biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[Model] Merge SupportsMultiModalWithRawInput with SupportsMultiModal (#23749)

Browse Source 
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
This commit is contained in:
Cyrus Leung
2025-08-28 01:01:50 +08:00
committed by GitHub
parent 4f35be10a9
commit 52883ed084
5 changed files with 30 additions and 50 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
vllm/v1/worker/gpu_model_runner.py
View File

@@ -139,8 +139,9 @@ class GPUModelRunner(LoRAModelRunnerMixin, KVConnectorModelRunnerMixin):
cache_config.cache_dtype]
self.is_pooling_model = model_config.pooler_config is not None
self.is_multimodal_raw_input_supported = (
model_config.is_multimodal_raw_input_supported)
self.is_multimodal_raw_input_only_model = (
model_config.is_multimodal_raw_input_only_model)
self.max_model_len = model_config.max_model_len
self.max_num_tokens = scheduler_config.max_num_batched_tokens
self.max_num_reqs = scheduler_config.max_num_seqs
@@ -612,7 +613,7 @@ class GPUModelRunner(LoRAModelRunnerMixin, KVConnectorModelRunnerMixin):
self,
scheduler_output: "SchedulerOutput",
) -> BatchedTensorInputs:
if not self.is_multimodal_raw_input_supported or not scheduler_output: # noqa: SIM102
if not scheduler_output or not self.is_multimodal_raw_input_only_model:
return {}
mm_kwargs = list[MultiModalKwargsItem]()
@@ -631,8 +632,9 @@ class GPUModelRunner(LoRAModelRunnerMixin, KVConnectorModelRunnerMixin):
return mm_kwargs_combined
def _dummy_mm_kwargs(self, num_seqs: int) -> BatchedTensorInputs:
if not self.is_multimodal_raw_input_supported:
if not self.is_multimodal_raw_input_only_model:
return {}
mm_budget = self.mm_budget
assert mm_budget is not None