HNHacker News
TopNewBestAskShowJobs

ConteMascetti71

165 karma · joined February 4, 2025

A seasoned tech geek with a passion that dates back to the Commodore 64 era. Born and raised in the beautiful region of Sicily, Vincenzo has transformed his early passion for computers into a career full of interesting tech experiences that has spanned over three decades.
submissionscomments

Inference-only expert boost saves 8.5% reasoning tokens in Qwen 35B MoE

zenodo.org·3 pts·ConteMascetti71·
0

What if LLMs escape through inferences itself? This is fiction. For now

agrillo.it·33 pts·ConteMascetti71·
83

Moe Estimator – Simulate decode speed with layer-major prefetch hiding

agrillo.it·2 pts·ConteMascetti71·
0

Vibe Coding at the End of 2023

antirez.com·2 pts·ConteMascetti71·
0

Gold Metal for "Future" Gpt5

github.com·3 pts·ConteMascetti71·
2

Latest VMware Critical vulnerabilities VMSA-2025-0013

support.broadcom.com·2 pts·ConteMascetti71·
1

Kimi K2 is a state-of-the-art mixture-of-experts (MoE) language model

github.com·352 pts·ConteMascetti71·
2

SOTA Model in 8B Size?

huggingface.co·2 pts·ConteMascetti71·
2

The technology behind ChatGPT4o's new image generation engine?

blog.paperspace.com·3 pts·ConteMascetti71·
0

DeepSeek Works on Function Calling

api-docs.deepseek.com·2 pts·ConteMascetti71·
1