W2V2_Bert_BIG_C_Bemba_20hr_v1

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: 0.6286
• Wer: 0.4136
• Cer: 0.1049

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: 5e-06
• train_batch_size: 4
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.025
• num_epochs: 100

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
2.1779	1.0	2546	0.8771	0.6175	0.1688
0.8215	2.0	5092	0.7074	0.5268	0.1557
0.7007	3.0	7638	0.6562	0.4932	0.1365
0.6419	4.0	10184	0.6431	0.4636	0.1263
0.5961	5.0	12730	0.6088	0.4746	0.1309
0.5581	6.0	15276	0.7341	0.4436	0.1240
0.5237	7.0	17822	0.6851	0.4406	0.1218
0.4935	8.0	20368	0.6648	0.4368	0.1206
0.4622	9.0	22914	0.6347	0.4320	0.1203
0.4284	10.0	25460	0.6552	0.4380	0.1206
0.4001	11.0	28006	0.6317	0.448	0.1259
0.3686	12.0	30552	0.6987	0.4329	0.1201
0.3397	13.0	33098	0.7235	0.4236	0.1183
0.3103	14.0	35644	0.7169	0.4386	0.1205
0.2839	15.0	38190	0.7628	0.4372	0.1226
0.2569	16.0	40736	0.7728	0.4476	0.1263
0.2316	17.0	43282	0.8605	0.4384	0.1255
0.2112	18.0	45828	0.8648	0.4619	0.1291
0.1889	19.0	48374	0.9461	0.4420	0.1280
0.1698	20.0	50920	1.0526	0.4482	0.1262
0.1529	21.0	53466	1.1109	0.4454	0.1249
0.136	22.0	56012	1.1421	0.4521	0.1283
0.1218	23.0	58558	1.1437	0.4585	0.1296

Framework versions

• Transformers 4.44.2
• Pytorch 2.2.0+cu121
• Datasets 3.0.0
• Tokenizers 0.19.1