view article Article Tensor Parallelism (TP) in Transformers: 5 Minutes to Understand qgallouedec • Dec 4, 2025 • 74
DSSD Collection Trained early exit head to be used with Dynamic Self-Speculative Decoding • 2 items • Updated Jan 8
view post Post 883 New research model out ! I uploaded a new Branchy model based on Phi-2 for faster inference using Early Exit. Check it out : https://huggingface.co/valcore/Branchy-Phi-2. I also uploaded a Hugging Face Space to try it out : https://huggingface.co/spaces/valcore/Branchy-phi-2, unfortunately inference is very slow on free tier. Let me know what you are thinking about it ! Reply
EERO: Early Exit with Reject Option for Efficient Classification with limited budget Paper • 2402.03779 • Published Feb 6, 2024