Skip to content

GRPO, vLLM

Latest
Latest
Compare
Choose a tag to compare
@danielhanchen danielhanchen released this 06 Feb 14:47
· 2 commits to nightly since this release
2025-02
066ec25

GRPO is in Unsloth!

  • Experience the "aha moment" from DeepSeek R1's paper now with Unsloth!
  • LoRA (16bit) / QLoRA (4bit) actually work for GRPO now!
  • Unsloth can do GRPO for Phi-4 14B Llama-3.1 8B in a free 15GB Colab GPU!
  • Unsloth now has native fast inference (20x more throughput) via vLLM! Use it via model.fast_generate after setting FastLanguageModel.from_pretrained(..., fast_inference = True) and installing vLLM via pip install vllm
  • Llama 3.3 70B QLoRA GRPO should fit in 1x 48GB (best 1x 80GB)
  • Update unsloth via pip install --upgrade --no-cache-dir --force-reinstall unsloth_zoo unsloth vllm

image

GRPO Notebooks

Model Type Colab Link
Phi 4 (14B) GRPO Open in Colab
Llama 3.1 (8B) GRPO Open in Colab
Qwen 2.5 (3B) GRPO Open in Colab

Minimal GRPO example (courtesy of Will Brown]

!pip install unsloth vllm
!pip install git+https://github.com/huggingface/trl.git

from unsloth import FastLanguageModel, PatchFastRL
PatchFastRL("GRPO", FastLanguageModel)

from unsloth import is_bfloat16_supported
import torch
max_seq_length = 512
lora_rank = 32

model, tokenizer = FastLanguageModel.from_pretrained(
    model_name = "meta-llama/meta-Llama-3.1-8B-Instruct",
    max_seq_length = max_seq_length,
    load_in_4bit = True,
    fast_inference = True,
    max_lora_rank = lora_rank,
    gpu_memory_utilization = 0.6,
)
model = FastLanguageModel.get_peft_model(
    model,
    r = lora_rank,
    lora_alpha = lora_rank,
)

import re
from datasets import load_dataset, Dataset

# 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('openai/gsm8k', 'main')[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()

# 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]

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,
    num_generations = 6,
    max_prompt_length = 256,
    max_completion_length = 200,
    # num_train_epochs = 1,
    max_steps = 250,
    save_steps = 250,
    max_grad_norm = 0.1,
    report_to = "none",
    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()

Bug Fixes

  • Gemma 2 should be fixed now
  • Mistral base mapping should be fixed
  • Some syntax warning issue fixes
  • And many many more bug fixes!

What's Changed

New Contributors

Full Changelog: 2025-01...2025-02

Contributors

niryuu, danielhanchen, and 9 other contributors
Assets 2
Loading