Add overview and update requirements.txt

Author: jusjinuk

README.md

@@ -5,10 +5,20 @@
 <h1 align="center">GuidedQuant</h1>
 </p>
 <p align="center"><b>Smarter LLM Post-Training Quantization using End Loss Guidance</b>, boosting the performance of <br> state-of-the-art <i>weight-only scalar</i>, <i>weight-only vector</i>, and <i>weight-and-activation</i> quantization methods.</p>
+<p align="center">
+<a href="https://arxiv.org/abs/2505.07004"><img src="https://img.shields.io/badge/arXiv-2505.07004-b31b1b.svg"></a>
+<a href="./LICENSE"><img src="https://img.shields.io/badge/License-MIT-yellow"></a>
+</p>
 
 # News
 - **May, 2025**: GuidedQuant is accepted to **ICML 2025**.
 
+# Overview
+![Light Mode](assets/objective-light.png#gh-light-mode-only)
+![Dark Mode](assets/objective-dark.png#gh-dark-mode-only)
+
+> *<b>GuidedQuant</b> enhances LLM quantization by integrating gradient information from the end loss into the quantization objective, boosting the performance of SOTA weight-only scalar, weight-only vector, and weight-and-activation quantization. Additionally, we introduce <b>LNQ</b>, a non-uniform scalar quantization algorithm which is guaranteed to monotonically decrease the quantization objective value.*
+
 # Installation & Usage
 
 To be released soon.

assets/objective-dark.png

@@ -1,16 +1,15 @@
-numpy~=1.26.4
-torch~=2.2.2
-transformers~=4.39.3
-tqdm~=4.66.2
-numba~=0.60.0
-datasets~=2.17.0
-accelerate~=0.29.2
-setuptools~=68.2.0
-pandas~=2.2.0
-safetensors~=0.4.2
-threadpoolctl~=3.2.0
-pyyaml~=6.0.1
-attributedict~=0.3.0
+numpy==1.26.4
+torch==2.5.1
+transformers==4.47.1
+tqdm==4.66.6
+numba==0.60.0
+datasets==3.2.0
+accelerate==0.29.3
+setuptools==68.2.2
+pandas==2.2.3
+safetensors==0.4.5
+threadpoolctl==3.2.0
+attributedict==0.3.0
 flash1dkmeans==0.1.4
 lm-eval==0.4.3
-peft==0.10.0
+peft==0.13.2