[Misc] add use_tqdm_on_load to reduce logs (#14407)

Signed-off-by: Aaron Pham <contact@aarnphm.xyz>
This commit is contained in:
Aaron Pham
2025-03-08 05:57:46 -08:00
committed by GitHub
parent 03fe18ae0f
commit 0b7f06b447
4 changed files with 54 additions and 22 deletions
+21 -5
View File
@@ -354,11 +354,18 @@ class DefaultModelLoader(BaseModelLoader):
self.load_config.download_dir,
hf_folder,
hf_weights_files,
self.load_config.use_tqdm_on_load,
)
elif use_safetensors:
weights_iterator = safetensors_weights_iterator(hf_weights_files)
weights_iterator = safetensors_weights_iterator(
hf_weights_files,
self.load_config.use_tqdm_on_load,
)
else:
weights_iterator = pt_weights_iterator(hf_weights_files)
weights_iterator = pt_weights_iterator(
hf_weights_files,
self.load_config.use_tqdm_on_load,
)
if current_platform.is_tpu():
# In PyTorch XLA, we should call `xm.mark_step` frequently so that
@@ -806,9 +813,15 @@ class BitsAndBytesModelLoader(BaseModelLoader):
def _hf_weight_iter(self, hf_weights_files, use_safetensors: bool):
if use_safetensors:
iterator = safetensors_weights_iterator(hf_weights_files)
iterator = safetensors_weights_iterator(
hf_weights_files,
self.load_config.use_tqdm_on_load,
)
else:
iterator = pt_weights_iterator(hf_weights_files)
iterator = pt_weights_iterator(
hf_weights_files,
self.load_config.use_tqdm_on_load,
)
for org_name, param in iterator:
# mapping weight names from transformers to vllm while preserving
# original names.
@@ -1396,7 +1409,10 @@ class RunaiModelStreamerLoader(BaseModelLoader):
revision: str) -> Generator[Tuple[str, torch.Tensor], None, None]:
"""Get an iterator for the model weights based on the load format."""
hf_weights_files = self._prepare_weights(model_or_path, revision)
return runai_safetensors_weights_iterator(hf_weights_files)
return runai_safetensors_weights_iterator(
hf_weights_files,
self.load_config.use_tqdm_on_load,
)
def download_model(self, model_config: ModelConfig) -> None:
"""Download model if necessary"""