Instructions to use wckwan/Search-R1-Qwen3-8B-Process-GRPO-PerTurnNorm with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use wckwan/Search-R1-Qwen3-8B-Process-GRPO-PerTurnNorm with Transformers:

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

pipe = pipeline("text-generation", model="wckwan/Search-R1-Qwen3-8B-Process-GRPO-PerTurnNorm")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("wckwan/Search-R1-Qwen3-8B-Process-GRPO-PerTurnNorm")
model = AutoModelForCausalLM.from_pretrained("wckwan/Search-R1-Qwen3-8B-Process-GRPO-PerTurnNorm", 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 wckwan/Search-R1-Qwen3-8B-Process-GRPO-PerTurnNorm with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "wckwan/Search-R1-Qwen3-8B-Process-GRPO-PerTurnNorm"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "wckwan/Search-R1-Qwen3-8B-Process-GRPO-PerTurnNorm",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/wckwan/Search-R1-Qwen3-8B-Process-GRPO-PerTurnNorm

SGLang

How to use wckwan/Search-R1-Qwen3-8B-Process-GRPO-PerTurnNorm 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 "wckwan/Search-R1-Qwen3-8B-Process-GRPO-PerTurnNorm" \
    --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": "wckwan/Search-R1-Qwen3-8B-Process-GRPO-PerTurnNorm",
		"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 "wckwan/Search-R1-Qwen3-8B-Process-GRPO-PerTurnNorm" \
        --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": "wckwan/Search-R1-Qwen3-8B-Process-GRPO-PerTurnNorm",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use wckwan/Search-R1-Qwen3-8B-Process-GRPO-PerTurnNorm with Docker Model Runner:
```
docker model run hf.co/wckwan/Search-R1-Qwen3-8B-Process-GRPO-PerTurnNorm
```

Search-R1-Qwen3-8B-Process-GRPO-PerTurnNorm

A Qwen/Qwen3-8B policy trained with GRPO on Search-R1 multi-turn retrieval QA with Process-GRPO: an Olmo-3-7B-Think verifier scores each turn, normalized per (group, turn-position).

The repository root holds the final policy from an earlier upload; every checkpoint added since lives in a step_<N>/ subfolder.

The root model and the step_{50..350} subfolders come from an earlier upload whose local copies have since been deleted; the remaining subfolders were added from local exports. Together they cover every 10 steps from 10 to 400.

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("wckwan/Search-R1-Qwen3-8B-Process-GRPO-PerTurnNorm")
tokenizer = AutoTokenizer.from_pretrained("wckwan/Search-R1-Qwen3-8B-Process-GRPO-PerTurnNorm")

# An intermediate checkpoint
model_step = AutoModelForCausalLM.from_pretrained("wckwan/Search-R1-Qwen3-8B-Process-GRPO-PerTurnNorm", subfolder="step_10")

Checkpoints

step_10/
step_20/
step_30/
step_40/
step_60/
step_70/
step_80/
step_90/
step_110/
step_120/
step_130/
step_140/
step_160/
step_170/
step_180/
step_190/
step_210/
step_220/
step_230/
step_240/
step_260/
step_270/
step_280/
step_290/
step_310/
step_320/
step_330/
step_340/
step_360/
step_370/
step_380/
step_390/
step_50/ (uploaded earlier; no local copy remains)
step_100/ (uploaded earlier; no local copy remains)
step_150/ (uploaded earlier; no local copy remains)
step_200/ (uploaded earlier; no local copy remains)
step_250/ (uploaded earlier; no local copy remains)
step_300/ (uploaded earlier; no local copy remains)
step_350/ (uploaded earlier; no local copy remains)

Raw FSDP checkpoints

fsdp/global_step_<N>/ holds the unmerged verl FSDP checkpoint (sharded fp32 model state, optimizer state and extra state) for step(s) 400. These are for resuming training, not for inference — use the merged exports above to load a policy.

Downloads last month: 430

Safetensors

Model size

8B params

Tensor type

BF16

Model tree for wckwan/Search-R1-Qwen3-8B-Process-GRPO-PerTurnNorm

Base model

Qwen/Qwen3-8B-Base

Finetuned

Qwen/Qwen3-8B

Finetuned

(1982)

this model