A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 12, 13:30 UTC
[BugFix] Fix initialization of draft model. (#29319)
Signed-off-by: Andrey Khalyavin <halyavin@yandex-team.ru> Signed-off-by: Tyler Michael Smith <tlrmchlsmth@gmail.com> Co-authored-by: Tyler Michael Smith <tlrmchlsmth@gmail.com>
Andrey Khalyavin committed
Nov 26, 2025 at 02:45 UTC
de75b0bb701c89c1b2bffe09b23f86446c951f73
1 file changed
+4
vllm/v1/worker/gpu_model_runner.py
+4
@@ -3460,6 +3460,10 @@ class GPUModelRunner(
3460
scope="local",
3461
)
3462
prepare_communication_buffer_for_model(self.model)
3463
+ if (drafter := getattr(self, "drafter", None)) and (
3464
+ drafter_model := getattr(drafter, "model", None)
3465
+ ):
3466
+ prepare_communication_buffer_for_model(drafter_model)
3467
mm_config = self.model_config.multimodal_config
3468
self.is_multimodal_pruning_enabled = (
3469
supports_multimodal_pruning(self.get_model())