How to use from
vLLM
Install from pip and serve model
# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "TreezzZ/r2s-webshop-qwen3-4b-student"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "TreezzZ/r2s-webshop-qwen3-4b-student",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'
Use Docker
docker model run hf.co/TreezzZ/r2s-webshop-qwen3-4b-student
Quick Links

R2S student: WebShop, Qwen3-4B (teacher Qwen3-30B-A3B-Instruct-2507)

Student checkpoint (final update, step 149) from the main results table of the R2S (Rollout-to-Scorer Distillation) paper, Capable Teacher, Myopic Scorer: Reward-Free Teacher Repair for Agentic On-Policy Distillation. It was distilled on-policy from the repaired teacher TreezzZ/r2s-webshop-qwen3-30b-a3b-teacher-for-qwen3-4b.

Environment WebShop (text agent, tool calls)
Student base Qwen/Qwen3-4B
Teacher Qwen3-30B-A3B-Instruct-2507, fine-tuned by R2S (teacher alone, mixing weight 1.0)
Objective on-policy distillation, sampled reverse KL, 150 updates, rollout temperature 0.4, seed 42
Checkpoint final update (149), bf16 safetensors

Results (main table)

WebShop success (%) on the 500 canonical goals. Mean ยฑ std over three decodes (seeds 42, 1001, 1002) of the final checkpoint.

Student alone Vanilla OPD R2S (this model)
15.5 ยฑ 1.2 24.7 ยฑ 1.1 28.5 ยฑ 0.6

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer
repo = "TreezzZ/r2s-webshop-qwen3-4b-student"
tok = AutoTokenizer.from_pretrained(repo)
model = AutoModelForCausalLM.from_pretrained(repo, dtype="bfloat16")

The checkpoint was trained with the environment prompts, tool schemas and rollout settings of the paper's code release; the chat template is the base model's.

Provenance

Trained with the paper's code release, Tree-Shu-Zhao/miles-agent at tag r2s (its README reproduces the main table; docs/r2s.md holds the details). The weights were exported from the training checkpoint and verified against the base model's tensor names and shapes.

Citation

@article{zhao2026capable,
  title={Capable Teacher, Myopic Scorer: Reward-Free Teacher Repair for Agentic On-Policy Distillation},
  author={Zhao, Shu and Yu, Tan},
  journal={arXiv preprint arXiv:2610.28586},
  year={2026}
}
Downloads last month
305
Safetensors
Model size
4B params
Tensor type
BF16
ยท
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Model tree for TreezzZ/r2s-webshop-qwen3-4b-student

Finetuned
Qwen/Qwen3-4B
Finetuned
(1,286)
this model

Collection including TreezzZ/r2s-webshop-qwen3-4b-student

Paper for TreezzZ/r2s-webshop-qwen3-4b-student