Hugging Face's logo

sanchit-gandhi
/
parler-tts-600M-cross-attention-rope

Text2Text Generation
Transformers
parler_tts
Inference Endpoints
Model card Files Community
sanchit-gandhi HF staff commited on
Commit
0c0e7ab
•
1 Parent(s): 61218be

Saving train state of step 40000

Browse files
Files changed (16) hide show
  1. checkpoint-40000-epoch-2/optimizer.bin +3 -0
  2. checkpoint-40000-epoch-2/pytorch_model.bin +3 -0
  3. checkpoint-40000-epoch-2/random_states_0.pkl +3 -0
  4. checkpoint-40000-epoch-2/random_states_1.pkl +3 -0
  5. checkpoint-40000-epoch-2/random_states_2.pkl +3 -0
  6. checkpoint-40000-epoch-2/random_states_3.pkl +3 -0
  7. checkpoint-40000-epoch-2/random_states_4.pkl +3 -0
  8. checkpoint-40000-epoch-2/random_states_5.pkl +3 -0
  9. checkpoint-40000-epoch-2/random_states_6.pkl +3 -0
  10. checkpoint-40000-epoch-2/random_states_7.pkl +3 -0
  11. checkpoint-40000-epoch-2/scheduler.bin +3 -0
  12. parler_tts/__pycache__/configuration_parler_tts.cpython-311.pyc +0 -0
  13. parler_tts/__pycache__/modeling_parler_tts.cpython-311.pyc +0 -0
  14. slurm_job.slurm +0 -1
  15. starting_point_0.01_rope.json +1 -2
  16. training/eval.py +3 -2
checkpoint-40000-epoch-2/optimizer.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e6645d44abbc401e120be5dc489364dff46c6bef4899ce4edf1b0c09c10713aa
3
+ size 3652769047
checkpoint-40000-epoch-2/pytorch_model.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a44c145e06a274382225a36112ccf24aebd12c5e123c08ced3e7fc450ec1bda1
3
+ size 2588465818
checkpoint-40000-epoch-2/random_states_0.pkl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4373bcf5389516c60ce27ee2770f4273321a8a399b28157b0b7a6a88b36f769a
3
+ size 16036
checkpoint-40000-epoch-2/random_states_1.pkl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f652448236a57a3fce494feeaca0a3aeb1dfcfcefbf87bd8830d9a7431c8879f
3
+ size 16036
checkpoint-40000-epoch-2/random_states_2.pkl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f7e0c39e93d05b96ee155ed4077d16db98441d2d8d8c3ff1cadd9957c13c8d29
3
+ size 16036
checkpoint-40000-epoch-2/random_states_3.pkl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1bf9eeeb4068d4f97ea9936d7f134238a8f1c6b55e04251765e5e949cb98847c
3
+ size 16100
checkpoint-40000-epoch-2/random_states_4.pkl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:def78a5322574acc6ada55a74473298a7ab0defa21ed086ebc32a7f541f4727a
3
+ size 16100
checkpoint-40000-epoch-2/random_states_5.pkl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:924d3701abf0ada3e0e8ebcd9b67e3a691c4d583480e8738584a24d86fcfb2d8
3
+ size 16100
checkpoint-40000-epoch-2/random_states_6.pkl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5ebd276339b71cac4a7347c4cbdce6537168ec50f1748b252d1e8e9f91639bf5
3
+ size 16100
checkpoint-40000-epoch-2/random_states_7.pkl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f9d13afe0f590dbbe29711ec15cbb9db7c54976da83b02bcc42d31b53b37230f
3
+ size 16100
checkpoint-40000-epoch-2/scheduler.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:93769288743218b7e58d6c9be3f8385f45d0347b2a28d09b0fbf0b28704da6e6
3
+ size 1000
parler_tts/__pycache__/configuration_parler_tts.cpython-311.pyc CHANGED
Binary files a/parler_tts/__pycache__/configuration_parler_tts.cpython-311.pyc and b/parler_tts/__pycache__/configuration_parler_tts.cpython-311.pyc differ
 
parler_tts/__pycache__/modeling_parler_tts.cpython-311.pyc CHANGED
Binary files a/parler_tts/__pycache__/modeling_parler_tts.cpython-311.pyc and b/parler_tts/__pycache__/modeling_parler_tts.cpython-311.pyc differ
 
slurm_job.slurm CHANGED
@@ -2,7 +2,6 @@
2
  #SBATCH --job-name=parler-tts
3
  #SBATCH --nodes=1
4
  # set 48h for job wall time limit
5
- #SBATCH --requeue
6
  #SBATCH --time=48:00:00
7
  #SBATCH --ntasks-per-node=1 # crucial - only 1 task per dist per node!
8
  #SBATCH --cpus-per-task=32
 
2
  #SBATCH --job-name=parler-tts
3
  #SBATCH --nodes=1
4
  # set 48h for job wall time limit
 
5
  #SBATCH --time=48:00:00
6
  #SBATCH --ntasks-per-node=1 # crucial - only 1 task per dist per node!
7
  #SBATCH --cpus-per-task=32
starting_point_0.01_rope.json CHANGED
@@ -1,9 +1,8 @@
1
  {
2
  "model_name_or_path": "parler-tts/parler-tts-untrained-600M-cross-attention-rope",
3
- "save_to_disk": "/scratch/tmp_dataset_audio/",
4
  "temporary_save_to_disk": "/scratch/tmp_dataset_audio/",
5
  "push_to_hub": true,
6
- "cache_dir": "/scratch/cache",
7
 
8
 
9
  "feature_extractor_name":"ylacombe/dac_44khZ_8kbps",
 
1
  {
2
  "model_name_or_path": "parler-tts/parler-tts-untrained-600M-cross-attention-rope",
3
+ "save_to_disk": "/fsx/yoach/tmp/artefacts/10k_hours_processed_punctuated/",
4
  "temporary_save_to_disk": "/scratch/tmp_dataset_audio/",
5
  "push_to_hub": true,
 
6
 
7
 
8
  "feature_extractor_name":"ylacombe/dac_44khZ_8kbps",
training/eval.py CHANGED
@@ -47,12 +47,13 @@ def wer(asr_model_name_or_path, prompts, audios, device, per_device_eval_batch_s
47
  normalized_references = []
48
 
49
  for pred, ref in zip(transcriptions, prompts):
50
- normalizer = english_normalizer if hasattr(pred, "language") and pred["language"] == "english" else basic_normalizer
 
51
  norm_ref = normalizer(ref)
52
  if len(norm_ref) > 0:
53
  norm_pred = normalizer(pred["text"])
54
  normalized_predictions.append(norm_pred)
55
- normalized_references.append(norm_pred)
56
 
57
  word_error = 100 * metric.compute(predictions=normalized_predictions, references=normalized_references)
58
 
 
47
  normalized_references = []
48
 
49
  for pred, ref in zip(transcriptions, prompts):
50
+ normalizer = english_normalizer
51
+
52
  norm_ref = normalizer(ref)
53
  if len(norm_ref) > 0:
54
  norm_pred = normalizer(pred["text"])
55
  normalized_predictions.append(norm_pred)
56
+ normalized_references.append(norm_ref)
57
 
58
  word_error = 100 * metric.compute(predictions=normalized_predictions, references=normalized_references)
59