Instructions to use Jinhe/ReflectRL-Qwen2.5-3B-Instruct-GRPO with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Jinhe/ReflectRL-Qwen2.5-3B-Instruct-GRPO with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="Jinhe/ReflectRL-Qwen2.5-3B-Instruct-GRPO")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("Jinhe/ReflectRL-Qwen2.5-3B-Instruct-GRPO")
model = AutoModelForCausalLM.from_pretrained("Jinhe/ReflectRL-Qwen2.5-3B-Instruct-GRPO", device_map="auto")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use Jinhe/ReflectRL-Qwen2.5-3B-Instruct-GRPO with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Jinhe/ReflectRL-Qwen2.5-3B-Instruct-GRPO"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Jinhe/ReflectRL-Qwen2.5-3B-Instruct-GRPO",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/Jinhe/ReflectRL-Qwen2.5-3B-Instruct-GRPO

SGLang

How to use Jinhe/ReflectRL-Qwen2.5-3B-Instruct-GRPO with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "Jinhe/ReflectRL-Qwen2.5-3B-Instruct-GRPO" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Jinhe/ReflectRL-Qwen2.5-3B-Instruct-GRPO",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "Jinhe/ReflectRL-Qwen2.5-3B-Instruct-GRPO" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Jinhe/ReflectRL-Qwen2.5-3B-Instruct-GRPO",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use Jinhe/ReflectRL-Qwen2.5-3B-Instruct-GRPO with Docker Model Runner:
```
docker model run hf.co/Jinhe/ReflectRL-Qwen2.5-3B-Instruct-GRPO
```

ReflectRL-Qwen2.5-3B-Instruct-GRPO-ReflectRL

This repository contains the checkpoint for ReflectRL-Qwen2.5-3B-Instruct-GRPO-ReflectRL, presented in the paper ReflectRL: Learning from Golden Negative Trajectories via Reflective-to-Direct Reasoning.

GitHub Repository: ReflectRL
Base Model: Qwen/Qwen2.5-3B-Instruct

Overview

ReflectRL is a lightweight framework for learning from Golden Negative Trajectories (GNTs) during on-policy post-training. Instead of imitating failed expert trajectories directly, ReflectRL uses them as reflective context during training and gradually transitions the policy back to direct reasoning for inference.

Motivated by the Reflection Advantage, ReflectRL allocates part of each rollout group to a reflective interface during training and gradually decays that allocation to zero, allowing the final model to be used directly with standard prompts.

Citation

@article{bi2026reflectrl,
  title={ReflectRL: Learning from Golden Negative Trajectories via Reflective-to-Direct Reasoning},
  author={Bi, Jinhe and Zhou, Chennan and Jin, Zengjie and Aniri and Lu, Shuo and Huang, Wenke and Cao, Hu and Xiao, Xun and Zhu, Zhihong and Tresp, Volker and Shen, Fei and Ma, Yunpu and Chua, Tat-Seng},
  journal={arXiv preprint arXiv:2608.03972},
  year={2026}
}