Instructions to use Yulin-Li/ReBalance with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use Yulin-Li/ReBalance with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("text-generation", model="Yulin-Li/ReBalance")# Load model directly from transformers import AutoModel model = AutoModel.from_pretrained("Yulin-Li/ReBalance", dtype="auto") - Notebooks
- Google Colab
- Kaggle
- Local Apps
- vLLM
How to use Yulin-Li/ReBalance with vLLM:
Install from pip and serve model
# Install vLLM from pip: pip install vllm # Start the vLLM server: vllm serve "Yulin-Li/ReBalance" # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:8000/v1/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "Yulin-Li/ReBalance", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }'Use Docker
docker model run hf.co/Yulin-Li/ReBalance
- SGLang
How to use Yulin-Li/ReBalance 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 "Yulin-Li/ReBalance" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "Yulin-Li/ReBalance", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }'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 "Yulin-Li/ReBalance" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "Yulin-Li/ReBalance", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }' - Docker Model Runner
How to use Yulin-Li/ReBalance with Docker Model Runner:
docker model run hf.co/Yulin-Li/ReBalance
Add pipeline_tag and library_name metadata
#1
by nielsr HF Staff - opened
README.md
CHANGED
|
@@ -1,17 +1,17 @@
|
|
| 1 |
---
|
|
|
|
|
|
|
| 2 |
license: mit
|
|
|
|
|
|
|
| 3 |
tags:
|
| 4 |
- rebalance
|
| 5 |
- steering-vector
|
| 6 |
- reasoning
|
| 7 |
- llm
|
| 8 |
- iclr-2026
|
| 9 |
-
language:
|
| 10 |
-
- en
|
| 11 |
---
|
| 12 |
|
| 13 |
-
|
| 14 |
-
|
| 15 |
<h1 align="center">ReBalance Steering Vectors</h1>
|
| 16 |
|
| 17 |
<p align="center">
|
|
@@ -26,6 +26,13 @@ language:
|
|
| 26 |
<a href="https://github.com/yu-lin-li/ReBalance/blob/main/LICENSE"><img src="https://img.shields.io/badge/License-MIT-yellow.svg" alt="License: MIT"></a>
|
| 27 |
</p>
|
| 28 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 29 |
## Overview
|
| 30 |
This repository provides **steering vectors only** for ReBalance.
|
| 31 |
|
|
|
|
| 1 |
---
|
| 2 |
+
language:
|
| 3 |
+
- en
|
| 4 |
license: mit
|
| 5 |
+
pipeline_tag: text-generation
|
| 6 |
+
library_name: transformers
|
| 7 |
tags:
|
| 8 |
- rebalance
|
| 9 |
- steering-vector
|
| 10 |
- reasoning
|
| 11 |
- llm
|
| 12 |
- iclr-2026
|
|
|
|
|
|
|
| 13 |
---
|
| 14 |
|
|
|
|
|
|
|
| 15 |
<h1 align="center">ReBalance Steering Vectors</h1>
|
| 16 |
|
| 17 |
<p align="center">
|
|
|
|
| 26 |
<a href="https://github.com/yu-lin-li/ReBalance/blob/main/LICENSE"><img src="https://img.shields.io/badge/License-MIT-yellow.svg" alt="License: MIT"></a>
|
| 27 |
</p>
|
| 28 |
|
| 29 |
+
## Description
|
| 30 |
+
This repository contains the steering vectors presented in the paper **"Efficient Reasoning with Balanced Thinking"** (ICLR 2026).
|
| 31 |
+
|
| 32 |
+
**Authors**: Yulin Li, Tengyao Tu, Li Ding, Junjie Wang, Huiling Zhen, Yixin Chen, Yong Li, Zhuotao Tian.
|
| 33 |
+
|
| 34 |
+
ReBalance is a training-free framework that achieves efficient reasoning by identifying and mitigating "overthinking" (redundant computational steps) and "underthinking" (insufficient exploration) in Large Reasoning Models (LRMs). It leverages confidence as a dynamic indicator to modulate reasoning trajectories using the steering vectors provided here.
|
| 35 |
+
|
| 36 |
## Overview
|
| 37 |
This repository provides **steering vectors only** for ReBalance.
|
| 38 |
|