一、定义

1.使用unsloth 进行强化学习训练grpo
2.训练后加载推理
3. 训练后模型+lora 合并

二、实现

1.使用unsloth 进行强化学习训练grpo

#coding="utf8"
from unsloth import FastLanguageModel, PatchFastRL
PatchFastRL("GRPO", FastLanguageModel)
import re
from datasets import load_dataset, Dataset


#加载模型
from unsloth import is_bfloat16_supported
import torch
max_seq_length = 512 # Can increase for longer reasoning traces
lora_rank = 32 # Larger rank = smarter, but slower

model, tokenizer = FastLanguageModel.from_pretrained(
    model_name = "/home/grpo_test/Qwen2.5-0.5B-Instruct",
    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,
)
print(model)

#加载数据集
#######################################################################################
# Load and prep dataset
SYSTEM_PROMPT = """
Respond in the following format:
<reasoning>
...
</reasoning>
<answer>
...
</answer>
"""

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

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()


# uncomment middle messages for 1-shot prompting
def get_gsm8k_questions(split = "train") -> Dataset:
    data = load_dataset('/home/grpo_test/openaigsm8k', 'main').select(range(1000))[split] # type: ignore
    data = data.map(lambda x: { # type: ignore
        'prompt': [
            {'role': 'system', 'content': SYSTEM_PROMPT},
            {'role': 'user', 'content': x['question']}
        ],
        'answer': extract_hash_answer(x['answer'])
    }) # type: ignore
    return data # type: ignore

dataset = get_gsm8k_questions()


#定义奖励规则
# Reward functions
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)]

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]

def strict_format_reward_func(completions, **kwargs) -> list[float]:
    """Reward function that checks if the completion has a specific format."""
    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]

def soft_format_reward_func(completions, **kwargs) -> list[float]:
    """Reward function that checks if the completion has a specific format."""
    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]

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]



if __name__ == "__main__":
    from transformers import set_seed
    set_seed(3407)

#训练模型
from trl import GRPOConfig, GRPOTrainer
training_args = GRPOConfig(
    use_vllm = True, # use vLLM for fast inference!
    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,
    bf16 = is_bfloat16_supported(),
    fp16 = not is_bfloat16_supported(),
    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 = 256,
    max_completion_length = 200,
    # 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,
)
trainer.train()

model.save_pretrained("lora_model") # Local saving
tokenizer.save_pretrained("lora_model")

2.训练后加载推理

1. 加载方式一： 以unsloth 原生接口加载，并进行推理， 该接口可以加载lora、qlora、非训练模型路径

from unsloth import FastLanguageModel
from unsloth import is_bfloat16_supported
import torch
max_seq_length = 512 # Can increase for longer reasoning traces
lora_rank = 32 # Larger rank = smarter, but slower

model, tokenizer = FastLanguageModel.from_pretrained(
    model_name = "outputs/checkpoint-250",
    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
)

FastLanguageModel.for_inference(model)

print(model)

SYSTEM_PROMPT = """
Respond in the following format:
<reasoning>
...
</reasoning>
<answer>
...
</answer>
"""

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

inputs = tokenizer(
[
   text,
], return_tensors = "pt").to("cuda")

from transformers import TextStreamer
text_streamer = TextStreamer(tokenizer)
outputs = model.generate(**inputs, max_new_tokens = 500)
print(outputs)
outputs = outputs[:, inputs['input_ids'].shape[1]:]
print(tokenizer.decode(outputs[0]))

2. 以transformer 原生方式加载推理

from peft import AutoPeftModelForCausalLM
from transformers import AutoTokenizer
model = AutoPeftModelForCausalLM.from_pretrained(
    "outputs/checkpoint-250", # YOUR MODEL YOU USED FOR TRAINING
    load_in_4bit = True,
)
tokenizer = AutoTokenizer.from_pretrained("outputs/checkpoint-250")

SYSTEM_PROMPT = """
Respond in the following format:
<reasoning>
...
</reasoning>
<answer>
...
</answer>
"""

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

inputs = tokenizer(
[
   text,
], return_tensors = "pt").to("cuda")

from transformers import TextStreamer
text_streamer = TextStreamer(tokenizer)
outputs = model.generate(**inputs, max_new_tokens = 128)
print(outputs)
outputs = outputs[:, inputs['input_ids'].shape[1]:]
print(tokenizer.decode(outputs[0]))

3.vllm + base Model+ lora

import torch
max_seq_length = 512 # Can increase for longer reasoning traces
lora_rank = 32 # Larger rank = smarter, but slower

from unsloth import FastLanguageModel
from unsloth import is_bfloat16_supported
import torch
max_seq_length = 512 # Can increase for longer reasoning traces
lora_rank = 32 # Larger rank = smarter, but slower

model, tokenizer = FastLanguageModel.from_pretrained(
    model_name = "outputs/checkpoint-250",
    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,
)


from vllm import LLM, SamplingParams
from transformers import AutoTokenizer

SYSTEM_PROMPT = """
Respond in the following format:
<reasoning>
...
</reasoning>
<answer>
...
</answer>
"""

text = tokenizer.apply_chat_template([
    {"role" : "system", "content" : SYSTEM_PROMPT},
    {"role" : "user", "content" : "How many r's are in strawberry?"},
], tokenize = False, add_generation_prompt = True)

from vllm import SamplingParams
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"),    # Load LoRA weights, 动态加载
)[0].outputs[0].text

print(output)

3. 训练后模型+lora 合并 与加载预测

from peft import AutoPeftModelForCausalLM
from transformers import AutoTokenizer
model = AutoPeftModelForCausalLM.from_pretrained(
    "outputs/checkpoint-250", # YOUR MODEL YOU USED FOR TRAINING
    load_in_4bit = True,
)
tokenizer = AutoTokenizer.from_pretrained("outputs/checkpoint-250")

model = model.merge_and_unload()
model.save_pretrained("merged-model")

tokenizer.save_pretrained("merged-model")