Zum Inhalt springen

Angelika Romanou

@agromanou.bsky.social did:plc:b6wriy4nqf3aqpromaz4o4cn

PhD candidate at EPFL doing research in #NLProc 👩🏻‍💻 https://agromanou.github.io/

Bei Bluesky seit

Follower
532
Folgt
316
Posts
19
Tage in Ranglisten
0

Zusammenfassung

Angelika Romanou (@agromanou.bsky.social) ist seit 17 November 2024 auf Bluesky. Das Konto hat 532 Follower, folgt 316 Konten und hat 19 Posts veröffentlicht.

Ranglisten-Platzierungen

Die besten Tage, an denen Posts dieses Kontos in Debyas Tagesliste kamen.

Dieses Konto war noch in keiner Tagesrangliste.

Zuletzt gelikte Posts

Öffentliche Likes aus dem Bluesky-Netzwerk.

  • MELT Workshop@meltworkshop.bsky.social

    Keynote talk: Apertus: Democratizing Open and Compliant LLMs for Global Language Environments. Imanol Schlag introduces Apertus, a fully open suite of LLMs with a focus on compliance, transparency, and multilingual representation training across 1000+ languages. 🌍🤖

    Post anzeigen
  • Deniz Bayazit@bayazitdeniz.bsky.social

    5/ Looking closer, feature sharing has limits: in Hindi & Arabic, overlap stays low even at 341B tokens. This may be due to richer agreement systems (e.g., verbs agreeing w/ subjects & objects) forcing BLOOM to keep language-specific features—or simply data scarcity!

    Post anzeigen
  • Deniz Bayazit@bayazitdeniz.bsky.social

    4/ In #multilingual models, cross-language feature overlap starts low and rises with training. At 6B tokens in BLOOM, most detectors are language-specific or for punctuation; by 341B tokens shared crosslingual features emerge, capturing syntactic abstractions over token patterns.

    Post anzeigen
  • Deniz Bayazit@bayazitdeniz.bsky.social

    3/ Which features matter early but fade, and which gain importance later? In Pythia, token-level detectors drop out, while higher-level grammatical features—like plural-noun detectors and nouns formed from verbs (e.g., runner from run)—strengthen by 286B tokens.

    Post anzeigen
  • Deniz Bayazit@bayazitdeniz.bsky.social

    2/ We align critical checkpoints for a task with sparse crosscoders, measure each feature’s causal role, and introduce RelIE to compare their influence across checkpoints. This lets us trace how internal features shift—and when they matter—in models like Pythia, OLMo, and BLOOM.

    Post anzeigen

Melde dich an, um mehr zu sehen

Blockierte Konten

Blockierungen sind auf Bluesky öffentlich.

Dieses Konto blockiert niemanden.