Fix fp8 kv replicate#349

Closed

quic-morteza wants to merge 3 commits intoquic:mainfrom

quic-morteza:fix_fp8_kv_replicate

Commits on Mar 25, 2025

the fix replicates biases too if they exist (e.g. Qwen)
quic-morteza
committed

Commits on Apr 7, 2025

Merge branch 'quic:main' into fix_fp8_kv_replicate
quic-morteza
authored

Commits on Apr 8, 2025

fix is needed/tested for replicate_kv_heads.py to work with FP8-Dynamic models
quic-morteza
committed