@tduncan / Nvidia-Optimization-Stack-VLLM Mirror · read-only / commits / fb30ee92ee
Fork 0

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 16, 00:45 UTC

[Bugfix] Fix BLIP-2 processing (#12412)

Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>

Cyrus Leung committed Jan 25, 2025 at 11:42 UTC fb30ee92eefec7eacc0d7483f9d07daa1206530f
1 file changed +2 -2
vllm/model_executor/models/blip2.py
+2 -2
@@ -481,14 +481,14 @@ class Blip2MultiModalProcessor(BaseMultiModalProcessor[Blip2ProcessingInfo]):
481 bos_token_id = tokenizer.bos_token_id
482 assert isinstance(bos_token_id, int)
483
484 - image_token_id = vocab["image"]
484 + image_token_id = vocab["<image>"]
485 num_image_tokens = self.info.get_num_image_tokens()
486 image_tokens = [image_token_id] * num_image_tokens
487
488 return [
489 PromptReplacement(
490 modality="image",
491 - target="</s>",
491 + target=[bos_token_id],
492 replacement=PromptReplacementDetails(
493 full=image_tokens + [bos_token_id],
494 features=image_tokens,