gemini-code-assist[bot] commented on code in PR #38110:
URL: https://github.com/apache/beam/pull/38110#discussion_r3189334925
##########
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,
+ model=model,
+ device=device,
+ load_pipeline_args=load_pipeline_args,
Review Comment:
Update this to pass `pipeline_kwargs` to the model handler to match the
underlying API.
```suggestion
pipeline_kwargs=pipeline_kwargs,
```
##########
sdks/python/apache_beam/yaml/tests/runinference_huggingface.yaml:
##########
@@ -0,0 +1,62 @@
+# Licensed to the Apache Software Foundation (ASF) under one or more
+# contributor license agreements. See the NOTICE file distributed with
+# this work for additional information regarding copyright ownership.
+# The ASF licenses this file to You under the Apache License, Version 2.0
+# (the "License"); you may not use this file except in compliance with
+# the License. You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+pipelines:
+ - pipeline:
+ type: chain
+ transforms:
+ - type: Create
+ config:
+ elements:
+ - text: "I love Apache Beam!"
+ - text: "I hate this error."
+ - type: RunInference
+ config:
+ model_handler:
+ type: "HuggingFacePipeline"
+ config:
+ task: "text-classification"
+ inference_fn:
+ callable: |
+ def real_inference(batch, pipeline, inference_args):
+ predictions = pipeline(batch, **inference_args)
+
+ # If it's a single dictionary (batch size of 1), wrap it
in a list
+ if isinstance(predictions, dict):
+ predictions = [predictions]
+
+ return {
+ 'label': [p['label'] for p in predictions],
+ 'score': [p['score'] for p in predictions]
+ }
+ preprocess:
+ callable: 'lambda x: x.text'
+ - type: MapToFields
+ config:
+ language: python
+ fields:
+ text: text
+ sentiment:
+ callable: 'lambda x: x.inference.inference["label"]'
Review Comment:
If the `inference_fn` is updated to return a list of dictionaries, the
prediction for each element will be a dictionary accessible via `x.inference`.
The double `.inference` access is only necessary when using the default handler
which returns `PredictionResult` objects. If you intend to use the default
handler's behavior, consider removing the custom `inference_fn` entirely.
```yaml
callable: 'lambda x: x.inference["label"]'
```
##########
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,
Review Comment:
The parameter name `load_pipeline_args` does not match the expected argument
`pipeline_kwargs` in the `HuggingFacePipelineModelHandler` constructor. To
ensure compatibility and clarity, it's better to use `pipeline_kwargs` or map
this parameter to it.
```suggestion
pipeline_kwargs: Optional[dict[str, Any]] = None,
```
##########
sdks/python/apache_beam/yaml/tests/runinference_huggingface.yaml:
##########
@@ -0,0 +1,62 @@
+# Licensed to the Apache Software Foundation (ASF) under one or more
+# contributor license agreements. See the NOTICE file distributed with
+# this work for additional information regarding copyright ownership.
+# The ASF licenses this file to You under the Apache License, Version 2.0
+# (the "License"); you may not use this file except in compliance with
+# the License. You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+pipelines:
+ - pipeline:
+ type: chain
+ transforms:
+ - type: Create
+ config:
+ elements:
+ - text: "I love Apache Beam!"
+ - text: "I hate this error."
+ - type: RunInference
+ config:
+ model_handler:
+ type: "HuggingFacePipeline"
+ config:
+ task: "text-classification"
+ inference_fn:
+ callable: |
+ def real_inference(batch, pipeline, inference_args):
+ predictions = pipeline(batch, **inference_args)
+
+ # If it's a single dictionary (batch size of 1), wrap it
in a list
+ if isinstance(predictions, dict):
+ predictions = [predictions]
+
+ return {
+ 'label': [p['label'] for p in predictions],
+ 'score': [p['score'] for p in predictions]
+ }
Review Comment:
The `inference_fn` must return an iterable (e.g., a list) of predictions
with the same length as the input batch. Currently, it returns a single
dictionary for the entire batch, which will cause `RunInference` to incorrectly
zip the results (e.g., zipping batch elements with the dictionary keys 'label'
and 'score' instead of the actual predictions).
```yaml
def real_inference(batch, pipeline, inference_args):
predictions = pipeline(batch, **inference_args)
# If it's a single dictionary (batch size of 1), wrap
it in a list
if isinstance(predictions, dict):
predictions = [predictions]
return predictions
```
--
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]
