Commit
β’
afda370
1
Parent(s):
3d06185
Update README.md (#144)
Browse files- Update README.md (29aee6b8a7c9c86b46f63bf6fc2331151935026b)
README.md
CHANGED
@@ -114,68 +114,39 @@ license: apache-2.0
|
|
114 |
|
115 |
# Whisper
|
116 |
|
117 |
-
Whisper is a
|
118 |
-
|
119 |
-
|
|
|
120 |
|
121 |
-
Whisper
|
122 |
-
|
123 |
|
124 |
-
|
125 |
-
|
126 |
-
1. The input uses 128 Mel frequency bins instead of 80
|
127 |
2. A new language token for Cantonese
|
128 |
|
129 |
-
The Whisper
|
130 |
-
The model was trained for 2.0 epochs over this mixture dataset.
|
131 |
-
|
132 |
-
The `large-v3` model shows improved performance over a wide variety of languages, showing 10% to 20% reduction of errors compared to Whisper `large-v2`.
|
133 |
-
|
134 |
-
|
135 |
-
**Disclaimer**: Content for this model card has partly been written by the Hugging Face team, and parts of it were
|
136 |
-
copied and pasted from the original model card.
|
137 |
-
|
138 |
-
## Model details
|
139 |
-
|
140 |
-
Whisper is a Transformer based encoder-decoder model, also referred to as a _sequence-to-sequence_ model.
|
141 |
-
It was trained on 1 million hours of weakly labeled audio and 4 million hours of pseudolabeled audio collected using Whisper `large-v2`.
|
142 |
-
|
143 |
-
The models were trained on either English-only data or multilingual data. The English-only models were trained
|
144 |
-
on the task of speech recognition. The multilingual models were trained on both speech recognition and speech
|
145 |
-
translation. For speech recognition, the model predicts transcriptions in the *same* language as the audio.
|
146 |
-
For speech translation, the model predicts transcriptions to a *different* language to the audio.
|
147 |
|
148 |
-
|
149 |
-
|
150 |
-
The largest checkpoints are multilingual only. All ten of the pre-trained checkpoints
|
151 |
-
are available on the [Hugging Face Hub](https://huggingface.co/models?search=openai/whisper). The
|
152 |
-
checkpoints are summarised in the following table with links to the models on the Hub:
|
153 |
|
154 |
-
|
155 |
-
|
156 |
-
| tiny | 39 M | [β](https://huggingface.co/openai/whisper-tiny.en) | [β](https://huggingface.co/openai/whisper-tiny) |
|
157 |
-
| base | 74 M | [β](https://huggingface.co/openai/whisper-base.en) | [β](https://huggingface.co/openai/whisper-base) |
|
158 |
-
| small | 244 M | [β](https://huggingface.co/openai/whisper-small.en) | [β](https://huggingface.co/openai/whisper-small) |
|
159 |
-
| medium | 769 M | [β](https://huggingface.co/openai/whisper-medium.en) | [β](https://huggingface.co/openai/whisper-medium) |
|
160 |
-
| large | 1550 M | x | [β](https://huggingface.co/openai/whisper-large) |
|
161 |
-
| large-v2 | 1550 M | x | [β](https://huggingface.co/openai/whisper-large-v2) |
|
162 |
-
| large-v3 | 1550 M | x | [β](https://huggingface.co/openai/whisper-large-v3) |
|
163 |
|
164 |
## Usage
|
165 |
|
166 |
-
Whisper
|
167 |
-
|
168 |
-
|
169 |
|
170 |
```bash
|
171 |
pip install --upgrade pip
|
172 |
-
pip install --upgrade
|
173 |
```
|
174 |
|
175 |
-
### Short-Form Transcription
|
176 |
-
|
177 |
The model can be used with the [`pipeline`](https://huggingface.co/docs/transformers/main_classes/pipelines#transformers.AutomaticSpeechRecognitionPipeline)
|
178 |
-
class to transcribe
|
179 |
|
180 |
```python
|
181 |
import torch
|
@@ -200,10 +171,6 @@ pipe = pipeline(
|
|
200 |
model=model,
|
201 |
tokenizer=processor.tokenizer,
|
202 |
feature_extractor=processor.feature_extractor,
|
203 |
-
max_new_tokens=128,
|
204 |
-
chunk_length_s=30,
|
205 |
-
batch_size=16,
|
206 |
-
return_timestamps=True,
|
207 |
torch_dtype=torch_dtype,
|
208 |
device=device,
|
209 |
)
|
@@ -216,9 +183,33 @@ print(result["text"])
|
|
216 |
```
|
217 |
|
218 |
To transcribe a local audio file, simply pass the path to your audio file when you call the pipeline:
|
219 |
-
|
220 |
-
|
221 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
222 |
```
|
223 |
|
224 |
Whisper predicts the language of the source audio automatically. If the source audio language is known *a-priori*, it
|
@@ -261,10 +252,6 @@ print(result["chunks"])
|
|
261 |
|
262 |
<summary> For more control over the generation parameters, use the model + processor API directly: </summary>
|
263 |
|
264 |
-
Ad-hoc generation arguments can be passed to `model.generate`, including `num_beams` for beam-search, `return_timestamps`
|
265 |
-
for segment-level timestamps, and `prompt_ids` for prompting. See the [docstrings](https://huggingface.co/docs/transformers/en/model_doc/whisper#transformers.WhisperForConditionalGeneration.generate)
|
266 |
-
for more details.
|
267 |
-
|
268 |
```python
|
269 |
import torch
|
270 |
from transformers import AutoModelForSpeechSeq2Seq, AutoProcessor
|
@@ -277,7 +264,7 @@ torch_dtype = torch.float16 if torch.cuda.is_available() else torch.float32
|
|
277 |
model_id = "openai/whisper-large-v3"
|
278 |
|
279 |
model = AutoModelForSpeechSeq2Seq.from_pretrained(
|
280 |
-
model_id, torch_dtype=torch_dtype, low_cpu_mem_usage=True
|
281 |
)
|
282 |
model.to(device)
|
283 |
|
@@ -287,37 +274,58 @@ dataset = load_dataset("hf-internal-testing/librispeech_asr_dummy", "clean", spl
|
|
287 |
dataset = dataset.cast_column("audio", Audio(processor.feature_extractor.sampling_rate))
|
288 |
sample = dataset[0]["audio"]
|
289 |
|
290 |
-
|
291 |
-
|
292 |
-
|
293 |
-
|
294 |
-
|
|
|
|
|
|
|
|
|
295 |
|
296 |
gen_kwargs = {
|
297 |
-
|
298 |
-
|
299 |
-
|
|
|
|
|
|
|
|
|
|
|
300 |
}
|
301 |
|
302 |
-
pred_ids = model.generate(
|
303 |
-
pred_text = processor.batch_decode(pred_ids, skip_special_tokens=True, decode_with_timestamps=
|
304 |
|
305 |
print(pred_text)
|
306 |
```
|
307 |
|
308 |
</details>
|
309 |
|
310 |
-
|
311 |
|
312 |
-
|
313 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
314 |
|
315 |
The sequential long-form algorithm should be used in either of the following scenarios:
|
316 |
-
1. Transcription accuracy is the most important factor, and
|
317 |
2. You are transcribing **batches** of long audio files, in which case the latency of sequential is comparable to chunked, while being up to 0.5% WER more accurate
|
318 |
|
319 |
-
|
320 |
-
|
|
|
|
|
|
|
|
|
|
|
321 |
|
322 |
```python
|
323 |
import torch
|
@@ -331,7 +339,7 @@ torch_dtype = torch.float16 if torch.cuda.is_available() else torch.float32
|
|
331 |
model_id = "openai/whisper-large-v3"
|
332 |
|
333 |
model = AutoModelForSpeechSeq2Seq.from_pretrained(
|
334 |
-
model_id, torch_dtype=torch_dtype, low_cpu_mem_usage=True
|
335 |
)
|
336 |
model.to(device)
|
337 |
|
@@ -342,7 +350,8 @@ pipe = pipeline(
|
|
342 |
model=model,
|
343 |
tokenizer=processor.tokenizer,
|
344 |
feature_extractor=processor.feature_extractor,
|
345 |
-
|
|
|
346 |
torch_dtype=torch_dtype,
|
347 |
device=device,
|
348 |
)
|
@@ -354,76 +363,21 @@ result = pipe(sample)
|
|
354 |
print(result["text"])
|
355 |
```
|
356 |
|
357 |
-
|
358 |
-
|
359 |
-
<summary> For more control over the generation parameters, use the model + processor API directly: </summary>
|
360 |
-
|
361 |
-
```python
|
362 |
-
import torch
|
363 |
-
from transformers import AutoModelForSpeechSeq2Seq, AutoProcessor
|
364 |
-
from datasets import Audio, load_dataset
|
365 |
-
|
366 |
-
|
367 |
-
device = "cuda:0" if torch.cuda.is_available() else "cpu"
|
368 |
-
torch_dtype = torch.float16 if torch.cuda.is_available() else torch.float32
|
369 |
-
|
370 |
-
model_id = "openai/whisper-large-v3"
|
371 |
-
|
372 |
-
model = AutoModelForSpeechSeq2Seq.from_pretrained(
|
373 |
-
model_id, torch_dtype=torch_dtype, low_cpu_mem_usage=True, use_safetensors=True
|
374 |
-
)
|
375 |
-
model.to(device)
|
376 |
-
|
377 |
-
processor = AutoProcessor.from_pretrained(model_id)
|
378 |
-
|
379 |
-
dataset = load_dataset("hf-internal-testing/librispeech_asr_dummy", "clean", split="validation")
|
380 |
-
dataset = dataset.cast_column("audio", Audio(processor.feature_extractor.sampling_rate))
|
381 |
-
sample = dataset[0]["audio"]
|
382 |
-
|
383 |
-
inputs = processor(
|
384 |
-
sample["array"],
|
385 |
-
sampling_rate=sample["sampling_rate"],
|
386 |
-
return_tensors="pt",
|
387 |
-
truncation=False,
|
388 |
-
padding="longest",
|
389 |
-
return_attention_mask=True,
|
390 |
-
)
|
391 |
-
inputs = inputs.to(device, dtype=torch_dtype)
|
392 |
-
|
393 |
-
gen_kwargs = {
|
394 |
-
"max_new_tokens": 448,
|
395 |
-
"num_beams": 1,
|
396 |
-
"condition_on_prev_tokens": False,
|
397 |
-
"compression_ratio_threshold": 1.35, # zlib compression ratio threshold (in token space)
|
398 |
-
"temperature": (0.0, 0.2, 0.4, 0.6, 0.8, 1.0),
|
399 |
-
"logprob_threshold": -1.0,
|
400 |
-
"no_speech_threshold": 0.6,
|
401 |
-
"return_timestamps": True,
|
402 |
-
}
|
403 |
-
|
404 |
-
pred_ids = model.generate(**i nputs, **gen_kwargs)
|
405 |
-
pred_text = processor.batch_decode(pred_ids, skip_special_tokens=True, decode_with_timestamps=False)
|
406 |
-
|
407 |
-
print(pred_text)
|
408 |
-
```
|
409 |
-
|
410 |
-
</details>
|
411 |
-
|
412 |
-
### Chunked Long-Form
|
413 |
|
414 |
-
|
415 |
-
|
416 |
-
the chunked algorithm is up to 9x faster than OpenAI's sequential long-form implementation (see Table 7 of the
|
417 |
-
[Distil-Whisper paper](https://arxiv.org/pdf/2311.00430.pdf)).
|
418 |
|
419 |
-
|
420 |
-
is optimal. To activate batching over long audio files, pass the argument `batch_size`:
|
421 |
|
422 |
```python
|
423 |
import torch
|
|
|
424 |
from transformers import AutoModelForSpeechSeq2Seq, AutoProcessor, pipeline
|
425 |
from datasets import load_dataset
|
|
|
426 |
|
|
|
427 |
|
428 |
device = "cuda:0" if torch.cuda.is_available() else "cpu"
|
429 |
torch_dtype = torch.float16 if torch.cuda.is_available() else torch.float32
|
@@ -431,9 +385,12 @@ torch_dtype = torch.float16 if torch.cuda.is_available() else torch.float32
|
|
431 |
model_id = "openai/whisper-large-v3"
|
432 |
|
433 |
model = AutoModelForSpeechSeq2Seq.from_pretrained(
|
434 |
-
model_id, torch_dtype=torch_dtype, low_cpu_mem_usage=True
|
435 |
-
)
|
436 |
-
|
|
|
|
|
|
|
437 |
|
438 |
processor = AutoProcessor.from_pretrained(model_id)
|
439 |
|
@@ -442,9 +399,6 @@ pipe = pipeline(
|
|
442 |
model=model,
|
443 |
tokenizer=processor.tokenizer,
|
444 |
feature_extractor=processor.feature_extractor,
|
445 |
-
max_new_tokens=128,
|
446 |
-
chunk_length_s=25,
|
447 |
-
batch_size=16,
|
448 |
torch_dtype=torch_dtype,
|
449 |
device=device,
|
450 |
)
|
@@ -452,20 +406,22 @@ pipe = pipeline(
|
|
452 |
dataset = load_dataset("distil-whisper/librispeech_long", "clean", split="validation")
|
453 |
sample = dataset[0]["audio"]
|
454 |
|
455 |
-
|
456 |
-
|
457 |
-
|
|
|
458 |
|
459 |
-
|
|
|
|
|
460 |
|
461 |
-
|
462 |
-
|
463 |
-
more efficient flash attention version.
|
464 |
|
465 |
#### Flash Attention 2
|
466 |
|
467 |
-
We recommend using [Flash-Attention 2](https://huggingface.co/docs/transformers/main/en/perf_infer_gpu_one#flashattention-2)
|
468 |
-
|
469 |
|
470 |
```
|
471 |
pip install flash-attn --no-build-isolation
|
@@ -473,9 +429,8 @@ pip install flash-attn --no-build-isolation
|
|
473 |
|
474 |
Then pass `attn_implementation="flash_attention_2"` to `from_pretrained`:
|
475 |
|
476 |
-
```
|
477 |
-
|
478 |
-
+ model = AutoModelForSpeechSeq2Seq.from_pretrained(model_id, torch_dtype=torch_dtype, low_cpu_mem_usage=True, use_safetensors=True, attn_implementation="flash_attention_2")
|
479 |
```
|
480 |
|
481 |
#### Torch Scale-Product-Attention (SDPA)
|
@@ -496,20 +451,36 @@ returns `False`, you need to upgrade your PyTorch version according to the [offi
|
|
496 |
Once a valid PyTorch version is installed, SDPA is activated by default. It can also be set explicitly by specifying
|
497 |
`attn_implementation="sdpa"` as follows:
|
498 |
|
499 |
-
```
|
500 |
-
|
501 |
-
+ model = AutoModelForSpeechSeq2Seq.from_pretrained(model_id, torch_dtype=torch_dtype, low_cpu_mem_usage=True, use_safetensors=True, attn_implementation="sdpa")
|
502 |
```
|
503 |
|
504 |
For more information about how to use the SDPA refer to the [Transformers SDPA documentation](https://huggingface.co/docs/transformers/en/perf_infer_gpu_one#pytorch-scaled-dot-product-attention).
|
505 |
|
506 |
-
#### Torch compile
|
507 |
|
508 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
509 |
|
510 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
511 |
|
512 |
-
Coming soon...
|
513 |
|
514 |
## Fine-Tuning
|
515 |
|
@@ -529,7 +500,7 @@ In particular, we caution against using Whisper models to transcribe recordings
|
|
529 |
|
530 |
## Training Data
|
531 |
|
532 |
-
The
|
533 |
|
534 |
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.
|
535 |
|
|
|
114 |
|
115 |
# Whisper
|
116 |
|
117 |
+
Whisper is a state-of-the-art model for automatic speech recognition (ASR) and speech translation, proposed in the paper
|
118 |
+
[Robust Speech Recognition via Large-Scale Weak Supervision](https://huggingface.co/papers/2212.04356) by Alec Radford
|
119 |
+
et al. from OpenAI. Trained on >5M hours of labeled data, Whisper demonstrates a strong ability to generalise to many
|
120 |
+
datasets and domains in a zero-shot setting.
|
121 |
|
122 |
+
Whisper large-v3 has the same architecture as the previous [large](https://huggingface.co/openai/whisper-large) and [large-v2](https://huggingface.co/openai/whisper-large-v2)
|
123 |
+
models, except for the following minor differences:
|
124 |
|
125 |
+
1. The spectrogram input uses 128 Mel frequency bins instead of 80
|
|
|
|
|
126 |
2. A new language token for Cantonese
|
127 |
|
128 |
+
The Whisper large-v3 model was trained on 1 million hours of weakly labeled audio and 4 million hours of pseudo-labeled
|
129 |
+
audio collected using Whisper [large-v2](https://huggingface.co/openai/whisper-large-v2) . The model was trained for 2.0 epochs over this mixture dataset.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
130 |
|
131 |
+
The large-v3 model shows improved performance over a wide variety of languages, showing 10% to 20% reduction of errors
|
132 |
+
compared to Whisper [large-v2](https://huggingface.co/openai/whisper-large-v2) . For more details on the different checkpoints available, refer to the section [Model details](#model-details).
|
|
|
|
|
|
|
133 |
|
134 |
+
**Disclaimer**: Content for this model card has partly been written by the π€ Hugging Face team, and partly copied and
|
135 |
+
pasted from the original model card.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
136 |
|
137 |
## Usage
|
138 |
|
139 |
+
Whisper large-v3 is supported in Hugging Face π€ Transformers. To run the model, first install the Transformers
|
140 |
+
library. For this example, we'll also install π€ Datasets to load toy audio dataset from the Hugging Face Hub, and
|
141 |
+
π€ Accelerate to reduce the model loading time:
|
142 |
|
143 |
```bash
|
144 |
pip install --upgrade pip
|
145 |
+
pip install --upgrade transformers datasets[audio] accelerate
|
146 |
```
|
147 |
|
|
|
|
|
148 |
The model can be used with the [`pipeline`](https://huggingface.co/docs/transformers/main_classes/pipelines#transformers.AutomaticSpeechRecognitionPipeline)
|
149 |
+
class to transcribe audios of arbitrary length:
|
150 |
|
151 |
```python
|
152 |
import torch
|
|
|
171 |
model=model,
|
172 |
tokenizer=processor.tokenizer,
|
173 |
feature_extractor=processor.feature_extractor,
|
|
|
|
|
|
|
|
|
174 |
torch_dtype=torch_dtype,
|
175 |
device=device,
|
176 |
)
|
|
|
183 |
```
|
184 |
|
185 |
To transcribe a local audio file, simply pass the path to your audio file when you call the pipeline:
|
186 |
+
|
187 |
+
```python
|
188 |
+
result = pipe("audio.mp3")
|
189 |
+
```
|
190 |
+
|
191 |
+
Multiple audio files can be transcribed in parallel by specifying them as a list and setting the `batch_size` parameter:
|
192 |
+
|
193 |
+
```python
|
194 |
+
result = pipe(["audio_1.mp3", "audio_2.mp3"], batch_size=2)
|
195 |
+
```
|
196 |
+
|
197 |
+
Transformers is compatible with all Whisper decoding strategies, such as temperature fallback and condition on previous
|
198 |
+
tokens. The following example demonstrates how to enable these heuristics:
|
199 |
+
|
200 |
+
```python
|
201 |
+
generate_kwargs = {
|
202 |
+
"max_new_tokens": 448,
|
203 |
+
"num_beams": 1,
|
204 |
+
"condition_on_prev_tokens": False,
|
205 |
+
"compression_ratio_threshold": 1.35, # zlib compression ratio threshold (in token space)
|
206 |
+
"temperature": (0.0, 0.2, 0.4, 0.6, 0.8, 1.0),
|
207 |
+
"logprob_threshold": -1.0,
|
208 |
+
"no_speech_threshold": 0.6,
|
209 |
+
"return_timestamps": True,
|
210 |
+
}
|
211 |
+
|
212 |
+
result = pipe(sample, generate_kwargs=generate_kwargs)
|
213 |
```
|
214 |
|
215 |
Whisper predicts the language of the source audio automatically. If the source audio language is known *a-priori*, it
|
|
|
252 |
|
253 |
<summary> For more control over the generation parameters, use the model + processor API directly: </summary>
|
254 |
|
|
|
|
|
|
|
|
|
255 |
```python
|
256 |
import torch
|
257 |
from transformers import AutoModelForSpeechSeq2Seq, AutoProcessor
|
|
|
264 |
model_id = "openai/whisper-large-v3"
|
265 |
|
266 |
model = AutoModelForSpeechSeq2Seq.from_pretrained(
|
267 |
+
model_id, torch_dtype=torch_dtype, low_cpu_mem_usage=True
|
268 |
)
|
269 |
model.to(device)
|
270 |
|
|
|
274 |
dataset = dataset.cast_column("audio", Audio(processor.feature_extractor.sampling_rate))
|
275 |
sample = dataset[0]["audio"]
|
276 |
|
277 |
+
inputs = processor(
|
278 |
+
sample["array"],
|
279 |
+
sampling_rate=sample["sampling_rate"],
|
280 |
+
return_tensors="pt",
|
281 |
+
truncation=False,
|
282 |
+
padding="longest",
|
283 |
+
return_attention_mask=True,
|
284 |
+
)
|
285 |
+
inputs = inputs.to(device, dtype=torch_dtype)
|
286 |
|
287 |
gen_kwargs = {
|
288 |
+
"max_new_tokens": 448,
|
289 |
+
"num_beams": 1,
|
290 |
+
"condition_on_prev_tokens": False,
|
291 |
+
"compression_ratio_threshold": 1.35, # zlib compression ratio threshold (in token space)
|
292 |
+
"temperature": (0.0, 0.2, 0.4, 0.6, 0.8, 1.0),
|
293 |
+
"logprob_threshold": -1.0,
|
294 |
+
"no_speech_threshold": 0.6,
|
295 |
+
"return_timestamps": True,
|
296 |
}
|
297 |
|
298 |
+
pred_ids = model.generate(**inputs, **gen_kwargs)
|
299 |
+
pred_text = processor.batch_decode(pred_ids, skip_special_tokens=True, decode_with_timestamps=False)
|
300 |
|
301 |
print(pred_text)
|
302 |
```
|
303 |
|
304 |
</details>
|
305 |
|
306 |
+
## Additional Speed & Memory Improvements
|
307 |
|
308 |
+
You can apply additional speed and memory improvements to Whisper to further reduce the inference speed and VRAM
|
309 |
+
requirements.
|
310 |
+
|
311 |
+
### Chunked Long-Form
|
312 |
+
|
313 |
+
Whisper has a receptive field of 30-seconds. To transcribe audios longer than this, one of two long-form algorithms are
|
314 |
+
required:
|
315 |
+
1. **Sequential:** uses a "sliding window" for buffered inference, transcribing 30-second slices one after the other
|
316 |
+
2. **Chunked:** splits long audio files into shorter ones (with a small overlap between segments), transcribes each segment independently, and stitches the resulting transcriptions at the boundaries
|
317 |
|
318 |
The sequential long-form algorithm should be used in either of the following scenarios:
|
319 |
+
1. Transcription accuracy is the most important factor, and speed is less of a consideration
|
320 |
2. You are transcribing **batches** of long audio files, in which case the latency of sequential is comparable to chunked, while being up to 0.5% WER more accurate
|
321 |
|
322 |
+
Conversely, the chunked algorithm should be used when:
|
323 |
+
1. Transcription speed is the most important factor
|
324 |
+
2. You are transcribing a **single** long audio file
|
325 |
+
|
326 |
+
By default, Transformers uses the sequential algorithm. To enable the chunked algorithm, pass the `chunk_length_s`
|
327 |
+
parameter to the `pipeline`. For large-v3, a chunk length of 30-seconds is optimal. To activate batching over long
|
328 |
+
audio files, pass the argument `batch_size`:
|
329 |
|
330 |
```python
|
331 |
import torch
|
|
|
339 |
model_id = "openai/whisper-large-v3"
|
340 |
|
341 |
model = AutoModelForSpeechSeq2Seq.from_pretrained(
|
342 |
+
model_id, torch_dtype=torch_dtype, low_cpu_mem_usage=True
|
343 |
)
|
344 |
model.to(device)
|
345 |
|
|
|
350 |
model=model,
|
351 |
tokenizer=processor.tokenizer,
|
352 |
feature_extractor=processor.feature_extractor,
|
353 |
+
chunk_length_s=30,
|
354 |
+
batch_size=16, # batch size for inference - set based on your device
|
355 |
torch_dtype=torch_dtype,
|
356 |
device=device,
|
357 |
)
|
|
|
363 |
print(result["text"])
|
364 |
```
|
365 |
|
366 |
+
#### Torch compile
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
367 |
|
368 |
+
The Whisper forward pass is compatible with [`torch.compile`](https://pytorch.org/docs/stable/generated/torch.compile.html)
|
369 |
+
for 4.5x speed-ups.
|
|
|
|
|
370 |
|
371 |
+
**Note:** `torch.compile` is currently not compatible with the Chunked long-form algorithm or Flash Attention 2 β οΈ
|
|
|
372 |
|
373 |
```python
|
374 |
import torch
|
375 |
+
from torch.nn.attention import SDPBackend, sdpa_kernel
|
376 |
from transformers import AutoModelForSpeechSeq2Seq, AutoProcessor, pipeline
|
377 |
from datasets import load_dataset
|
378 |
+
from tqdm import tqdm
|
379 |
|
380 |
+
torch.set_float32_matmul_precision("high")
|
381 |
|
382 |
device = "cuda:0" if torch.cuda.is_available() else "cpu"
|
383 |
torch_dtype = torch.float16 if torch.cuda.is_available() else torch.float32
|
|
|
385 |
model_id = "openai/whisper-large-v3"
|
386 |
|
387 |
model = AutoModelForSpeechSeq2Seq.from_pretrained(
|
388 |
+
model_id, torch_dtype=torch_dtype, low_cpu_mem_usage=True
|
389 |
+
).to(device)
|
390 |
+
|
391 |
+
# Enable static cache and compile the forward pass
|
392 |
+
model.generation_config.cache_implementation = "static"
|
393 |
+
model.forward = torch.compile(model.forward, mode="reduce-overhead", fullgraph=True)
|
394 |
|
395 |
processor = AutoProcessor.from_pretrained(model_id)
|
396 |
|
|
|
399 |
model=model,
|
400 |
tokenizer=processor.tokenizer,
|
401 |
feature_extractor=processor.feature_extractor,
|
|
|
|
|
|
|
402 |
torch_dtype=torch_dtype,
|
403 |
device=device,
|
404 |
)
|
|
|
406 |
dataset = load_dataset("distil-whisper/librispeech_long", "clean", split="validation")
|
407 |
sample = dataset[0]["audio"]
|
408 |
|
409 |
+
# 2 warmup steps
|
410 |
+
for _ in tqdm(range(2), desc="Warm-up step"):
|
411 |
+
with sdpa_kernel(SDPBackend.MATH):
|
412 |
+
result = pipe(sample.copy())
|
413 |
|
414 |
+
# fast run
|
415 |
+
with sdpa_kernel(SDPBackend.MATH):
|
416 |
+
result = pipe(sample.copy())
|
417 |
|
418 |
+
print(result["text"])
|
419 |
+
```
|
|
|
420 |
|
421 |
#### Flash Attention 2
|
422 |
|
423 |
+
We recommend using [Flash-Attention 2](https://huggingface.co/docs/transformers/main/en/perf_infer_gpu_one#flashattention-2) if your GPU supports it and you are not using [torch.compile](#torch-compile).
|
424 |
+
To do so, first install [Flash Attention](https://github.com/Dao-AILab/flash-attention):
|
425 |
|
426 |
```
|
427 |
pip install flash-attn --no-build-isolation
|
|
|
429 |
|
430 |
Then pass `attn_implementation="flash_attention_2"` to `from_pretrained`:
|
431 |
|
432 |
+
```python
|
433 |
+
model = AutoModelForSpeechSeq2Seq.from_pretrained(model_id, torch_dtype=torch_dtype, low_cpu_mem_usage=True, attn_implementation="flash_attention_2")
|
|
|
434 |
```
|
435 |
|
436 |
#### Torch Scale-Product-Attention (SDPA)
|
|
|
451 |
Once a valid PyTorch version is installed, SDPA is activated by default. It can also be set explicitly by specifying
|
452 |
`attn_implementation="sdpa"` as follows:
|
453 |
|
454 |
+
```python
|
455 |
+
model = AutoModelForSpeechSeq2Seq.from_pretrained(model_id, torch_dtype=torch_dtype, low_cpu_mem_usage=True, attn_implementation="sdpa")
|
|
|
456 |
```
|
457 |
|
458 |
For more information about how to use the SDPA refer to the [Transformers SDPA documentation](https://huggingface.co/docs/transformers/en/perf_infer_gpu_one#pytorch-scaled-dot-product-attention).
|
459 |
|
|
|
460 |
|
461 |
+
## Model details
|
462 |
+
|
463 |
+
Whisper is a Transformer based encoder-decoder model, also referred to as a _sequence-to-sequence_ model. There are two
|
464 |
+
flavours of Whisper model: English-only and multilingual. The English-only models were trained on the task of English
|
465 |
+
speech recognition. The multilingual models were trained simultaneously on multilingual speech recognition and speech
|
466 |
+
translation. For speech recognition, the model predicts transcriptions in the *same* language as the audio. For speech
|
467 |
+
translation, the model predicts transcriptions to a *different* language to the audio.
|
468 |
+
|
469 |
+
Whisper checkpoints come in five configurations of varying model sizes. The smallest four are available as English-only
|
470 |
+
and multilingual. The largest checkpoints are multilingual only. All ten of the pre-trained checkpoints
|
471 |
+
are available on the [Hugging Face Hub](https://huggingface.co/models?search=openai/whisper). The
|
472 |
+
checkpoints are summarised in the following table with links to the models on the Hub:
|
473 |
|
474 |
+
| Size | Parameters | English-only | Multilingual |
|
475 |
+
|----------|------------|------------------------------------------------------|-----------------------------------------------------|
|
476 |
+
| tiny | 39 M | [β](https://huggingface.co/openai/whisper-tiny.en) | [β](https://huggingface.co/openai/whisper-tiny) |
|
477 |
+
| base | 74 M | [β](https://huggingface.co/openai/whisper-base.en) | [β](https://huggingface.co/openai/whisper-base) |
|
478 |
+
| small | 244 M | [β](https://huggingface.co/openai/whisper-small.en) | [β](https://huggingface.co/openai/whisper-small) |
|
479 |
+
| medium | 769 M | [β](https://huggingface.co/openai/whisper-medium.en) | [β](https://huggingface.co/openai/whisper-medium) |
|
480 |
+
| large | 1550 M | x | [β](https://huggingface.co/openai/whisper-large) |
|
481 |
+
| large-v2 | 1550 M | x | [β](https://huggingface.co/openai/whisper-large-v2) |
|
482 |
+
| large-v3 | 1550 M | x | [β](https://huggingface.co/openai/whisper-large-v3) |
|
483 |
|
|
|
484 |
|
485 |
## Fine-Tuning
|
486 |
|
|
|
500 |
|
501 |
## Training Data
|
502 |
|
503 |
+
The large-v3 checkpoint is trained on 1 million hours of weakly labeled audio and 4 million hours of pseudo-labeled audio collected using Whisper large-v2.
|
504 |
|
505 |
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.
|
506 |
|