Instructions to use Siyuc/INFUSER-Qwen3-4B-base with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Siyuc/INFUSER-Qwen3-4B-base with Transformers:

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

pipe = pipeline("text-generation", model="Siyuc/INFUSER-Qwen3-4B-base")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForMultimodalLM

tokenizer = AutoTokenizer.from_pretrained("Siyuc/INFUSER-Qwen3-4B-base")
model = AutoModelForMultimodalLM.from_pretrained("Siyuc/INFUSER-Qwen3-4B-base")
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 Siyuc/INFUSER-Qwen3-4B-base with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Siyuc/INFUSER-Qwen3-4B-base"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Siyuc/INFUSER-Qwen3-4B-base",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/Siyuc/INFUSER-Qwen3-4B-base

SGLang

How to use Siyuc/INFUSER-Qwen3-4B-base 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 "Siyuc/INFUSER-Qwen3-4B-base" \
    --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": "Siyuc/INFUSER-Qwen3-4B-base",
		"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 "Siyuc/INFUSER-Qwen3-4B-base" \
        --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": "Siyuc/INFUSER-Qwen3-4B-base",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use Siyuc/INFUSER-Qwen3-4B-base with Docker Model Runner:
```
docker model run hf.co/Siyuc/INFUSER-Qwen3-4B-base
```

INFUSER-Qwen3-4B-base

This repository contains the selected INFUSER solver checkpoint for Qwen3-4B Base, introduced in the paper INFUSER: Influence-Guided Self-Evolution Improves Reasoning.

The model is uploaded in a flat layout so it can be loaded directly with AutoModelForCausalLM.from_pretrained("Siyuc/INFUSER-Qwen3-4B-base") and AutoTokenizer.from_pretrained("Siyuc/INFUSER-Qwen3-4B-base").

Summary

Paper: INFUSER: Influence-Guided Self-Evolution Improves Reasoning
Blog: y-agent.github.io/infuser
Base model: Qwen/Qwen3-4B-Base
Method: INFUSER
Code repo: https://github.com/FFishy-git/INFUSER
Data repo: Siyuc/infuser-data

Evaluation

Released Checkpoint Scores

Category	Benchmark	Score
General	MMLU-Pro	60.68%
General	GPQA-Diamond	35.35%
General	SuperGPQA	33.90%
General	BBEH	12.57%
Math & physics	MATH500	77.90%
Math & physics	AIME2024	11.87%
Math & physics	AIME2025	11.56%
Math & physics	HMMT	3.19%
Math & physics	OlympiadBench (Math)	42.14%
Math & physics	OlympiadBench (Phys)	8.90%
Medical	MedQA	59.47%
Medical	MedXpertQA	13.80%
Coding	HumanEval+	75.23%
Coding	LiveCodeBench v1-5	23.01%

Comparison Summary

Category and overall means are computed over the same benchmark groups as the paper's main table. INFUSER avg is the average over three seeded INFUSER runs. R-Few (paper) and SPICE (paper) are self-reported values from their original papers, so missing categories are shown as -.

Category	This model	INFUSER avg	Base	R-Zero	AZR	R-Few (paper)	SPICE (paper)
General reasoning	35.62%	35.43%	29.46%	32.18%	32.93%	34.33%	35.00%
Math & physics reasoning	25.93%	25.73%	21.34%	25.12%	26.51%	-	-
Medical	36.63%	36.32%	34.24%	36.75%	36.14%	-	-
Coding	49.12%	48.63%	45.47%	47.65%	47.49%	-	-
Overall (14 benchmarks)	33.54%	33.28%	28.95%	32.01%	32.71%	-	-

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer

repo_id = "Siyuc/INFUSER-Qwen3-4B-base"
tokenizer = AutoTokenizer.from_pretrained(repo_id)
model = AutoModelForCausalLM.from_pretrained(repo_id)

Citation

@misc{chen2026infuser,
  title        = {INFUSER: Influence-Guided Self-Evolution Improves Reasoning},
  author       = {Siyu Chen and Miao Lu and Beining Wu and Heejune Sheen and Fengzhuo Zhang and Shuangning Li and Zhiyuan Li and Jose Blanchet and Tianhao Wang and Zhuoran Yang},
  year         = {2026},
  eprint       = {2606.09052},
  archivePrefix = {arXiv},
  primaryClass = {cs.LG},
  url          = {https://arxiv.org/abs/2606.09052}
}

Downloads last month: 35

Safetensors

Model size

4B params

Tensor type

BF16

Model tree for Siyuc/INFUSER-Qwen3-4B-base

Base model

Qwen/Qwen3-4B-Base

Finetuned

(308)

this model

Paper for Siyuc/INFUSER-Qwen3-4B-base

INFUSER: Influence-Guided Self-Evolution Improves Reasoning

Paper • 2606.09052 • Published 3 days ago