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

[Refactor] Pass tokenizer explicitly instead of binding to prompt update (#23542)

Browse Source 
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
This commit is contained in:
Cyrus Leung
2025-08-25 21:31:57 +08:00
committed by GitHub
parent e269be2ba2
commit 6879cd80ae
4 changed files with 95 additions and 144 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
vllm/model_executor/models/gemma3n_mm.py
View File

@@ -35,7 +35,6 @@ from vllm.multimodal.processing import (BaseMultiModalProcessor,
PlaceholderFeaturesInfo,
PromptReplacement, PromptUpdate,
PromptUpdateDetails,
find_mm_placeholders,
replace_token_matches)
# yapf: enable
from vllm.multimodal.profiling import BaseDummyInputsBuilder
@@ -318,7 +317,8 @@ class Gemma3nMultiModalProcessor(BaseMultiModalProcessor[Gemma3nProcessingInfo]
repl_token_ids.extend(repl_toks)
repl_orig_idxs.extend(orig_idx for _ in range(len(repl_toks)))
repls = find_mm_placeholders(repl_token_ids, mm_prompt_updates)
repls = super()._find_mm_placeholders(repl_token_ids,
mm_prompt_updates)
return {
modality: [