Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

fix: clamp keep input size in update_cache for causal conv #5732

Closed
wants to merge 3 commits into from
Closed

fix: clamp keep input size in update_cache for causal conv #5732

wants to merge 3 commits into from

Conversation

messiaen
Copy link
Contributor

@messiaen messiaen commented Jan 4, 2023

What does this PR do ?

Sometimes in CausalConv1D.update_cache input_x_keep ends up having no frames (ie size of [M, N, 0]). Make sure that the we have at least one frame.

Collection: asr

Changelog

  • Add specific line by line info of high level changes in this PR.

Usage

  • You can potentially add a usage example below
# Add a code snippet demonstrating how to use this

Before your PR is "Ready for review"

Pre checks:

  • Make sure you read and followed Contributor guidelines
  • Did you write any new necessary tests?
  • Did you add or update any necessary documentation?
  • Does the PR affect components that are optional to install? (Ex: Numba, Pynini, Apex etc)
    • Reviewer: Does the PR have correct import guards for all optional libraries?

PR Type:

  • New Feature
  • Bugfix
  • Documentation

If you haven't finished some of the above items you can still open "Draft" PR.

Who can review?

Anyone in the community is free to review the PR once the checks have passed.
Contributor guidelines contains specific people who can review PRs to various areas.
@VahidooX

Additional Information

To reproduce the original issue in main run:

#!/bin/bash
python examples/asr/asr_cache_aware_streaming/speech_to_text_cache_aware_streaming_infer.py \
    --asr_model=stt_en_conformer_ctc_small \
    --chunk_size=100 \
    --shift_size=50 \
    --left_chunks=2 \
    --online_normalization \
    --manifest_file=/datasets/ls_test_other/transcripts.local.json \
    --batch_size=16 \
    --compare_vs_offline \
    --use_amp \
    --debug_mode

Error output:

...
Traceback (most recent call last):
  File "examples/asr/asr_cache_aware_streaming/speech_to_text_cache_aware_streaming_infer.py", line 393, in <module>
    main()
  File "examples/asr/asr_cache_aware_streaming/speech_to_text_cache_aware_streaming_infer.py", line 349, in main
    streaming_tran, offline_tran = perform_streaming(
  File "examples/asr/asr_cache_aware_streaming/speech_to_text_cache_aware_streaming_infer.py", line 154, in perform_streaming
    ) = asr_model.conformer_stream_step(
  File "/home/grclark/code/NeMo.git/streaming-conformer/nemo/collections/asr/parts/mixins/mixins.py", line 475, in conformer_stream_step
    (encoded, encoded_len, cache_last_channel_next, cache_last_time_next) = self.encoder.cache_aware_stream_step(
  File "/home/grclark/code/NeMo.git/streaming-conformer/nemo/collections/asr/parts/mixins/streaming.py", line 61, in cache_aware_stream_step
    encoder_output = self(
  File "/usr/local/lib/python3.8/dist-packages/torch/nn/modules/module.py", line 1190, in _call_impl
    return forward_call(*input, **kwargs)
  File "/home/grclark/code/NeMo.git/streaming-conformer/nemo/core/classes/common.py", line 1087, in __call__
    outputs = wrapped(*args, **kwargs)
  File "/home/grclark/code/NeMo.git/streaming-conformer/nemo/collections/asr/modules/conformer_encoder.py", line 471, in forward
    audio_signal = layer(
  File "/usr/local/lib/python3.8/dist-packages/torch/nn/modules/module.py", line 1190, in _call_impl
    return forward_call(*input, **kwargs)
  File "/home/grclark/code/NeMo.git/streaming-conformer/nemo/collections/asr/parts/submodules/conformer_modules.py", line 191, in forward
    x = self.conv(x, pad_mask=pad_mask, cache=cache_last_time, cache_next=cache_last_time_next)
  File "/usr/local/lib/python3.8/dist-packages/torch/nn/modules/module.py", line 1190, in _call_impl
    return forward_call(*input, **kwargs)
  File "/home/grclark/code/NeMo.git/streaming-conformer/nemo/collections/asr/parts/submodules/conformer_modules.py", line 350, in forward
    x = self.depthwise_conv(x, cache=cache, cache_next=cache_next)
  File "/usr/local/lib/python3.8/dist-packages/torch/nn/modules/module.py", line 1190, in _call_impl
    return forward_call(*input, **kwargs)
  File "/home/grclark/code/NeMo.git/streaming-conformer/nemo/collections/asr/parts/submodules/causal_convs.py", line 162, in forward
    x = self.update_cache(x, cache=cache, cache_next=cache_next)
  File "/home/grclark/code/NeMo.git/streaming-conformer/nemo/collections/asr/parts/submodules/causal_convs.py", line 158, in update_cache
    cache_next[self._cache_id, :, :, -cache_keep_size:] = input_x_kept[:, :, -cache_keep_size:]
RuntimeError: The expanded size of the tensor (1) must match the existing size (0) at non-singleton dimension 2.  Target sizes: [16, 176, 1].  Tensor sizes: [16, 176, 0]

@github-actions github-actions bot added the ASR label Jan 4, 2023
github-advanced-security[bot]
github-advanced-security bot found potential problems Jan 4, 2023
View reviewed changes
nemo/collections/asr/parts/submodules/causal_convs.py
@@ -148,7 +148,9 @@
x = torch.cat((needed_cache, x), dim=-1)

if cache_next is not None:
input_x_kept = input_x[:, :, : input_x.size(-1) - self.cache_drop_size]
input_x_size = torch.tensor(input_x.size(-1) - self.cache_drop_size, dtype=torch.int64)

Check failure

Code scanning / CodeQL

Potentially uninitialized local variable

Local variable 'input_x' may be used before it is initialized.
VahidooX
VahidooX requested changes Jan 5, 2023
View reviewed changes
nemo/collections/asr/parts/submodules/causal_convs.py
@@ -148,7 +148,9 @@ def update_cache(self, x, cache=None, cache_next=None):
x = torch.cat((needed_cache, x), dim=-1)

if cache_next is not None:
input_x_kept = input_x[:, :, : input_x.size(-1) - self.cache_drop_size]
input_x_size = torch.tensor(input_x.size(-1) - self.cache_drop_size, dtype=torch.int64)
input_x_size = input_x_size.clip(min=1, max=input_x.size(-1))
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

There is no need to specify max for the clip.

nemo/collections/asr/parts/submodules/causal_convs.py
input_x_kept = input_x[:, :, : input_x.size(-1) - self.cache_drop_size]
input_x_size = torch.tensor(input_x.size(-1) - self.cache_drop_size, dtype=torch.int64)
input_x_size = input_x_size.clip(min=1, max=input_x.size(-1))
input_x_kept = input_x[:, :, :input_x_size]
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Would you please check it with export nemo to make sure the ONNX conversion is still working?

nemo/collections/asr/parts/submodules/causal_convs.py
@@ -148,7 +148,9 @@ def update_cache(self, x, cache=None, cache_next=None):
x = torch.cat((needed_cache, x), dim=-1)

if cache_next is not None:
input_x_kept = input_x[:, :, : input_x.size(-1) - self.cache_drop_size]
input_x_size = torch.tensor(input_x.size(-1) - self.cache_drop_size, dtype=torch.int64)
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Do you get the error with a specific file or any causes this?
What are the values of self.cache_drop_size and input_x.size(-1) here?

@github-actions
Copy link
Contributor

This PR is stale because it has been open for 14 days with no activity. Remove stale label or comment or update or this will be closed in 7 days.

@github-actions github-actions bot added the stale label Jan 19, 2023
@messiaen
Copy link
Contributor Author

Closing. Will recreate.

@messiaen messiaen closed this Jan 20, 2023
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@VahidooX VahidooX VahidooX requested changes

@titu1994 titu1994 titu1994 approved these changes

Assignees
No one assigned
Labels
ASR stale
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.

None yet
3 participants
@messiaen @titu1994 @VahidooX