problem history
[Bug]: Qwen3.5 (hybrid GDN) models cannot run dflash speculative decoding with pipeline parallelism (missing supports_aux_hidden_states_over_pp opt-in)
Revision reasons are not recorded in this slice.