Spaces:
Running
on
Zero
Running
on
Zero
da03
commited on
Commit
·
0ec8782
1
Parent(s):
ee2ee31
app.py
CHANGED
@@ -141,8 +141,8 @@ demo = gr.Interface(
|
|
141 |
gr.HighlightedText(label='No CoT Predicted Product', combine_adjacent=False, show_legend=False, color_map=color_map, show_inline_category=False),
|
142 |
gr.HighlightedText(label='Explicit CoT Predicted Product', combine_adjacent=False, show_legend=False, color_map=color_map, show_inline_category=False),
|
143 |
],
|
144 |
-
title='
|
145 |
-
description='This demo
|
146 |
article="""
|
147 |
- [Paper: From Explicit CoT to Implicit CoT: Learning to Internalize CoT Step by Step](https://arxiv.org/pdf/2405.14838)
|
148 |
- [Code Repository](https://github.com/da03/Internalize_CoT_Step_by_Step)
|
|
|
141 |
gr.HighlightedText(label='No CoT Predicted Product', combine_adjacent=False, show_legend=False, color_map=color_map, show_inline_category=False),
|
142 |
gr.HighlightedText(label='Explicit CoT Predicted Product', combine_adjacent=False, show_legend=False, color_map=color_map, show_inline_category=False),
|
143 |
],
|
144 |
+
title='Can GPT2 Predict Multiplication of Two Numbers Without Intermediate Steps?',
|
145 |
+
description='This demo demonstrates GPT2\'s ability to directly predict the product of two large numbers without intermediate reasoning steps. The GPT2 has been finetuned to internalize chain-of-thought (CoT) reasoning within its hidden states through our stepwise internalization approach. The results demonstrate the effectiveness of implicit CoT (our approach, accurate and fast), compared to no CoT (fast but inaccurate) and explicit CoT (accurate but slow).',
|
146 |
article="""
|
147 |
- [Paper: From Explicit CoT to Implicit CoT: Learning to Internalize CoT Step by Step](https://arxiv.org/pdf/2405.14838)
|
148 |
- [Code Repository](https://github.com/da03/Internalize_CoT_Step_by_Step)
|