Migrate transducer input checks to C++ #1391
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
carolineechen
force-pushed
the
migrate_transducer_input_checks
branch
from
2d15d0f to
ce61ffe
Compare
mthrok
reviewed
Mar 15, 2021
mthrok
approved these changes
Mar 15, 2021
vincentqb
reviewed
Mar 15, 2021
torchaudio/csrc/transducer.cpp
Outdated
Comment on lines
31
to
36
| TORCH_CHECK( | ||
| input_lengths.size(0) == acts.size(0), | ||
| "each output sequence must have a length"); | ||
| TORCH_CHECK( | ||
| label_lengths.size(0) == acts.size(0), | ||
| "each example must have a label length"); |
There was a problem hiding this comment.
Let's be explicit and consistent with python naming:
"batch dimension mismatch between acts and act_lens: each example must have a length"
"batch dimension mismatch between acts and label_lens: each example must have a label length"
There was a problem hiding this comment.
(As follow-up to this PR: we'll change the names of the variables in the C++ function to match those of the python function:
"""
Args:
acts (Tensor): Tensor of dimension (batch, time, label, class) containing output from network
before applying ``torch.nn.functional.log_softmax``.
labels (Tensor): Tensor of dimension (batch, max label length) containing the labels padded by zero
act_lens (Tensor): Tensor of dimension (batch) containing the length of each output sequence
label_lens (Tensor): Tensor of dimension (batch) containing the length of each output sequence
"""
and, also, I just realized the two last descriptions are the same in the python documentation :)
"""
Args:
acts (Tensor): Tensor of dimension (batch, time, label, class) containing output *sequence* from network
before applying ``torch.nn.functional.log_softmax``.
labels (Tensor): Tensor of dimension (batch, max label length) containing the labels padded by zero
act_lens (Tensor): Tensor of dimension (batch) containing the length of each output sequence
label_lens (Tensor): Tensor of dimension (batch) containing the length of each *label*
"""
added * for suggested change.)
| int maxT = acts.size(1); | ||
| int maxU = acts.size(2); | ||
| int minibatch_size = acts.size(0); | ||
| int alphabet_size = acts.size(3); | ||
|
|
||
| TORCH_CHECK( | ||
| at::max(input_lengths).item().toInt() == maxT, "input length mismatch"); |
There was a problem hiding this comment.
(Follow-up beyond this PR: let's improve the readability here: "The maximum length of a sequence in acts must be equal to the maximal value given in act_lens" ?)
| at::max(input_lengths).item().toInt() == maxT, "input length mismatch"); | ||
| TORCH_CHECK( | ||
| at::max(label_lengths).item().toInt() + 1 == maxU, | ||
| "output length mismatch"); |
There was a problem hiding this comment.
(Follow-up beyond this PR: we'll want to improve this message too)
carolineechen
force-pushed
the
migrate_transducer_input_checks
branch
from
dbaf97c to
21b5abf
Compare
carolineechen
force-pushed
the
migrate_transducer_input_checks
branch
from
21b5abf to
9b72d80
Compare
mthrok
pushed a commit
to mthrok/audio
that referenced
this pull request
Dec 13, 2022
* Update index.rst * Update layout.html
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Move input checks for RNNT from Python to C++
RNN Transducer Loss Issue: #1240