Skip to content

Commit ecb5914

Browse files
committed
fix: Correct logic for checkpoint_format handling in create_input_processor
- Updated the condition to check for "HF" explicitly, removing the None case handling. Signed-off-by: Robin Kobus <[email protected]>
1 parent e1ea0b8 commit ecb5914

File tree

1 file changed

+2
-2
lines changed

1 file changed

+2
-2
lines changed

tensorrt_llm/inputs/registry.py

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -449,7 +449,7 @@ def create_input_processor(
449449
model_path_or_dir: Path or repo id used to locate pretrained config/tokenizer.
450450
tokenizer: Tokenizer instance.
451451
checkpoint_format: Checkpoint format identifier. "HF" uses Hugging Face-style
452-
config loading; any other value skips HF config loading. None is treated as "HF".
452+
config loading; any other value skips HF config loading.
453453
454454
Returns:
455455
An InputProcessor implementation (model-specific if registered; otherwise DefaultInputProcessor).
@@ -459,7 +459,7 @@ def create_input_processor(
459459

460460
model_config = None
461461

462-
if checkpoint_format is None or checkpoint_format == "HF":
462+
if checkpoint_format == "HF":
463463
try:
464464
config = ModelConfig.from_pretrained(model_path_or_dir,
465465
trust_remote_code=True)

0 commit comments

Comments
 (0)