[MM] Cache common token sequences (#53560)

Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
Signed-off-by: Cyrus Leung <tlleungac@connect.ust.hk>
This commit is contained in:
Cyrus Leung
2026-08-24 07:48:10 -07:00
committed by GitHub
parent 4ca856b0b5
commit ecfa7bb373
14 changed files with 40 additions and 23 deletions
+3 -1
View File
@@ -91,7 +91,9 @@ class BaseDummyInputsBuilder(ABC, Generic[_I]):
# prompt tokens.
dummy_prompt = []
else:
dummy_prompt = tokenizer.encode(dummy_text, truncation=False)
from .processor import cached_encode
dummy_prompt = cached_encode(tokenizer, dummy_text, truncation=False)
return ProcessorInputs(
prompt=dummy_prompt,
+6 -1
View File
@@ -57,10 +57,15 @@ def cached_encode(
tokenizer: TokenizerLike,
text: str,
*,
truncation: bool | None = None,
add_special_tokens: bool = True,
) -> list[int]:
"""Encode text while caching repeated tokenizer calls."""
return tokenizer.encode(text, add_special_tokens=add_special_tokens)
return tokenizer.encode(
text,
truncation=truncation,
add_special_tokens=add_special_tokens,
)
class _GetMatchIndex(Protocol):