Unloth

Unsloth 是一个加速 LLM 微调的库，它使得训练模型更快，并减少计算资源的需求。Unsloth 与 TRL 集成。

可在 Google Colab T4 GPU 上免费运行。

pip install unsloth vllm
pip install --upgrade pillow

Setup unsloth 加载模型

from unsloth import FastLanguageModel 这个类将 transformers 与 Unsloth 优化集成。

加载 Google 的 Gemma 3 1B Instruct 模型并配置它进行微调。

from unsloth import FastLanguageModel
import torch

max_seq_length = 1024  # Can increase for longer reasoning traces
lora_rank = 32  # Larger rank = smarter, but slower

model, tokenizer = FastLanguageModel.from_pretrained(
    model_name="google/gemma-3-1b-it",
    max_seq_length=max_seq_length,
    load_in_4bit=True,  # False for LoRA 16bit
    fast_inference=True,  # Enable vLLM fast inference
    max_lora_rank=lora_rank,
    gpu_memory_utilization=0.6,  # Reduce if out of memory
)

model = FastLanguageModel.get_peft_model(
    model,
    r=lora_rank,  # 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",
    ],  # Remove QKVO if out of memory
    lora_alpha=lora_rank,
    use_gradient_checkpointing="unsloth",  # Enable long context finetuning
    random_state=3407,
)

用 4 位量化方式加载模型以节省内存，并应用 LoRA 低秩适配进行高效微调。 target_modules 参数指定要微调模型的哪些层， use_gradient_checkpointing 参数启用使用更长的上下文进行训练。

准备数据

使用 GSM8K 数据集，其包含小学数学问题。我们将格式化数据，以鼓励模型在给出答案之前展示其推理过程。首先，定义提示和答案的格式：

# Define the system prompt that instructs the model to use a specific format
SYSTEM_PROMPT = """
Respond in the following format:
<reasoning>
...
</reasoning>
<answer>
...
</answer>
"""

XML_COT_FORMAT = """\
<reasoning>
{reasoning}
</reasoning>
<answer>
{answer}
</answer>
"""

然后准备数据集，从数据集中提取答案并将其格式化为字符串：

import re
from datasets import load_dataset, Dataset

# Helper functions to extract answers from different formats
def extract_xml_answer(text: str) -> str:
    answer = text.split("<answer>")[-1]
    answer = answer.split("</answer>")[0]
    return answer.strip()

def extract_hash_answer(text: str) -> str | None:
    if "####" not in text:
        return None
    return text.split("####")[1].strip()

# Function to prepare the GSM8K dataset
def get_gsm8k_questions(split="train") -> Dataset:
    data = load_dataset("openai/gsm8k", "main")[split]
    data = data.map(
        lambda x: {
            "prompt": [
                {"role": "system", "content": SYSTEM_PROMPT},
                {"role": "user", "content": x["question"]},
            ],
            "answer": extract_hash_answer(x["answer"]),
        }
    )
    return data

dataset = get_gsm8k_questions()

定义 Reward function

回顾下， GRPO 可以使用奖励函数根据可验证的标准（如长度和格式）来指导模型的学习。

# 奖励模型当其答案与正确答案匹配时
def correctness_reward_func(prompts, completions, answer, **kwargs) -> list[float]:
    responses = [completion[0]["content"] for completion in completions]
    q = prompts[0][-1]["content"]
    extracted_responses = [extract_xml_answer(r) for r in responses]
    print(
        "-" * 20,
        f"Question:\n{q}",
        f"\nAnswer:\n{answer[0]}",
        f"\nResponse:\n{responses[0]}",
        f"\nExtracted:\n{extracted_responses[0]}",
    )
    return [2.0 if r == a else 0.0 for r, a in zip(extracted_responses, answer)]


# 奖励模型提供 数值答案
def int_reward_func(completions, **kwargs) -> list[float]:
    responses = [completion[0]["content"] for completion in completions]
    extracted_responses = [extract_xml_answer(r) for r in responses]
    return [0.5 if r.isdigit() else 0.0 for r in extracted_responses]


# 奖励模型遵循指定格式
def strict_format_reward_func(completions, **kwargs) -> list[float]:
    pattern = r"^<reasoning>\n.*?\n</reasoning>\n<answer>\n.*?\n</answer>\n$"
    responses = [completion[0]["content"] for completion in completions]
    matches = [re.match(pattern, r) for r in responses]
    return [0.5 if match else 0.0 for match in matches]


# 奖励模型遵循指定格式
def soft_format_reward_func(completions, **kwargs) -> list[float]:
    pattern = r"<reasoning>.*?</reasoning>\s*<answer>.*?</answer>"
    responses = [completion[0]["content"] for completion in completions]
    matches = [re.match(pattern, r) for r in responses]
    return [0.5 if match else 0.0 for match in matches]


# 奖励正确的 XML 标签使用，并惩罚闭合标签后的额外内容
def count_xml(text) -> float:
    count = 0.0
    if text.count("<reasoning>\n") == 1:
        count += 0.125
    if text.count("\n</reasoning>\n") == 1:
        count += 0.125
    if text.count("\n<answer>\n") == 1:
        count += 0.125
        count -= len(text.split("\n</answer>\n")[-1]) * 0.001
    if text.count("\n</answer>") == 1:
        count += 0.125
        count -= (len(text.split("\n</answer>")[-1]) - 1) * 0.001
    return count


def xmlcount_reward_func(completions, **kwargs) -> list[float]:
    contents = [completion[0]["content"] for completion in completions]
    return [count_xml(c) for c in contents]

使用 GRPO 进行训练

现在使用得到的模型、分词器和奖励函数来设置 GRPO 训练器：

from trl import GRPOConfig, GRPOTrainer

max_prompt_length = 256

training_args = GRPOConfig(
    learning_rate=5e-6,
    adam_beta1=0.9,
    adam_beta2=0.99,
    weight_decay=0.1,
    warmup_ratio=0.1,
    lr_scheduler_type="cosine",
    optim="paged_adamw_8bit",
    logging_steps=1,
    per_device_train_batch_size=1,
    gradient_accumulation_steps=1,  # Increase to 4 for smoother training
    num_generations=6,  # Decrease if out of memory
    max_prompt_length=max_prompt_length,
    max_completion_length=max_seq_length - max_prompt_length,
    # num_train_epochs = 1, # Set to 1 for a full training run
    max_steps=250,
    save_steps=250,
    max_grad_norm=0.1,
    report_to="none",  # Can use Weights & Biases
    output_dir="outputs",
)

trainer = GRPOTrainer(
    model=model,
    processing_class=tokenizer,
    reward_funcs=[
        xmlcount_reward_func,
        soft_format_reward_func,
        strict_format_reward_func,
        int_reward_func,
        correctness_reward_func,
    ],
    args=training_args,
    train_dataset=dataset,
)

开始训练：

trainer.train()

测试模型

model.save_lora("grpo_saved_lora")

from vllm import SamplingParams

# 计算pi
text = tokenizer.apply_chat_template(
    [
        {"role": "system", "content": SYSTEM_PROMPT},
        {"role": "user", "content": "Calculate pi."},
    ],
    tokenize=False,
    add_generation_prompt=True,
)

sampling_params = SamplingParams(
    temperature=0.8,
    top_p=0.95,
    max_tokens=1024,
)
output = (
    model.fast_generate(
        text,
        sampling_params=sampling_params,
        lora_request=model.load_lora("grpo_saved_lora"),
    )[0]
    .outputs[0]
    .text
)

print(output)

保存模型

Unsloth 提供了多种保存您微调模型的选项：

保存为 16-bit precision

model.save_pretrained_merged("model", tokenizer, save_method="merged_16bit")

push 到 HF

model.push_to_hub_merged(
    "your-username/model-name", tokenizer, save_method="merged_16bit", token="your-token"
)

保存为 GGUF 格式，用于与 llama.cpp 一起使用

model.push_to_hub_gguf(
    "your-username/model-name",
    tokenizer,
    quantization_method=["q4_k_m", "q8_0", "q5_k_m"],
    token="your-token",
)

然后可以使用 llama.cpp 运行模型： llama-cli -m my_model.gguf。

Unloth#

Setup unsloth 加载模型#

准备数据#

定义 Reward function#

使用 GRPO 进行训练#

测试模型#

保存模型#