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

[VLM] Avoid unnecessary tokenization (#12310)

Browse Source 
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
This commit is contained in:
Cyrus Leung
2025-01-22 19:08:31 +08:00
committed by GitHub
parent 68ad4e3a8d
commit cd7b6f0857
9 changed files with 71 additions and 40 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
vllm/model_executor/models/fuyu.py
View File

@@ -183,7 +183,9 @@ class FuyuMultiModalProcessor(BaseMultiModalProcessor[FuyuProcessingInfo]):
) -> list[int]:
# HF processor adds boa_token_id
tokenizer = self.info.get_tokenizer()
boa_token_id: int = tokenizer.vocab["<0x04>"] # type: ignore
vocab = tokenizer.get_vocab()
boa_token_id = vocab["<0x04>"]
return prompt_tokens + [boa_token_id]
@@ -202,6 +204,7 @@ class FuyuMultiModalProcessor(BaseMultiModalProcessor[FuyuProcessingInfo]):
) -> list[PromptReplacement]:
hf_config = self.info.get_hf_config()
bos_token_id = hf_config.bos_token_id
assert isinstance(bos_token_id, int)
tokenizer = self.info.get_tokenizer()
eot_token_id = tokenizer.bos_token_id