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

[Misc] Clean up input processing (#17582)

Browse Source 
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
This commit is contained in:
Cyrus Leung
2025-05-02 23:11:53 +08:00
committed by GitHub
parent 3a500cd0b6
commit cb234955df
9 changed files with 359 additions and 285 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
vllm/multimodal/processing.py
View File

@@ -1670,15 +1670,17 @@ class BaseMultiModalProcessor(ABC, Generic[_I]):
placeholders = mm_placeholders.get(modality, [])
if len(placeholders) != item_count:
# NOTE: If you are a model developer, this can also arise from
# an inconsistency between `_call_hf_processor` and
# `_get_mm_fields_config` implementations
raise RuntimeError(
f"Expected there to be {item_count} prompt updates "
f"corresponding to {item_count} {modality} items, but "
f"instead found {len(placeholders)} prompt updates! "
"Either the prompt text has missing/incorrect tokens for "
"multi-modal inputs, or there is a problem with your "
"implementation of merged multi-modal processor for this "
"model (usually arising from an inconsistency between "
"`_call_hf_processor` and `_get_prompt_updates`).")
"This is likely because you forgot to include input "
"placeholder tokens (e.g., `<image>`, `<|image_pad|>`) "
"in the prompt. If the model has a chat template, make "
"sure you have applied it before calling `LLM.generate`.")
def _maybe_apply_prompt_updates(
self,