marinone94
commited on
Commit
•
073a108
1
Parent(s):
3583a24
End of training
Browse files- all_results.json +18 -3
- eval_pretrained_results.json +7 -0
- huggingface_training.py +37 -37
- test_finetuned_results.json +8 -0
- test_pretrained_results.json +7 -0
- train_results.json +3 -3
- trainer_state.json +9 -9
- training_args.bin +1 -1
all_results.json
CHANGED
@@ -1,17 +1,32 @@
|
|
1 |
{
|
2 |
"epoch": 1.0,
|
3 |
"eval_loss": 1.6191972494125366,
|
|
|
|
|
|
|
|
|
|
|
4 |
"eval_runtime": 56.3363,
|
5 |
"eval_samples_per_second": 0.071,
|
6 |
"eval_steps_per_second": 0.036,
|
7 |
"eval_wer": 153.2258064516129,
|
|
|
|
|
|
|
|
|
|
|
8 |
"test_loss": 1.7568330764770508,
|
|
|
|
|
|
|
|
|
|
|
9 |
"test_runtime": 37.8582,
|
10 |
"test_samples_per_second": 0.106,
|
11 |
"test_steps_per_second": 0.053,
|
12 |
"test_wer": 138.5964912280702,
|
13 |
"train_loss": 1.4339025020599365,
|
14 |
-
"train_runtime":
|
15 |
-
"train_samples_per_second": 0.
|
16 |
-
"train_steps_per_second": 0.
|
17 |
}
|
|
|
1 |
{
|
2 |
"epoch": 1.0,
|
3 |
"eval_loss": 1.6191972494125366,
|
4 |
+
"eval_pretrained_loss": 1.6191972494125366,
|
5 |
+
"eval_pretrained_runtime": 59.7459,
|
6 |
+
"eval_pretrained_samples_per_second": 0.067,
|
7 |
+
"eval_pretrained_steps_per_second": 0.033,
|
8 |
+
"eval_pretrained_wer": 153.2258064516129,
|
9 |
"eval_runtime": 56.3363,
|
10 |
"eval_samples_per_second": 0.071,
|
11 |
"eval_steps_per_second": 0.036,
|
12 |
"eval_wer": 153.2258064516129,
|
13 |
+
"test_finetuned_loss": 1.7568330764770508,
|
14 |
+
"test_finetuned_runtime": 39.6579,
|
15 |
+
"test_finetuned_samples_per_second": 0.101,
|
16 |
+
"test_finetuned_steps_per_second": 0.05,
|
17 |
+
"test_finetuned_wer": 138.5964912280702,
|
18 |
"test_loss": 1.7568330764770508,
|
19 |
+
"test_pretrained_loss": 1.7568330764770508,
|
20 |
+
"test_pretrained_runtime": 42.5376,
|
21 |
+
"test_pretrained_samples_per_second": 0.094,
|
22 |
+
"test_pretrained_steps_per_second": 0.047,
|
23 |
+
"test_pretrained_wer": 138.5964912280702,
|
24 |
"test_runtime": 37.8582,
|
25 |
"test_samples_per_second": 0.106,
|
26 |
"test_steps_per_second": 0.053,
|
27 |
"test_wer": 138.5964912280702,
|
28 |
"train_loss": 1.4339025020599365,
|
29 |
+
"train_runtime": 102.2429,
|
30 |
+
"train_samples_per_second": 0.078,
|
31 |
+
"train_steps_per_second": 0.02
|
32 |
}
|
eval_pretrained_results.json
ADDED
@@ -0,0 +1,7 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"eval_pretrained_loss": 1.6191972494125366,
|
3 |
+
"eval_pretrained_runtime": 59.7459,
|
4 |
+
"eval_pretrained_samples_per_second": 0.067,
|
5 |
+
"eval_pretrained_steps_per_second": 0.033,
|
6 |
+
"eval_pretrained_wer": 153.2258064516129
|
7 |
+
}
|
huggingface_training.py
CHANGED
@@ -322,55 +322,55 @@ I hope you haven't left yet. If you have, bad for you, as we are ready for train
|
|
322 |
As Whisper is a pretrained model ready to be used off-the-shelf, it is advisable to evaluate it before training on both the validation and test sets. Let's make sure we make no harm to it.
|
323 |
"""
|
324 |
|
325 |
-
|
326 |
-
|
327 |
-
|
328 |
-
|
329 |
-
|
330 |
-
#
|
331 |
-
|
332 |
|
333 |
-
|
334 |
-
|
335 |
-
|
336 |
|
337 |
-
|
338 |
-
|
339 |
-
|
340 |
-
|
341 |
-
|
342 |
-
#
|
343 |
-
|
344 |
|
345 |
-
|
346 |
-
|
347 |
-
|
348 |
|
349 |
-
|
350 |
-
|
351 |
|
352 |
-
|
353 |
-
|
354 |
-
|
355 |
-
|
356 |
-
|
357 |
|
358 |
# """ADD SOMETHING ABOUT THE TRAINING.
|
359 |
|
360 |
# Now let's evaluate the
|
361 |
# """
|
362 |
|
363 |
-
|
364 |
-
|
365 |
-
|
366 |
-
|
367 |
-
|
368 |
-
#
|
369 |
-
|
370 |
|
371 |
-
|
372 |
-
|
373 |
-
|
374 |
|
375 |
# Pushing to hub during training slows down training
|
376 |
# so we push it only in the end.
|
|
|
322 |
As Whisper is a pretrained model ready to be used off-the-shelf, it is advisable to evaluate it before training on both the validation and test sets. Let's make sure we make no harm to it.
|
323 |
"""
|
324 |
|
325 |
+
eval_metrics = trainer.evaluate(
|
326 |
+
eval_dataset=preprocessed_dataset["validation"],
|
327 |
+
metric_key_prefix="eval_pretrained",
|
328 |
+
max_length=448,
|
329 |
+
num_beams=1,
|
330 |
+
# gen_kwargs={"key": value} to provide additional generation specific arguments by keyword
|
331 |
+
)
|
332 |
|
333 |
+
trainer.log_metrics("eval_pretrained", eval_metrics)
|
334 |
+
trainer.save_metrics("eval_pretrained", eval_metrics)
|
335 |
+
print(eval_metrics)
|
336 |
|
337 |
+
test_metrics = trainer.evaluate(
|
338 |
+
eval_dataset=preprocessed_dataset["test"],
|
339 |
+
metric_key_prefix="test_pretrained",
|
340 |
+
max_length=448,
|
341 |
+
num_beams=1,
|
342 |
+
# gen_kwargs={"key": value} to provide additional generation specific arguments by keyword
|
343 |
+
)
|
344 |
|
345 |
+
trainer.log_metrics("test_pretrained", test_metrics)
|
346 |
+
trainer.save_metrics("test_pretrained", test_metrics)
|
347 |
+
print(test_metrics)
|
348 |
|
349 |
+
train_result = trainer.train()
|
350 |
+
trainer.save_model()
|
351 |
|
352 |
+
metrics = train_result.metrics
|
353 |
+
trainer.log_metrics("train", metrics)
|
354 |
+
trainer.save_metrics("train", metrics)
|
355 |
+
trainer.save_state()
|
356 |
+
print(metrics)
|
357 |
|
358 |
# """ADD SOMETHING ABOUT THE TRAINING.
|
359 |
|
360 |
# Now let's evaluate the
|
361 |
# """
|
362 |
|
363 |
+
final_metrics = trainer.evaluate(
|
364 |
+
eval_dataset=preprocessed_dataset["test"],
|
365 |
+
metric_key_prefix="test_finetuned",
|
366 |
+
max_length=448,
|
367 |
+
num_beams=1,
|
368 |
+
# gen_kwargs={"key": value} to provide additional generation specific arguments by keyword
|
369 |
+
)
|
370 |
|
371 |
+
trainer.log_metrics("test_finetuned", final_metrics)
|
372 |
+
trainer.save_metrics("test_finetuned", final_metrics)
|
373 |
+
print(final_metrics)
|
374 |
|
375 |
# Pushing to hub during training slows down training
|
376 |
# so we push it only in the end.
|
test_finetuned_results.json
ADDED
@@ -0,0 +1,8 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"epoch": 1.0,
|
3 |
+
"test_finetuned_loss": 1.7568330764770508,
|
4 |
+
"test_finetuned_runtime": 39.6579,
|
5 |
+
"test_finetuned_samples_per_second": 0.101,
|
6 |
+
"test_finetuned_steps_per_second": 0.05,
|
7 |
+
"test_finetuned_wer": 138.5964912280702
|
8 |
+
}
|
test_pretrained_results.json
ADDED
@@ -0,0 +1,7 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"test_pretrained_loss": 1.7568330764770508,
|
3 |
+
"test_pretrained_runtime": 42.5376,
|
4 |
+
"test_pretrained_samples_per_second": 0.094,
|
5 |
+
"test_pretrained_steps_per_second": 0.047,
|
6 |
+
"test_pretrained_wer": 138.5964912280702
|
7 |
+
}
|
train_results.json
CHANGED
@@ -1,7 +1,7 @@
|
|
1 |
{
|
2 |
"epoch": 1.0,
|
3 |
"train_loss": 1.4339025020599365,
|
4 |
-
"train_runtime":
|
5 |
-
"train_samples_per_second": 0.
|
6 |
-
"train_steps_per_second": 0.
|
7 |
}
|
|
|
1 |
{
|
2 |
"epoch": 1.0,
|
3 |
"train_loss": 1.4339025020599365,
|
4 |
+
"train_runtime": 102.2429,
|
5 |
+
"train_samples_per_second": 0.078,
|
6 |
+
"train_steps_per_second": 0.02
|
7 |
}
|
trainer_state.json
CHANGED
@@ -16,9 +16,9 @@
|
|
16 |
{
|
17 |
"epoch": 0.5,
|
18 |
"eval_loss": 1.6191972494125366,
|
19 |
-
"eval_runtime":
|
20 |
-
"eval_samples_per_second": 0.
|
21 |
-
"eval_steps_per_second": 0.
|
22 |
"eval_wer": 153.2258064516129,
|
23 |
"step": 1
|
24 |
},
|
@@ -31,9 +31,9 @@
|
|
31 |
{
|
32 |
"epoch": 1.0,
|
33 |
"eval_loss": 1.6191972494125366,
|
34 |
-
"eval_runtime":
|
35 |
-
"eval_samples_per_second": 0.
|
36 |
-
"eval_steps_per_second": 0.
|
37 |
"eval_wer": 153.2258064516129,
|
38 |
"step": 2
|
39 |
},
|
@@ -42,9 +42,9 @@
|
|
42 |
"step": 2,
|
43 |
"total_flos": 196951080960000.0,
|
44 |
"train_loss": 1.4339025020599365,
|
45 |
-
"train_runtime":
|
46 |
-
"train_samples_per_second": 0.
|
47 |
-
"train_steps_per_second": 0.
|
48 |
}
|
49 |
],
|
50 |
"max_steps": 2,
|
|
|
16 |
{
|
17 |
"epoch": 0.5,
|
18 |
"eval_loss": 1.6191972494125366,
|
19 |
+
"eval_runtime": 38.6993,
|
20 |
+
"eval_samples_per_second": 0.103,
|
21 |
+
"eval_steps_per_second": 0.052,
|
22 |
"eval_wer": 153.2258064516129,
|
23 |
"step": 1
|
24 |
},
|
|
|
31 |
{
|
32 |
"epoch": 1.0,
|
33 |
"eval_loss": 1.6191972494125366,
|
34 |
+
"eval_runtime": 39.2638,
|
35 |
+
"eval_samples_per_second": 0.102,
|
36 |
+
"eval_steps_per_second": 0.051,
|
37 |
"eval_wer": 153.2258064516129,
|
38 |
"step": 2
|
39 |
},
|
|
|
42 |
"step": 2,
|
43 |
"total_flos": 196951080960000.0,
|
44 |
"train_loss": 1.4339025020599365,
|
45 |
+
"train_runtime": 102.2429,
|
46 |
+
"train_samples_per_second": 0.078,
|
47 |
+
"train_steps_per_second": 0.02
|
48 |
}
|
49 |
],
|
50 |
"max_steps": 2,
|
training_args.bin
CHANGED
@@ -1,3 +1,3 @@
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:
|
3 |
size 3579
|
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:c9cf1b9241b15d95f3310024d7bec9fb6d139a94c9d760c0be51787d41a93fc3
|
3 |
size 3579
|