llama.cpp
fix: speculative decoding broken on hybrid SSM/MoE (Qwen3.5 MoE)
#20075
Closed
Closed
fix: speculative decoding broken on hybrid SSM/MoE (Qwen3.5 MoE) #20075
eauchs
wants to merge
4 commits into
ggml-org:master
from
eauchs:feat/qwen-moe-speculative-decoding
fix: implement synchronous recurrent state checkpointing for hybrid m…
04e2fb15
eauchs
changed the title
eauchs
changed the title
eauchs
changed the title 
fix: replace soft rollback with proper failure in recurrent seq_rm
9a04ac4e
Revert "fix: replace soft rollback with proper failure in recurrent s…
8fe0e035
fix: implement recurrent state checkpointing for has_cell=true path
8a6b1c86
Login to write a write a comment.
Login via GitHub