riaz.somc/transformers
1
0
Fork 0
mirror of https://github.com/huggingface/transformers.git synced 2025-08-02 03:01:07 +06:00
Code Issues Actions 3 Packages Projects Releases Wiki Activity

Better defaults (#34026)

Browse Source 
* be nice to our usres

* nit

* fixup

* default to -1

* oups

* turbo nit

* auto infer framework
This commit is contained in:
Arthur 2024-10-24 11:11:55 +02:00 committed by GitHub
parent 65753d6065
commit 05863817d6
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
2 changed files with 6 additions and 13 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
src/transformers/generation/utils.py
View File

@ -1440,6 +1440,8 @@ class GenerationMixin:
and not self.config.is_encoder_decoder and not self.config.is_encoder_decoder
): ):
generation_config.max_length -= inputs_tensor.shape[1] generation_config.max_length -= inputs_tensor.shape[1]
else: # by default let's always generate 10 new tokens
generation_config.max_length = generation_config.max_length + input_ids_length
# same for min length # same for min length
if generation_config.min_new_tokens is not None: if generation_config.min_new_tokens is not None:

17
src/transformers/pipelines/base.py
View File

@ -881,18 +881,7 @@ class Pipeline(_ScikitCompat, PushToHubMixin):
# Take the first device used by `accelerate`. # Take the first device used by `accelerate`.
device = next(iter(hf_device_map.values())) device = next(iter(hf_device_map.values()))
else: else:
device = -1 device = 0
if (
is_torch_mlu_available()
or is_torch_cuda_available()
or is_torch_npu_available()
or is_torch_xpu_available(check_device=True)
or is_torch_mps_available()
):
logger.warning(
"Hardware accelerator e.g. GPU is available in the environment, but no `device` argument"
" is passed to the `Pipeline` object. Model will be on CPU."
)
if is_torch_available() and self.framework == "pt": if is_torch_available() and self.framework == "pt":
if device == -1 and self.model.device is not None: if device == -1 and self.model.device is not None:
@ -920,10 +909,12 @@ class Pipeline(_ScikitCompat, PushToHubMixin):
elif is_torch_mps_available(): elif is_torch_mps_available():
self.device = torch.device(f"mps:{device}") self.device = torch.device(f"mps:{device}")
else: else:
raise ValueError(f"{device} unrecognized or not available.") self.device = torch.device("cpu")
else: else:
self.device = device if device is not None else -1 self.device = device if device is not None else -1
logger.warning(f"Device set to use {self.device}")
self.binary_output = binary_output self.binary_output = binary_output
# We shouldn't call `model.to()` for models loaded with accelerate as well as the case that model is already on device # We shouldn't call `model.to()` for models loaded with accelerate as well as the case that model is already on device
if ( if (