vriveras commited on
Commit
e1f72dc
1 Parent(s): bbb1c85

Initial models

Browse files
README.md ADDED
@@ -0,0 +1,445 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ language:
3
+ - en
4
+ - zh
5
+ - de
6
+ - es
7
+ - ru
8
+ - ko
9
+ - fr
10
+ - ja
11
+ - pt
12
+ - tr
13
+ - pl
14
+ - ca
15
+ - nl
16
+ - ar
17
+ - sv
18
+ - it
19
+ - id
20
+ - hi
21
+ - fi
22
+ - vi
23
+ - he
24
+ - uk
25
+ - el
26
+ - ms
27
+ - cs
28
+ - ro
29
+ - da
30
+ - hu
31
+ - ta
32
+ - no
33
+ - th
34
+ - ur
35
+ - hr
36
+ - bg
37
+ - lt
38
+ - la
39
+ - mi
40
+ - ml
41
+ - cy
42
+ - sk
43
+ - te
44
+ - fa
45
+ - lv
46
+ - bn
47
+ - sr
48
+ - az
49
+ - sl
50
+ - kn
51
+ - et
52
+ - mk
53
+ - br
54
+ - eu
55
+ - is
56
+ - hy
57
+ - ne
58
+ - mn
59
+ - bs
60
+ - kk
61
+ - sq
62
+ - sw
63
+ - gl
64
+ - mr
65
+ - pa
66
+ - si
67
+ - km
68
+ - sn
69
+ - yo
70
+ - so
71
+ - af
72
+ - oc
73
+ - ka
74
+ - be
75
+ - tg
76
+ - sd
77
+ - gu
78
+ - am
79
+ - yi
80
+ - lo
81
+ - uz
82
+ - fo
83
+ - ht
84
+ - ps
85
+ - tk
86
+ - nn
87
+ - mt
88
+ - sa
89
+ - lb
90
+ - my
91
+ - bo
92
+ - tl
93
+ - mg
94
+ - as
95
+ - tt
96
+ - haw
97
+ - ln
98
+ - ha
99
+ - ba
100
+ - jw
101
+ - su
102
+ tags:
103
+ - audio
104
+ - automatic-speech-recognition
105
+ - hf-asr-leaderboard
106
+ widget:
107
+ - example_title: Librispeech sample 1
108
+ src: https://cdn-media.huggingface.co/speech_samples/sample1.flac
109
+ - example_title: Librispeech sample 2
110
+ src: https://cdn-media.huggingface.co/speech_samples/sample2.flac
111
+ model-index:
112
+ - name: whisper-base
113
+ results:
114
+ - task:
115
+ name: Automatic Speech Recognition
116
+ type: automatic-speech-recognition
117
+ dataset:
118
+ name: LibriSpeech (clean)
119
+ type: librispeech_asr
120
+ config: clean
121
+ split: test
122
+ args:
123
+ language: en
124
+ metrics:
125
+ - name: Test WER
126
+ type: wer
127
+ value: 5.008769117619326
128
+ - task:
129
+ name: Automatic Speech Recognition
130
+ type: automatic-speech-recognition
131
+ dataset:
132
+ name: LibriSpeech (other)
133
+ type: librispeech_asr
134
+ config: other
135
+ split: test
136
+ args:
137
+ language: en
138
+ metrics:
139
+ - name: Test WER
140
+ type: wer
141
+ value: 12.84936273212057
142
+ - task:
143
+ name: Automatic Speech Recognition
144
+ type: automatic-speech-recognition
145
+ dataset:
146
+ name: Common Voice 11.0
147
+ type: mozilla-foundation/common_voice_11_0
148
+ config: hi
149
+ split: test
150
+ args:
151
+ language: hi
152
+ metrics:
153
+ - name: Test WER
154
+ type: wer
155
+ value: 131
156
+ pipeline_tag: automatic-speech-recognition
157
+ license: apache-2.0
158
+ ---
159
+
160
+ # Model summary
161
+ This Whisper-base model has been optimized to work with WebNN. This model is licensed under the [Apache-2.0 license](https://huggingface.co/datasets/choosealicense/licenses/blob/main/markdown/apache-2.0.md). For terms of use, please visit the [Intended use](https://huggingface.co/openai/whisper-base#evaluated-use). If you comply with the license and terms of use, you have the rights described therin. By using this Model, you accept the terms.
162
+
163
+ Whisper-base-WebNN is meant to be used with the corresponding sample [here](https://microsoft.github.io/webnn-developer-preview/) for educational or testing purposes only.
164
+
165
+ # WebNN changes
166
+ This original model is [Whisper-base](https://huggingface.co/openai/whisper-base). Whisper-base-WebNN is an ONNX version of the Whisper-base model that optimizes for WebNN by using static input shapes and eliminates operators that are not in use.
167
+
168
+ # Whisper-base Model card
169
+
170
+ Whisper is a pre-trained model for automatic speech recognition (ASR) and speech translation. Trained on 680k hours
171
+ of labelled data, Whisper models demonstrate a strong ability to generalise to many datasets and domains **without** the need
172
+ for fine-tuning.
173
+
174
+ Whisper was proposed in the paper [Robust Speech Recognition via Large-Scale Weak Supervision](https://arxiv.org/abs/2212.04356)
175
+ by Alec Radford et al from OpenAI. The original code repository can be found [here](https://github.com/openai/whisper).
176
+
177
+ **Disclaimer**: Content for this model card has partly been written by the Hugging Face team, and parts of it were
178
+ copied and pasted from the original model card.
179
+
180
+ ## Model details
181
+
182
+ Whisper is a Transformer based encoder-decoder model, also referred to as a _sequence-to-sequence_ model.
183
+ It was trained on 680k hours of labelled speech data annotated using large-scale weak supervision.
184
+
185
+ The models were trained on either English-only data or multilingual data. The English-only models were trained
186
+ on the task of speech recognition. The multilingual models were trained on both speech recognition and speech
187
+ translation. For speech recognition, the model predicts transcriptions in the *same* language as the audio.
188
+ For speech translation, the model predicts transcriptions to a *different* language to the audio.
189
+
190
+ Whisper checkpoints come in five configurations of varying model sizes.
191
+ The smallest four are trained on either English-only or multilingual data.
192
+ The largest checkpoints are multilingual only. All ten of the pre-trained checkpoints
193
+ are available on the [Hugging Face Hub](https://huggingface.co/models?search=openai/whisper). The
194
+ checkpoints are summarised in the following table with links to the models on the Hub:
195
+
196
+ | Size | Parameters | English-only | Multilingual |
197
+ |----------|------------|------------------------------------------------------|-----------------------------------------------------|
198
+ | tiny | 39 M | [✓](https://huggingface.co/openai/whisper-tiny.en) | [���](https://huggingface.co/openai/whisper-tiny) |
199
+ | base | 74 M | [✓](https://huggingface.co/openai/whisper-base.en) | [✓](https://huggingface.co/openai/whisper-base) |
200
+ | small | 244 M | [✓](https://huggingface.co/openai/whisper-small.en) | [✓](https://huggingface.co/openai/whisper-small) |
201
+ | medium | 769 M | [✓](https://huggingface.co/openai/whisper-medium.en) | [✓](https://huggingface.co/openai/whisper-medium) |
202
+ | large | 1550 M | x | [✓](https://huggingface.co/openai/whisper-large) |
203
+ | large-v2 | 1550 M | x | [✓](https://huggingface.co/openai/whisper-large-v2) |
204
+
205
+ # Usage
206
+
207
+ To transcribe audio samples, the model has to be used alongside a [`WhisperProcessor`](https://huggingface.co/docs/transformers/model_doc/whisper#transformers.WhisperProcessor).
208
+
209
+ The `WhisperProcessor` is used to:
210
+ 1. Pre-process the audio inputs (converting them to log-Mel spectrograms for the model)
211
+ 2. Post-process the model outputs (converting them from tokens to text)
212
+
213
+ The model is informed of which task to perform (transcription or translation) by passing the appropriate "context tokens". These context tokens
214
+ are a sequence of tokens that are given to the decoder at the start of the decoding process, and take the following order:
215
+ 1. The transcription always starts with the `<|startoftranscript|>` token
216
+ 2. The second token is the language token (e.g. `<|en|>` for English)
217
+ 3. The third token is the "task token". It can take one of two values: `<|transcribe|>` for speech recognition or `<|translate|>` for speech translation
218
+ 4. In addition, a `<|notimestamps|>` token is added if the model should not include timestamp prediction
219
+
220
+ Thus, a typical sequence of context tokens might look as follows:
221
+ ```
222
+ <|startoftranscript|> <|en|> <|transcribe|> <|notimestamps|>
223
+ ```
224
+ Which tells the model to decode in English, under the task of speech recognition, and not to predict timestamps.
225
+
226
+ These tokens can either be forced or un-forced. If they are forced, the model is made to predict each token at
227
+ each position. This allows one to control the output language and task for the Whisper model. If they are un-forced,
228
+ the Whisper model will automatically predict the output langauge and task itself.
229
+
230
+ The context tokens can be set accordingly:
231
+
232
+ ```python
233
+ model.config.forced_decoder_ids = WhisperProcessor.get_decoder_prompt_ids(language="english", task="transcribe")
234
+ ```
235
+
236
+ Which forces the model to predict in English under the task of speech recognition.
237
+
238
+ ## Transcription
239
+
240
+ ### English to English
241
+ In this example, the context tokens are 'unforced', meaning the model automatically predicts the output language
242
+ (English) and task (transcribe).
243
+
244
+ ```python
245
+ >>> from transformers import WhisperProcessor, WhisperForConditionalGeneration
246
+ >>> from datasets import load_dataset
247
+
248
+ >>> # load model and processor
249
+ >>> processor = WhisperProcessor.from_pretrained("openai/whisper-base")
250
+ >>> model = WhisperForConditionalGeneration.from_pretrained("openai/whisper-base")
251
+ >>> model.config.forced_decoder_ids = None
252
+
253
+ >>> # load dummy dataset and read audio files
254
+ >>> ds = load_dataset("hf-internal-testing/librispeech_asr_dummy", "clean", split="validation")
255
+ >>> sample = ds[0]["audio"]
256
+ >>> input_features = processor(sample["array"], sampling_rate=sample["sampling_rate"], return_tensors="pt").input_features
257
+
258
+ >>> # generate token ids
259
+ >>> predicted_ids = model.generate(input_features)
260
+ >>> # decode token ids to text
261
+ >>> transcription = processor.batch_decode(predicted_ids, skip_special_tokens=False)
262
+ ['<|startoftranscript|><|en|><|transcribe|><|notimestamps|> Mr. Quilter is the apostle of the middle classes and we are glad to welcome his gospel.<|endoftext|>']
263
+
264
+ >>> transcription = processor.batch_decode(predicted_ids, skip_special_tokens=True)
265
+ [' Mr. Quilter is the apostle of the middle classes and we are glad to welcome his gospel.']
266
+ ```
267
+ The context tokens can be removed from the start of the transcription by setting `skip_special_tokens=True`.
268
+
269
+ ### French to French
270
+ The following example demonstrates French to French transcription by setting the decoder ids appropriately.
271
+
272
+ ```python
273
+ >>> from transformers import WhisperProcessor, WhisperForConditionalGeneration
274
+ >>> from datasets import Audio, load_dataset
275
+
276
+ >>> # load model and processor
277
+ >>> processor = WhisperProcessor.from_pretrained("openai/whisper-base")
278
+ >>> model = WhisperForConditionalGeneration.from_pretrained("openai/whisper-base")
279
+ >>> forced_decoder_ids = processor.get_decoder_prompt_ids(language="french", task="transcribe")
280
+
281
+ >>> # load streaming dataset and read first audio sample
282
+ >>> ds = load_dataset("common_voice", "fr", split="test", streaming=True)
283
+ >>> ds = ds.cast_column("audio", Audio(sampling_rate=16_000))
284
+ >>> input_speech = next(iter(ds))["audio"]
285
+ >>> input_features = processor(input_speech["array"], sampling_rate=input_speech["sampling_rate"], return_tensors="pt").input_features
286
+
287
+ >>> # generate token ids
288
+ >>> predicted_ids = model.generate(input_features, forced_decoder_ids=forced_decoder_ids)
289
+ >>> # decode token ids to text
290
+ >>> transcription = processor.batch_decode(predicted_ids)
291
+ ['<|startoftranscript|><|fr|><|transcribe|><|notimestamps|> Un vrai travail intéressant va enfin être mené sur ce sujet.<|endoftext|>']
292
+
293
+ >>> transcription = processor.batch_decode(predicted_ids, skip_special_tokens=True)
294
+ [' Un vrai travail intéressant va enfin être mené sur ce sujet.']
295
+ ```
296
+
297
+ ## Translation
298
+ Setting the task to "translate" forces the Whisper model to perform speech translation.
299
+
300
+ ### French to English
301
+
302
+ ```python
303
+ >>> from transformers import WhisperProcessor, WhisperForConditionalGeneration
304
+ >>> from datasets import Audio, load_dataset
305
+
306
+ >>> # load model and processor
307
+ >>> processor = WhisperProcessor.from_pretrained("openai/whisper-base")
308
+ >>> model = WhisperForConditionalGeneration.from_pretrained("openai/whisper-base")
309
+ >>> forced_decoder_ids = processor.get_decoder_prompt_ids(language="french", task="translate")
310
+
311
+ >>> # load streaming dataset and read first audio sample
312
+ >>> ds = load_dataset("common_voice", "fr", split="test", streaming=True)
313
+ >>> ds = ds.cast_column("audio", Audio(sampling_rate=16_000))
314
+ >>> input_speech = next(iter(ds))["audio"]
315
+ >>> input_features = processor(input_speech["array"], sampling_rate=input_speech["sampling_rate"], return_tensors="pt").input_features
316
+
317
+ >>> # generate token ids
318
+ >>> predicted_ids = model.generate(input_features, forced_decoder_ids=forced_decoder_ids)
319
+ >>> # decode token ids to text
320
+ >>> transcription = processor.batch_decode(predicted_ids, skip_special_tokens=True)
321
+ [' A very interesting work, we will finally be given on this subject.']
322
+ ```
323
+
324
+ ## Evaluation
325
+
326
+ This code snippet shows how to evaluate Whisper Base on [LibriSpeech test-clean](https://huggingface.co/datasets/librispeech_asr):
327
+
328
+ ```python
329
+ >>> from datasets import load_dataset
330
+ >>> from transformers import WhisperForConditionalGeneration, WhisperProcessor
331
+ >>> import torch
332
+ >>> from evaluate import load
333
+
334
+ >>> librispeech_test_clean = load_dataset("librispeech_asr", "clean", split="test")
335
+
336
+ >>> processor = WhisperProcessor.from_pretrained("openai/whisper-base")
337
+ >>> model = WhisperForConditionalGeneration.from_pretrained("openai/whisper-base").to("cuda")
338
+
339
+ >>> def map_to_pred(batch):
340
+ >>> audio = batch["audio"]
341
+ >>> input_features = processor(audio["array"], sampling_rate=audio["sampling_rate"], return_tensors="pt").input_features
342
+ >>> batch["reference"] = processor.tokenizer._normalize(batch['text'])
343
+ >>>
344
+ >>> with torch.no_grad():
345
+ >>> predicted_ids = model.generate(input_features.to("cuda"))[0]
346
+ >>> transcription = processor.decode(predicted_ids)
347
+ >>> batch["prediction"] = processor.tokenizer._normalize(transcription)
348
+ >>> return batch
349
+
350
+ >>> result = librispeech_test_clean.map(map_to_pred)
351
+
352
+ >>> wer = load("wer")
353
+ >>> print(100 * wer.compute(references=result["reference"], predictions=result["prediction"]))
354
+ 5.082316555716899
355
+ ```
356
+
357
+ ## Long-Form Transcription
358
+
359
+ The Whisper model is intrinsically designed to work on audio samples of up to 30s in duration. However, by using a chunking
360
+ algorithm, it can be used to transcribe audio samples of up to arbitrary length. This is possible through Transformers
361
+ [`pipeline`](https://huggingface.co/docs/transformers/main_classes/pipelines#transformers.AutomaticSpeechRecognitionPipeline)
362
+ method. Chunking is enabled by setting `chunk_length_s=30` when instantiating the pipeline. With chunking enabled, the pipeline
363
+ can be run with batched inference. It can also be extended to predict sequence level timestamps by passing `return_timestamps=True`:
364
+
365
+ ```python
366
+ >>> import torch
367
+ >>> from transformers import pipeline
368
+ >>> from datasets import load_dataset
369
+
370
+ >>> device = "cuda:0" if torch.cuda.is_available() else "cpu"
371
+
372
+ >>> pipe = pipeline(
373
+ >>> "automatic-speech-recognition",
374
+ >>> model="openai/whisper-base",
375
+ >>> chunk_length_s=30,
376
+ >>> device=device,
377
+ >>> )
378
+
379
+ >>> ds = load_dataset("hf-internal-testing/librispeech_asr_dummy", "clean", split="validation")
380
+ >>> sample = ds[0]["audio"]
381
+
382
+ >>> prediction = pipe(sample.copy(), batch_size=8)["text"]
383
+ " Mr. Quilter is the apostle of the middle classes, and we are glad to welcome his gospel."
384
+
385
+ >>> # we can also return timestamps for the predictions
386
+ >>> prediction = pipe(sample.copy(), batch_size=8, return_timestamps=True)["chunks"]
387
+ [{'text': ' Mr. Quilter is the apostle of the middle classes and we are glad to welcome his gospel.',
388
+ 'timestamp': (0.0, 5.44)}]
389
+ ```
390
+
391
+ Refer to the blog post [ASR Chunking](https://huggingface.co/blog/asr-chunking) for more details on the chunking algorithm.
392
+
393
+ ## Fine-Tuning
394
+
395
+ The pre-trained Whisper model demonstrates a strong ability to generalise to different datasets and domains. However,
396
+ its predictive capabilities can be improved further for certain languages and tasks through *fine-tuning*. The blog
397
+ post [Fine-Tune Whisper with 🤗 Transformers](https://huggingface.co/blog/fine-tune-whisper) provides a step-by-step
398
+ guide to fine-tuning the Whisper model with as little as 5 hours of labelled data.
399
+
400
+ ### Evaluated Use
401
+
402
+ The primary intended users of these models are AI researchers studying robustness, generalization, capabilities, biases, and constraints of the current model. However, Whisper is also potentially quite useful as an ASR solution for developers, especially for English speech recognition. We recognize that once models are released, it is impossible to restrict access to only “intended” uses or to draw reasonable guidelines around what is or is not research.
403
+
404
+ The models are primarily trained and evaluated on ASR and speech translation to English tasks. They show strong ASR results in ~10 languages. They may exhibit additional capabilities, particularly if fine-tuned on certain tasks like voice activity detection, speaker classification, or speaker diarization but have not been robustly evaluated in these areas. We strongly recommend that users perform robust evaluations of the models in a particular context and domain before deploying them.
405
+
406
+ In particular, we caution against using Whisper models to transcribe recordings of individuals taken without their consent or purporting to use these models for any kind of subjective classification. We recommend against use in high-risk domains like decision-making contexts, where flaws in accuracy can lead to pronounced flaws in outcomes. The models are intended to transcribe and translate speech, use of the model for classification is not only not evaluated but also not appropriate, particularly to infer human attributes.
407
+
408
+
409
+ ## Training Data
410
+
411
+ The models are trained on 680,000 hours of audio and the corresponding transcripts collected from the internet. 65% of this data (or 438,000 hours) represents English-language audio and matched English transcripts, roughly 18% (or 126,000 hours) represents non-English audio and English transcripts, while the final 17% (or 117,000 hours) represents non-English audio and the corresponding transcript. This non-English data represents 98 different languages.
412
+
413
+ As discussed in [the accompanying paper](https://cdn.openai.com/papers/whisper.pdf), we see that performance on transcription in a given language is directly correlated with the amount of training data we employ in that language.
414
+
415
+
416
+ ## Performance and Limitations
417
+
418
+ Our studies show that, over many existing ASR systems, the models exhibit improved robustness to accents, background noise, technical language, as well as zero shot translation from multiple languages into English; and that accuracy on speech recognition and translation is near the state-of-the-art level.
419
+
420
+ However, because the models are trained in a weakly supervised manner using large-scale noisy data, the predictions may include texts that are not actually spoken in the audio input (i.e. hallucination). We hypothesize that this happens because, given their general knowledge of language, the models combine trying to predict the next word in audio with trying to transcribe the audio itself.
421
+
422
+ Our models perform unevenly across languages, and we observe lower accuracy on low-resource and/or low-discoverability languages or languages where we have less training data. The models also exhibit disparate performance on different accents and dialects of particular languages, which may include higher word error rate across speakers of different genders, races, ages, or other demographic criteria. Our full evaluation results are presented in [the paper accompanying this release](https://cdn.openai.com/papers/whisper.pdf).
423
+
424
+ In addition, the sequence-to-sequence architecture of the model makes it prone to generating repetitive texts, which can be mitigated to some degree by beam search and temperature scheduling but not perfectly. Further analysis on these limitations are provided in [the paper](https://cdn.openai.com/papers/whisper.pdf). It is likely that this behavior and hallucinations may be worse on lower-resource and/or lower-discoverability languages.
425
+
426
+
427
+ ## Broader Implications
428
+
429
+ We anticipate that Whisper models’ transcription capabilities may be used for improving accessibility tools. While Whisper models cannot be used for real-time transcription out of the box – their speed and size suggest that others may be able to build applications on top of them that allow for near-real-time speech recognition and translation. The real value of beneficial applications built on top of Whisper models suggests that the disparate performance of these models may have real economic implications.
430
+
431
+ There are also potential dual use concerns that come with releasing Whisper. While we hope the technology will be used primarily for beneficial purposes, making ASR technology more accessible could enable more actors to build capable surveillance technologies or scale up existing surveillance efforts, as the speed and accuracy allow for affordable automatic transcription and translation of large volumes of audio communication. Moreover, these models may have some capabilities to recognize specific individuals out of the box, which in turn presents safety concerns related both to dual use and disparate performance. In practice, we expect that the cost of transcription is not the limiting factor of scaling up surveillance projects.
432
+
433
+
434
+ ### BibTeX entry and citation info
435
+ ```bibtex
436
+ @misc{radford2022whisper,
437
+ doi = {10.48550/ARXIV.2212.04356},
438
+ url = {https://arxiv.org/abs/2212.04356},
439
+ author = {Radford, Alec and Kim, Jong Wook and Xu, Tao and Brockman, Greg and McLeavey, Christine and Sutskever, Ilya},
440
+ title = {Robust Speech Recognition via Large-Scale Weak Supervision},
441
+ publisher = {arXiv},
442
+ year = {2022},
443
+ copyright = {arXiv.org perpetual, non-exclusive license}
444
+ }
445
+ ```
processer/resolve/main/.gitkeep ADDED
File without changes
processer/resolve/main/preprocessor_config.json ADDED
The diff for this file is too large to render. See raw diff
 
tokenizer/resolve/main/.gitkeep ADDED
File without changes
tokenizer/resolve/main/tokenizer.json ADDED
The diff for this file is too large to render. See raw diff
 
tokenizer/resolve/main/tokenizer_config.json ADDED
The diff for this file is too large to render. See raw diff
 
whisper_base_decoder_static_kvcache_128_lm_fp16_layernorm.onnx ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f0d073987c98193e456cf0964d5e150f97032bb125b1296ca06856d9dbba5841
3
+ size 150945858
whisper_base_decoder_static_kvcache_128_lm_fp16_layernorm_4dmask.onnx ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:402fe671970ab2862b7e3e5e8525a729dd2cbaf0b2ee3dc1beb24d52f6e16b90
3
+ size 150899784
whisper_base_decoder_static_kvcache_128_lm_fp16_layernorm_gelu_4dmask.onnx ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d2096c8b32e193ef32640a40aa41f7489652df2bb6efb7cf34bf5fa0e899efe1
3
+ size 150892897
whisper_base_decoder_static_non_kvcache_lm_fp16_layernorm.onnx ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a3717fa06607ad4fd04d4f65a09b21a444f8e226bf5cf6de2d2aa43dc0ae33ed
3
+ size 156797289
whisper_base_decoder_static_non_kvcache_lm_fp16_layernorm_4dmask.onnx ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:13de18ba5175327de7d2e72438693c815ff4474e2d0c1c4703322749532fe9e8
3
+ size 156748222
whisper_base_decoder_static_non_kvcache_lm_fp16_layernorm_gelu_4dmask.onnx ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:060be4a0cf2b4bfba0048f2368085a22c9cd1699b31364cb39e0c65e95d3de2e
3
+ size 156741064
whisper_base_encoder_lm_fp16_layernorm.onnx ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:905fc54f7ffb82a364e30a857ab3f6c7cc8ad0bb3c3b31d4940029b6929cdb04
3
+ size 41248401
whisper_base_encoder_lm_fp16_layernorm_gelu.onnx ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6e1cc2bfac4a2731dfd1852bbc98f338c8f0d02ff23af69f687f886f0062cbf8
3
+ size 41244536