[Spec][V2] Support MTP speculative decoding under pipeline parallelism (#46994)

Signed-off-by: Chris Eastwood <chris.eastwood@pwn4g3.dev>
Signed-off-by: Nick Hill <nickhill123@gmail.com>
Co-authored-by: Nick Hill <nickhill123@gmail.com>
Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
This commit is contained in:
Chris Eastwood
2026-09-11 17:55:21 +00:00
committed by GitHub
co-authored by Nick Hill mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
parent 9a35c081e8
commit 6fe67cbbf3
12 changed files with 167 additions and 49 deletions
+7 -1
View File
@@ -2030,7 +2030,13 @@ class GPUModelRunner(LoRAModelRunnerMixin):
)
mm_inputs: tuple[list[torch.Tensor], torch.Tensor] | None = None
if self.speculator is not None and self.speculator.supports_mm_inputs:
# The encoder runner exists only on the first PP rank, so later ranks
# have no cached embeddings to gather.
if (
self.speculator is not None
and self.speculator.supports_mm_inputs
and self.model_state.supports_mm_inputs
):
# Get cached multimodal embeddings for draft forward.
# NOTE: This is done here because postprocess updates
# num_computed_prefill_tokens.