Add interactive demo link to model card
Browse files
README.md
CHANGED
|
@@ -1,8 +1,7 @@
|
|
| 1 |
---
|
| 2 |
-
title: HALT-CoT
|
| 3 |
-
sdk: gradio
|
| 4 |
-
app_file: app.py
|
| 5 |
license: mit
|
|
|
|
|
|
|
| 6 |
tags:
|
| 7 |
- chain-of-thought
|
| 8 |
- reasoning
|
|
@@ -14,6 +13,8 @@ tags:
|
|
| 14 |
|
| 15 |
# HALT-CoT
|
| 16 |
|
|
|
|
|
|
|
| 17 |
HALT-CoT is an inference-time early stopping method for chain-of-thought reasoning. After each generated reasoning step, it computes the Shannon entropy of the model's answer distribution over candidate answers. When entropy stays below a threshold, generation stops and the current most likely answer is returned.
|
| 18 |
|
| 19 |
This repository packages the method from:
|
|
|
|
| 1 |
---
|
|
|
|
|
|
|
|
|
|
| 2 |
license: mit
|
| 3 |
+
library_name: transformers
|
| 4 |
+
pipeline_tag: text-generation
|
| 5 |
tags:
|
| 6 |
- chain-of-thought
|
| 7 |
- reasoning
|
|
|
|
| 13 |
|
| 14 |
# HALT-CoT
|
| 15 |
|
| 16 |
+
> ▶️ **Try the interactive demo:** [**HALT-CoT Space**](https://huggingface.co/spaces/yass4/halt-cot) — run entropy-based early stopping live in your browser.
|
| 17 |
+
|
| 18 |
HALT-CoT is an inference-time early stopping method for chain-of-thought reasoning. After each generated reasoning step, it computes the Shannon entropy of the model's answer distribution over candidate answers. When entropy stays below a threshold, generation stops and the current most likely answer is returned.
|
| 19 |
|
| 20 |
This repository packages the method from:
|