w2v-bert-2.0-am-amharic-dataset-100hr-v2

KYAGABA/w2v-bert-2.0-am-amharic-dataset-100hr-v2

w2v-bert-2.0-am-amharic-dataset-100hr-v2 is a automatic speech recognition model from KYAGABA. Use it when you need speech turned into text. It is set up for transformers. The card lists the license as mit.

should probably proofread and complete it, then remove this comment. --

Downloads · 30 days

13% of all-time downloads

All-time downloads

Public

Parameters

581M

181 GB on disk

Likes

Public

Hugging Face

Repo makeup

Click a slice to open those files.

.safetensors2.3 GB · 100%

At a glance

Task: Automatic Speech Recognition
Library: transformers
License: mit
Model type: wav2vec2-bert
Access: Public
Created: Sep 3, 2024
Updated: Sep 11, 2024
SHA: 63a8fc00

Base models

facebook/w2v-bert-2.0

Model card

From the Hugging Face model README

w2v-bert-2.0-am-amharic-dataset-100hr-v2

This model is a fine-tuned version of facebook/w2v-bert-2.0 on an unknown dataset. It achieves the following results on the evaluation set:

Loss: 1.0133
Model Preparation Time: 0.0105
Wer: 0.4316
Cer: 0.1447

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 2e-05
train_batch_size: 32
eval_batch_size: 8
seed: 42
gradient_accumulation_steps: 2
total_train_batch_size: 64
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: linear
lr_scheduler_warmup_ratio: 0.05
num_epochs: 100
mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Model Preparation Time	Wer	Cer
28.1746	1.0	507	24.1319	0.0105	1.0	1.0
21.7375	2.0	1014	19.4962	0.0105	1.0	1.0
15.2036	3.0	1521	10.7916	0.0105	1.0	1.0
6.8433	4.0	2028	4.4202	0.0105	1.0	1.0
4.2252	5.0	2535	4.1002	0.0105	1.0	1.0
4.1549	6.0	3042	4.0767	0.0105	1.0	1.0
4.1323	7.0	3549	4.0700	0.0105	1.0	1.0
4.1051	8.0	4056	3.9675	0.0105	1.0	0.9825
3.0619	9.0	4563	2.1299	0.0105	0.7964	0.2808
1.7589	10.0	5070	1.3116	0.0105	0.4970	0.1352
1.21	11.0	5577	0.9147	0.0105	0.3922	0.1032
0.8527	12.0	6084	0.6185	0.0105	0.3401	0.0868
0.62	13.0	6591	0.4633	0.0105	0.2990	0.0755
0.4706	14.0	7098	0.3360	0.0105	0.2420	0.0606
0.3786	15.0	7605	0.2689	0.0105	0.2196	0.0549
0.3197	16.0	8112	0.2195	0.0105	0.1872	0.0447
0.2773	17.0	8619	0.1918	0.0105	0.1646	0.0392
0.2495	18.0	9126	0.1789	0.0105	0.1532	0.0368
0.2293	19.0	9633	0.1623	0.0105	0.1388	0.0332
0.211	20.0	10140	0.1561	0.0105	0.1346	0.0323
0.1969	21.0	10647	0.1537	0.0105	0.1297	0.0313
0.1811	22.0	11154	0.1447	0.0105	0.1251	0.0296
0.169	23.0	11661	0.1438	0.0105	0.1227	0.0296
0.159	24.0	12168	0.1478	0.0105	0.1202	0.0294
0.1478	25.0	12675	0.1419	0.0105	0.1145	0.0278
0.1374	26.0	13182	0.1387	0.0105	0.1124	0.0272
0.1275	27.0	13689	0.1498	0.0105	0.1146	0.0281
0.1189	28.0	14196	0.1395	0.0105	0.1107	0.0269
0.107	29.0	14703	0.1391	0.0105	0.1095	0.0260
0.0993	30.0	15210	0.1411	0.0105	0.1050	0.0252
0.0907	31.0	15717	0.1431	0.0105	0.1092	0.0259
0.0833	32.0	16224	0.1514	0.0105	0.1103	0.0264
0.081	33.0	16731	0.1447	0.0105	0.1087	0.0263
0.0753	34.0	17238	0.1454	0.0105	0.1045	0.0252
0.069	35.0	17745	0.1470	0.0105	0.1054	0.0250
0.0636	36.0	18252	0.1516	0.0105	0.1095	0.0256
0.0593	37.0	18759	0.1568	0.0105	0.1129	0.0267
0.0557	38.0	19266	0.1637	0.0105	0.1089	0.0258
0.0524	39.0	19773	0.1651	0.0105	0.1057	0.0254
0.0497	40.0	20280	0.1642	0.0105	0.1106	0.0263
0.0462	41.0	20787	0.1684	0.0105	0.1079	0.0259
0.0444	42.0	21294	0.1702	0.0105	0.1093	0.0259
0.0409	43.0	21801	0.1787	0.0105	0.1110	0.0271
0.0389	44.0	22308	0.1744	0.0105	0.1054	0.0253
0.036	45.0	22815	0.1748	0.0105	0.1075	0.0255
0.0341	46.0	23322	0.1808	0.0105	0.1123	0.0263
0.0313	47.0	23829	0.1833	0.0105	0.1121	0.0263
0.0295	48.0	24336	0.1752	0.0105	0.1068	0.0250
0.0274	49.0	24843	0.1864	0.0105	0.1056	0.0255

Framework versions

Transformers 4.44.2
Pytorch 2.1.0+cu118
Datasets 2.20.0
Tokenizers 0.19.1

Task: Automatic Speech Recognition
Library: transformers
Type: wav2vec2-bert
License: mit
Created: Sep 3, 2024
Updated: Sep 11, 2024