Spaces:
Runtime error
Runtime error
Replace deprecated `t5x.partitioning.ModelBasedPjitPartitioner` with `t5x.partitioning.PjitPartitioner`
#2
by
deleted
- opened
app.py
CHANGED
|
@@ -85,7 +85,7 @@ class InferenceModel(object):
|
|
| 85 |
self.sequence_length = {'inputs': self.inputs_length,
|
| 86 |
'targets': self.outputs_length}
|
| 87 |
|
| 88 |
-
self.partitioner = t5x.partitioning.
|
| 89 |
model_parallel_submesh=(1, 1, 1, 1), num_partitions=1)
|
| 90 |
|
| 91 |
# Build Codecs and Vocabularies.
|
|
@@ -178,7 +178,6 @@ class InferenceModel(object):
|
|
| 178 |
|
| 179 |
Args:
|
| 180 |
audio: 1-d numpy array of audio samples (16kHz) for a single example.
|
| 181 |
-
|
| 182 |
Returns:
|
| 183 |
A note_sequence of the transcribed audio.
|
| 184 |
"""
|
|
|
|
| 85 |
self.sequence_length = {'inputs': self.inputs_length,
|
| 86 |
'targets': self.outputs_length}
|
| 87 |
|
| 88 |
+
self.partitioner = t5x.partitioning.PjitPartitioner(
|
| 89 |
model_parallel_submesh=(1, 1, 1, 1), num_partitions=1)
|
| 90 |
|
| 91 |
# Build Codecs and Vocabularies.
|
|
|
|
| 178 |
|
| 179 |
Args:
|
| 180 |
audio: 1-d numpy array of audio samples (16kHz) for a single example.
|
|
|
|
| 181 |
Returns:
|
| 182 |
A note_sequence of the transcribed audio.
|
| 183 |
"""
|