[
https://issues.apache.org/jira/browse/BEAM-14337?focusedWorklogId=772796&page=com.atlassian.jira.plugin.system.issuetabpanels:worklog-tabpanel#worklog-772796
]
ASF GitHub Bot logged work on BEAM-14337:
-----------------------------------------
Author: ASF GitHub Bot
Created on: 20/May/22 12:17
Start Date: 20/May/22 12:17
Worklog Time Spent: 10m
Work Description: yeandy commented on code in PR #17470:
URL: https://github.com/apache/beam/pull/17470#discussion_r878083266
##########
sdks/python/apache_beam/ml/inference/base_test.py:
##########
@@ -39,7 +41,12 @@ class FakeInferenceRunner(base.InferenceRunner):
def __init__(self, clock=None):
self._mock_clock = clock
- def run_inference(self, batch: Any, model: Any) -> Iterable[Any]:
+ def run_inference(
+ self,
+ batch: Any,
+ model: Any,
+ prediction_params: Optional[Dict[str, Any]] = None,
Review Comment:
If I take out `prediction_params`, I get
```
result_generator = self._inference_runner.run_inference(
> examples, self._model, self._prediction_params)
E TypeError: run_inference() takes 3 positional arguments but 4 were
given [while running 'RunInference/ParDo(_RunInferenceDoFn)']
apache_beam/ml/inference/base.py:216: TypeError
```
If we allow this not to be a required param, then maybe we can check if
`_prediction_params` is `None` like so?
```
if not self._prediction_params:
result_generator = self._inference_runner.run_inference(
examples, self._model, self._prediction_params)
else:
result_generator = self._inference_runner.run_inference(
examples, self._model)
```
That feels a bit messy though, and breaks our `run_inference` interface,
assuming we want to use this modification:
```
class InferenceRunner():
def run_inference(
self,
batch: List[Any],
model: Any,
prediction_params: Optional[Dict[str, Any]] = None) -> Iterable[Any]:
"""Runs inferences on a batch of examples and
returns an Iterable of Predictions."""
raise NotImplementedError(type(self))
```
Issue Time Tracking
-------------------
Worklog Id: (was: 772796)
Time Spent: 2h 50m (was: 2h 40m)
> Support **kwargs for PyTorch models.
> ------------------------------------
>
> Key: BEAM-14337
> URL: https://issues.apache.org/jira/browse/BEAM-14337
> Project: Beam
> Issue Type: Sub-task
> Components: sdk-py-core
> Reporter: Anand Inguva
> Assignee: Andy Ye
> Priority: P2
> Time Spent: 2h 50m
> Remaining Estimate: 0h
>
> Some models in Pytorch instantiating from torch.nn.Module, has extra
> parameters in the forward function call. These extra parameters can be passed
> as Dict or as positional arguments.
> Example of PyTorch models supported by Hugging Face ->
> [https://huggingface.co/bert-base-uncased]
> [Some torch models on Hugging
> face|https://github.com/huggingface/transformers/blob/main/src/transformers/models/bert/modeling_bert.py]
> Eg:
> [https://huggingface.co/docs/transformers/model_doc/bert#transformers.BertModel]
> {code:java}
> inputs = {
> input_ids: Tensor1,
> attention_mask: Tensor2,
> token_type_ids: Tensor3,
> }
> model = BertModel.from_pretrained("bert-base-uncased") # which is a
> # subclass of torch.nn.Module
> outputs = model(**inputs) # model forward method should be expecting the keys
> in the inputs as the positional arguments.{code}
>
> [Transformers|https://pytorch.org/hub/huggingface_pytorch-transformers/]
> integrated in Pytorch is supported by Hugging Face as well.
>
--
This message was sent by Atlassian Jira
(v8.20.7#820007)
