--

Author: asudeh

imgs/rsr.png

@@ -118,7 +118,7 @@
                         Born from high-end research, Needle is designed to be accessible to everyone while delivering top-notch performance. 
                         Whether you are a researcher, developer, or an enthusiast, Needle opens up innovative ways to explore your image datasets. ✨
                       <div class="project-links">
-                        <a href="https://www.cs.uic.edu/~indexlab/Needle/" class="btn project-btn" target="_blank">
+                        <a href="https://www.cs.uic.edu/~indexlab/Needle" class="btn project-btn" target="_blank">
                           🌐 Visit Project
                         </a>
                         <a href="https://github.com/UIC-InDeXLab/Needle" class="btn github-btn" target="_blank">
@@ -138,16 +138,17 @@
             </div>
             <hr/>
             <div class="row no-gutters">
-                  <!--<div class="col-md-3 text-center">
+                <div class="col-md-3 text-center">
                     <p></p>
-                    <img width="90%" src="imgs/needlelogo.png"/>
-                  </div>-->
-                    <div class="card-body">
+                    <img width="90%" src="imgs/rsr.png"/>
+                  </div>
+                  <div class="card-body col-md-9">
                         <h6><b>RSR 🧮</b>: Efficient Matrix Multiplication for Accelerating Inference in Binary and Ternary Neural Networks</h6>
-                        The codebase provides ready-to-use <b>NumPy</b>-based, <b>Torch</b>-based, and native <b>C++</b> implementations of our <i>RSR</i> and <i>RSR++</i> algorithms. 
-                        It contains codes for the inference on CPU and GPU and sample experiments on `1.58bit` models, including
-                        <a href="https://huggingface.co/HF1BitLLM/Llama3-8B-1.58-100B-tokens" target="_blank">[`Llama3-8B-1.58bit`]</a>, <a href="https://huggingface.co/tiiuae/Falcon3-10B-Instruct-1.58bit" target="_blank">[`Falcon3-10B-1.58bit`]</a>, and <a href="https://huggingface.co/tiiuae/Falcon3-3B-Instruct-1.58bit" target="_blank">[`Falcon3-3B-1.58bit`]</a>.
-                        ✨
+                        This project aims to provide a fast and efficient approach to low-bit matrix multiplication.
+                        The code repository implements Redundant Segment Reduction (RSR), a fast matrix multiplication algorithm designed for matrices in binary and ternary networks. 
+                        The RSR method optimizes computation efficiency by a log(n) factor, making it particularly useful for applications in low-bit deep learning and efficient inference.
+                        The codebase provides ready-to-use <b>C++</b> and <b>NumPy</b>-based implementations, as well as <b>PyTorch</b> implementations with both <b>CPU</b> and <b>GPU</b> support, enabling scalable and optimized matrix operations in deep learning environments. 
+                        It includes sample experiments on various `1.58bit` models and LLMs.✨
                         <div class="project-links">
                             <a href="https://github.com/UIC-InDeXLab/RSR/" class="btn github-btn" target="_blank">
                             🛠️ View on GitHub
@@ -156,8 +157,8 @@ <h6><b>RSR 🧮</b>: Efficient Matrix Multiplication for Accelerating Inference
                                 📄 Read Publication (Preprint)
                             </a>
                         </div>
-                    </div>
-            </div>
+                  </div>
+                </div>
             </div>
         </div>