zagreus-0.4B post-trained to 0.4787 on ITALIC (base 0.28, prior leader 0.372) via committee soft answer-KD, under 10 H100h.
Ruggero Marino Lazzaroni PRO
ruggsea
AI & ML interests
NLP in any form
Recent Activity
authored a paper about 5 hours ago
Infini-News: Efficiently Queryable Access to 1.3 Billion Processed Common Crawl News Articles authored a paper about 5 hours ago
Shaping Explanations: Semantic Reward Modeling with Encoder-Only Transformers for GRPO updated a dataset about 21 hours ago
ruggsea/infini-news-indexOrganizations
