w2v-bert-grain-lg_cv_only_v2

This model is a fine-tuned version of facebook/w2v-bert-2.0 on the common_voice_17_0 dataset. It achieves the following results on the evaluation set:

• Loss: 0.6384
• Wer: 0.2320
• Cer: 0.0721

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: 0.0001
• train_batch_size: 8
• eval_batch_size: 4
• seed: 42
• optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• num_epochs: 80
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
0.3102	1.0	8884	0.4540	0.3644	0.1028
0.2032	2.0	17768	0.3881	0.3005	0.0845
0.1687	3.0	26652	0.4061	0.3139	0.0883
0.148	4.0	35536	0.4048	0.2879	0.0825
0.1327	5.0	44420	0.4136	0.2860	0.0831
0.1191	6.0	53304	0.3685	0.2889	0.0843
0.1087	7.0	62188	0.4108	0.2630	0.0810
0.0996	8.0	71072	0.3682	0.2628	0.0789
0.0918	9.0	79956	0.4126	0.2672	0.0779
0.0854	10.0	88840	0.3482	0.2628	0.0783
0.0778	11.0	97724	0.3948	0.2540	0.0773
0.0719	12.0	106608	0.3530	0.2477	0.0740
0.066	13.0	115492	0.4267	0.2604	0.0785
0.0595	14.0	124376	0.3779	0.2467	0.0727
0.0541	15.0	133260	0.4424	0.2622	0.0813
0.0485	16.0	142144	0.3848	0.2500	0.0755
0.044	17.0	151028	0.3752	0.2465	0.0736
0.0391	18.0	159912	0.3722	0.2524	0.0753
0.0347	19.0	168796	0.4386	0.2481	0.0762
0.0309	20.0	177680	0.4647	0.2552	0.0788
0.0273	21.0	186564	0.4453	0.2468	0.0736
0.0252	22.0	195448	0.4612	0.2450	0.0750
0.0229	23.0	204332	0.4624	0.2510	0.0750
0.0209	24.0	213216	0.4640	0.2535	0.0739
0.0186	25.0	222100	0.4309	0.2542	0.0747
0.0173	26.0	230984	0.4339	0.2490	0.0734
0.016	27.0	239868	0.4463	0.2477	0.0740
0.0143	28.0	248752	0.5788	0.2432	0.0784
0.0135	29.0	257636	0.4962	0.2482	0.0745
0.0124	30.0	266520	0.5620	0.2448	0.0794
0.0116	31.0	275404	0.5030	0.2419	0.0749
0.0108	32.0	284288	0.4731	0.2374	0.0729
0.0099	33.0	293172	0.4890	0.2425	0.0736
0.0095	34.0	302056	0.5449	0.2449	0.0783
0.0086	35.0	310940	0.5007	0.2355	0.0726
0.0082	36.0	319824	0.4715	0.2372	0.0738
0.0079	37.0	328708	0.5407	0.2430	0.0731
0.0072	38.0	337592	0.5361	0.2374	0.0738
0.0068	39.0	346476	0.5152	0.2459	0.0755
0.0063	40.0	355360	0.4737	0.2316	0.0715
0.0058	41.0	364244	0.5980	0.2391	0.0779
0.0052	42.0	373128	0.5633	0.2360	0.0727
0.0051	43.0	382012	0.5640	0.2352	0.0732
0.0046	44.0	390896	0.5674	0.2270	0.0710
0.0044	45.0	399780	0.5487	0.2352	0.0717
0.0042	46.0	408664	0.6279	0.2436	0.0786
0.0039	47.0	417548	0.6260	0.2438	0.0770
0.0038	48.0	426432	0.5995	0.2328	0.0763
0.0036	49.0	435316	0.6540	0.2403	0.0776
0.0031	50.0	444200	0.5347	0.2370	0.0747
0.0028	51.0	453084	0.6086	0.2490	0.0739
0.0026	52.0	461968	0.5515	0.2287	0.0693
0.0025	53.0	470852	0.6788	0.2414	0.0793
0.0023	54.0	479736	0.6384	0.2320	0.0721

Framework versions

• Transformers 4.46.1
• Pytorch 2.1.0+cu118
• Datasets 3.1.0
• Tokenizers 0.20.1