SVD-LLM
SVD-LLM copied to clipboard
AIoT-MLSys-Lab
Official Code for "SVD-LLM: Truncation-aware Singular Value Decomposition for Large Language Model Compression"
## Issue: Converting Compressed LLaMA2 Model to Hugging Face-Compatible Format ### Description We have successfully compressed a LLaMA2 model with 4.4 billion parameters. However, I am encountering issues when trying...
Hi, could you please release the codes of svdllm plus LoRA? Thanks, Xin
Hi, thank you for the great work. I followed the homogeneous compression ratio approach in SVD-LLM and applied a 20% compression to each layer. However, I obtained a perplexity of...
Dear authors, We are wondering when the newest version of the code can be released. Thank you so much for your great paper and repo!
if loss < 1.0, the result will become a negative number, then the compress ratio can not be calculated correctly.
Hello! Thanks for your work, I wonder if you could share the exact compression configuration for ASVD Llama 3 8B model as you report in Table 2 of [your preprint](https://arxiv.org/abs/2503.12340)....
Hi, thanks for your great work! But is there any update on the code of the efficiency evaluation part? Cause I saw it was uploaded 11 months ago, and I...
Thank you for your outstanding work. However, we are unable to reproduce the results of **PARAMETER UPDATE WITH SEQUENTIAL LOW-RANK APPROXIMATION** based on the code you published. Our steps are...
Hi @kar-m 🤗 I'm Niels and work as part of the open-source team at Hugging Face. I discovered your work through Hugging Face's daily papers as yours got featured: https://huggingface.co/papers/2403.07378....
Our team highly recognizes your work. You are currently the best at matrix decomposition. How can I reproduce the generation task in Table 10? I have read your code in...
Metadata
Owner
Metadata
Official Code for "SVD-LLM: Truncation-aware Singular Value Decomposition for Large Language Model Compression"