Greedy decoding from large language models is commonly treated as deterministic. We show it is not precision-invariant: the same model, prompt, and decoding algorithm produce different outputs in BF16 versus FP16 on…
0 comments
No comments yet.
Related stories
- Hacker News · 5 points · 8 days ago
- Hacker News · 553 points · 10 days ago
- How AI is impacting who gets to cross bordersnewsroom.taylorandfrancisgroup.comHacker News · 1 points · 9 days ago
- Hacker News · 2 points · 9 days ago
- Hacker News · 2 points · 10 days ago
- DEV Community · 5 points · 8 days ago