riaz.somc/transformers
1
0
Fork 0
mirror of https://github.com/huggingface/transformers.git synced 2025-07-03 12:50:06 +06:00
Code Issues Actions 28 Packages Projects Releases Wiki Activity

[Whisper] another patch (#21324)

Browse Source 
* another patch

* fix timestamp test modeling

* let it be negative when the token is None
This commit is contained in:
Arthur 2023-01-27 16:35:16 +01:00 committed by GitHub
parent e5eb3e22ea
commit 0dff407d71
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
3 changed files with 5 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
src/transformers/generation/tf_logits_process.py
View File

@ -557,7 +557,8 @@ class TFForceTokensLogitsProcessor(TFLogitsProcessor):
# Indexes without forced tokens will have an negative value.
force_token_array = np.ones((max(force_token_map.keys()) + 1), dtype=np.int32) * -1
for index, token in force_token_map.items():
force_token_array[index] = token
if token is not None:
force_token_array[index] = token
self.force_token_array = tf.convert_to_tensor(force_token_array, dtype=tf.int32)
def __call__(self, input_ids: tf.Tensor, scores: tf.Tensor, cur_len: int) -> tf.Tensor:

2
src/transformers/pipelines/automatic_speech_recognition.py
View File

@ -101,7 +101,7 @@ def _find_timestamp_sequence(sequences, tokenizer, feature_extractor, max_source
chunk_len, stride_left, stride_right = stride
sequence = sequence.squeeze(0)
# get rid of the `forced_decoder_idx` that are use to parametrize the generation
begin_idx = np.where(sequence == timestamp_begin)[0].item() if timestamp_begin in sequence else 0
begin_idx = np.where(sequence == timestamp_begin)[0][0] if timestamp_begin in sequence else 0
sequence = sequence[begin_idx:]
timestamp_tokens = sequence >= timestamp_begin

6
tests/models/whisper/test_modeling_whisper.py
View File

@ -46,7 +46,6 @@ if is_torch_available():
WhisperProcessor,
set_seed,
)
from transformers.generation.logits_process import WhisperTimeStampLogitsProcessor
from transformers.models.whisper.modeling_whisper import WhisperDecoder, WhisperEncoder
@ -1077,9 +1076,8 @@ class WhisperModelIntegrationTests(unittest.TestCase):
input_features = processor.feature_extractor(raw_speech=input_speech, return_tensors="pt").input_features.to(
torch_device
)
model.config.forced_decoder_ids = [(1, 50259), (2, 50359), (3, 50364)]
timestamp_processor = [WhisperTimeStampLogitsProcessor(len(model.config.forced_decoder_ids))]
generated_ids = model.generate(input_features, max_length=448, logits_processor=timestamp_processor).to("cpu")
generated_ids = model.generate(input_features, max_length=448, return_timestamps=True).to("cpu")
# fmt: off
EXPECTED_OUTPUT = torch.tensor([50258, 50259, 50359, 50364, 2221, 13, 2326, 388, 391, 307, 264, 50244, 295, 264, 2808, 5359, 11, 293, 321, 366, 5404, 281, 2928, 702, 14943, 13, 50692, 50692, 6966, 307, 2221, 13, 2326, 388, 391, 311, 9060, 1570, 1880, 813, 702, 1871, 13, 50926, 50926, 634, 5112, 505, 300, 412, 341, 42729, 3196, 295, 264, 1064, 11, 365, 5272, 293, 12904, 9256, 450, 10539, 51208, 51208, 949, 505, 11, 14138, 10117, 490, 3936, 293, 1080, 3542, 5160, 881, 26336, 281, 264, 1575, 13, 51552, 51552, 634, 575, 12525, 22618, 1968, 6144, 35617, 7354, 1292, 6, 589, 307, 534, 10281, 934, 439, 11, 293, 51836, 51836, 50257])