行以下代码安装必要的库。由于 Colab 默认 PyTorch 版本可能不兼容部分包，需单独安装。

%%capture
import torch
major_version, minor_version = torch.cuda.get_device_capability()
# Must install separately since Colab has torch 2.2.1, which breaks packages
!pip install "unsloth[colab-new] @ git+https://github.com/unslothai/unsloth.git"
if major_version >= 8:
    # Use this for new GPUs like Ampere, Hopper GPUs (RTX 30xx, RTX 40xx, A100, H100, L40)
    !pip install --no-deps packaging ninja einops flash-attn xformers trl peft accelerate bitsandbytes
else:
    # Use this for older GPUs (V100, Tesla T4, RTX 20xx)
    !pip install --no-deps xformers trl peft accelerate bitsandbytes

等待安装完成，确认输出无报错。

三、加载模型

使用 Unsloth 的 FastLanguageModel 加载量化后的 Llama-3 8b 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.

# 4bit pre quantized models we support for 4x faster downloading + no OOMs.
fourbit_models = [
    "unsloth/mistral-7b-bnb-4bit",
    "unsloth/mistral-7b-instruct-v0.2-bnb-4bit",
    "unsloth/llama-2-7b-bnb-4bit",
    "unsloth/gemma-7b-bnb-4bit",
    "unsloth/gemma-7b-it-bnb-4bit", # Instruct version of Gemma 7b
    "unsloth/gemma-2b-bnb-4bit",
    "unsloth/gemma-2b-it-bnb-4bit", # Instruct version of Gemma 2b
    "unsloth/llama-3-8b-bnb-4bit", # [NEW] 15 Trillion token Llama-3
]

model, tokenizer = FastLanguageModel.from_pretrained(
    model_name = "unsloth/llama-3-8b-bnb-4bit",
    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
)

此处需下载模型，约需要几分钟，耐心等待即可。

四、配置 PEFT (LoRA)

为模型添加 LoRA 适配器以进行高效微调。

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
)

五、准备数据集

使用 Hugging Face 上的公开数据集 yahma/alpaca-cleaned 进行训练。

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
dataset = load_dataset("yahma/alpaca-cleaned", split = "train")
dataset = dataset.map(formatting_prompts_func, batched = True,)

六、训练配置与执行

配置 SFTTrainer 并开始训练。

from trl import SFTTrainer
from transformers import TrainingArguments

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,
        max_steps = 60,
        learning_rate = 2e-4,
        fp16 = not torch.cuda.is_bf16_supported(),
        bf16 = torch.cuda.is_bf16_supported(),
        logging_steps = 1,
        optim = "adamw_8bit",
        weight_decay = 0.01,
        lr_scheduler_type = "linear",
        seed = 3407,
        output_dir = "outputs",
    ),
)

trainer_stats = trainer.train()

七、推理测试

训练完成后，可以使用模型进行生成测试。

# alpaca_prompt = Copied from above
FastLanguageModel.for_inference(model) # Enable native 2x faster inference
inputs = tokenizer(
[
    alpaca_prompt.format(
        "Continue the Fibonacci 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)

如需流式输出，可使用 TextStreamer：

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

from transformers import TextStreamer
text_streamer = TextStreamer(tokenizer)
_ = model.generate(**inputs, streamer = text_streamer, max_new_tokens = 128)

八、自定义问答体验

你可以修改指令来测试模型效果，例如让模型为小说《麦田里的守望者》写一篇短评。

if False:
    from unsloth import FastLanguageModel
    model, tokenizer = FastLanguageModel.from_pretrained(
        model_name = "lora_model", # YOUR MODEL YOU USED FOR TRAINING
        max_seq_length = max_seq_length,
        dtype = dtype,
        load_in_4bit = load_in_4bit,
    )
    FastLanguageModel.for_inference(model) # Enable native 2x faster inference

    # alpaca_prompt = You MUST copy from above!

    inputs = tokenizer(
[
    alpaca_prompt.format(
        "为小说《麦田里的守望者》写一篇短评", # instruction
        "", # input
        "", # output - leave this blank for generation!
    )
], return_tensors = "pt").to("cuda")

    outputs = model.generate(**inputs, max_new_tokens = 300, use_cache = True)
tokenizer.batch_decode(outputs)

可以看到执行步骤都非常的简单，当我们真实的需要去应用和训练时，简单的微调，也只需要了解和修改对应的参数和配置即可。Hugging Face 上有非常多的数据集，大家有兴趣可以自行尝试，使用别的数据集进行训练。

使用 Google Colab 免费算力微调 Llama-3 8b 大模型教程

使用 Google Colab 免费算力微调 Llama-3 8b 大模型

一、准备工作

二、安装依赖

三、加载模型

四、配置 PEFT (LoRA)

五、准备数据集

六、训练配置与执行

七、推理测试

八、自定义问答体验

更多推荐文章

相关免费在线工具

使用 Google Colab 免费算力微调 Llama-3 8b 大模型教程

使用 Google Colab 免费算力微调 Llama-3 8b 大模型

一、准备工作

二、安装依赖

三、加载模型

四、配置 PEFT (LoRA)

五、准备数据集

六、训练配置与执行

七、推理测试

八、自定义问答体验

微信扫一扫，关注极客日志

更多推荐文章

相关免费在线工具