Commit History

Update: --num-draft-tokens 1 is a net speedup (1.06x), not the net slowdown found at the default N=2
4bcdb44
verified

filipemiguelmartins commited on

Distilled EuroLLM-1.7B draft model for AMALIA-9B speculative decoding, with honest mixed results
8c98e0b
verified

filipemiguelmartins commited on