sptest-rinna-w2v2-base-kanji-RSs-0112

This model is a fine-tuned version of rinna/japanese-wav2vec2-base on the None dataset. It achieves the following results on the evaluation set:

• Loss: 2.3348
• Cer: 0.4308
• Wer: 1.0

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-05
• train_batch_size: 4
• eval_batch_size: 1
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 8
• optimizer: Use OptimizerNames.ADAMW_HF with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 38750
• num_epochs: 50
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer	Wer
67.7037	1.0	3878	53.2477	0.9833	1.0
5.235	2.0	7756	7.4842	0.9833	1.0
4.9859	3.0	11634	7.2036	0.9833	1.0
4.4234	4.0	15512	6.3211	0.9184	1.0
3.4727	5.0	19390	4.7569	0.7974	1.0
2.4254	6.0	23268	3.2544	0.6374	1.0
1.7561	7.0	27146	2.5051	0.5555	1.0
1.3871	8.0	31024	2.0472	0.4927	1.0
1.2092	9.0	34902	2.0813	0.5507	1.0
1.0836	10.0	38780	1.7959	0.4693	1.0
0.9881	11.0	42658	1.6362	0.4419	1.0
0.9027	12.0	46536	1.6809	0.4509	1.0
0.8279	13.0	50414	1.6806	0.4437	1.0
0.7737	14.0	54292	1.6882	0.4390	1.0
0.7186	15.0	58170	1.6895	0.4339	1.0
0.6666	16.0	62048	1.7808	0.4459	1.0
0.6231	17.0	65926	1.7291	0.4400	1.0
0.577	18.0	69804	1.8842	0.4855	1.0
0.5412	19.0	73682	1.7856	0.4580	1.0
0.5075	20.0	77560	1.7994	0.4405	1.0
0.4859	21.0	81438	1.8741	0.4423	1.0
0.4559	22.0	85316	1.9860	0.4439	1.0
0.4245	23.0	89194	1.9108	0.4594	1.0
0.3977	24.0	93072	1.9349	0.4439	1.0
0.3669	25.0	96950	2.0505	0.4435	1.0
0.3544	26.0	100828	2.1177	0.4476	1.0
0.3274	27.0	104706	2.2021	0.4611	1.0
0.3108	28.0	108584	2.1119	0.4445	1.0
0.2923	29.0	112462	2.0917	0.4361	1.0
0.2832	30.0	116340	2.0835	0.4373	1.0
0.2595	31.0	120218	2.0585	0.4402	1.0
0.2378	32.0	124096	2.1043	0.4348	1.0
0.2306	33.0	127974	2.1317	0.4324	1.0
0.217	34.0	131852	2.1674	0.4418	1.0
0.2091	35.0	135730	2.2216	0.4466	1.0
0.193	36.0	139608	2.2977	0.4400	1.0
0.1903	37.0	143486	2.2430	0.4318	1.0
0.1769	38.0	147364	2.2251	0.4299	1.0
0.1684	39.0	151242	2.2592	0.4357	1.0
0.1635	40.0	155120	2.2510	0.4302	1.0
0.1567	41.0	158998	2.3432	0.4347	1.0
0.1518	42.0	162876	2.3085	0.4314	1.0
0.1424	43.0	166754	2.2984	0.4314	1.0
0.1448	44.0	170632	2.3247	0.4315	1.0
0.1401	45.0	174510	2.3353	0.4322	1.0
0.1389	46.0	178388	2.3436	0.4298	1.0
0.1353	47.0	182266	2.3329	0.4302	1.0
0.1337	48.0	186144	2.3272	0.4304	1.0
0.1332	49.0	190022	2.3333	0.4301	1.0
0.1365	50.0	193900	2.3348	0.4308	1.0

Framework versions

• Transformers 4.46.3
• Pytorch 2.3.1+cu121
• Datasets 3.2.0
• Tokenizers 0.20.3