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

[Multimodal] Always enable hashing mm data (#23308)

Browse Source 
Signed-off-by: Roger Wang <hey@rogerw.io>
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
Co-authored-by: DarkLight1337 <tlleungac@connect.ust.hk>
This commit is contained in:
Roger Wang
2025-08-21 07:23:28 -07:00
committed by GitHub
parent f8daddcc4c
commit 79f05e4436
15 changed files with 94 additions and 148 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
vllm/model_executor/models/deepseek_vl2.py
View File

@@ -290,8 +290,6 @@ class DeepseekVL2MultiModalProcessor(
mm_data_items: MultiModalDataItems,
hf_processor_mm_kwargs: Mapping[str, object],
tokenization_kwargs: Mapping[str, object],
*,
return_mm_hashes: bool,
) -> tuple[list[int], MultiModalProcessingInfo, bool]:
# The processor logic is different for len(images) <= 2 vs > 2
# Since the processing cache assumes that the processor output is
@@ -303,7 +301,6 @@ class DeepseekVL2MultiModalProcessor(
mm_data_items=mm_data_items,
hf_processor_mm_kwargs=hf_processor_mm_kwargs,
tokenization_kwargs=tokenization_kwargs,
return_mm_hashes=return_mm_hashes,
)
return super()._cached_apply_hf_processor(
@@ -311,7 +308,6 @@ class DeepseekVL2MultiModalProcessor(
mm_data_items=mm_data_items,
hf_processor_mm_kwargs=hf_processor_mm_kwargs,
tokenization_kwargs=tokenization_kwargs,
return_mm_hashes=return_mm_hashes,
)