diff --git a/python/sglang/srt/server_args.py b/python/sglang/srt/server_args.py index 4eac6e339..1d2444762 100644 --- a/python/sglang/srt/server_args.py +++ b/python/sglang/srt/server_args.py @@ -5250,6 +5250,19 @@ class ServerArgs: assert view.mamba_track_interval % view.page_size == 0 assert self.mamba_cache_chunk_size is not None + if ( + view.chunked_prefill_size is not None + and 0 < view.chunked_prefill_size < self.mamba_cache_chunk_size + ): + logger.warning( + "Mamba radix extra-buffer is enabled with chunked_prefill_size=%s " + "smaller than mamba_cache_chunk_size=%s. This can make " + "mamba_track_mask false for unfinished chunked-prefill handoff " + "and skip Mamba state checkpoints.", + view.chunked_prefill_size, + self.mamba_cache_chunk_size, + ) + def _handle_mamba_radix_cache(self, model_arch: str): # Resolution moved to the resolution pipeline (arg_groups/overrides.py: # _mamba_radix_cache_resolution), invoked here at each legacy call