gemini-code-assist[bot] commented on code in PR #38110:
URL: https://github.com/apache/beam/pull/38110#discussion_r3185444977
##########
sdks/python/apache_beam/yaml/yaml_ml.py:
##########
@@ -282,6 +282,55 @@ def inference_output_type(self):
('model_id', Optional[str])])
[email protected]_handler_type('HuggingFacePipeline')
+class HuggingFacePipelineProvider(ModelHandlerProvider):
+ def __init__(
+ self,
+ task: Optional[str] = None,
+ model: Optional[str] = None,
+ preprocess: Optional[dict[str, str]] = None,
+ postprocess: Optional[dict[str, str]] = None,
+ device: Optional[Any] = None,
+ inference_fn: Optional[dict[str, str]] = None,
+ load_pipeline_args: Optional[dict[str, Any]] = None,
+ **kwargs):
+ try:
+ from apache_beam.ml.inference.huggingface_inference import
HuggingFacePipelineModelHandler
+ except ImportError:
+ raise ValueError(
+ 'Unable to import HuggingFacePipelineModelHandler. Please '
+ 'install transformers dependencies.')
+
+ kwargs = {k: v for k, v in kwargs.items() if not k.startswith('_')}
+
+ inference_fn_obj = self.parse_processing_transform(
+ inference_fn, 'inference_fn') if inference_fn else None
+
+ handler_kwargs = {}
+ if inference_fn_obj:
+ handler_kwargs['inference_fn'] = inference_fn_obj
+
+ _handler = HuggingFacePipelineModelHandler(
+ task=task or "",
+ model=model or "",
Review Comment:
Using `or ""` converts `None` values to empty strings. However, the
underlying `HuggingFacePipelineModelHandler` and the `transformers` library
typically expect `None` for optional parameters to trigger default behavior
(such as inferring the task from the model). Passing an empty string `""` as a
task or model name will likely cause a runtime error in the `transformers`
pipeline (e.g., a `KeyError` or model not found error).
```suggestion
task=task,
model=model,
```
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
