model-index: - name: tamasheq-99-final results: []

tamasheq-99-final

This model is a fine-tuned version of jonatasgrosman/wav2vec2-large-xlsr-53-arabic on the None dataset. It achieves the following results on the evaluation set:

• Cer: 16.2959
• Wer: 55.5334

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: 3e-05
• train_batch_size: 16
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500

Training results

step	tamasheq_wer	arabic_wer	tamasheq_cer	arabic_cer
Before train	104.985	23.1305	67.4458	7.30972
step 300	99.5513	23.0544	49.7078	7.1043
step 600	95.1147	22.5267	41.4515	6.0098
step 900	93.5194	21.0404	38.0867	5.52939
step 1200	92.5723	20.6224	37.0877	5.39751
step 1500	92.3009	20.9238	36.9915	5.6718
step 1800	92.0738	21.2699	36.3713	6.08877
step 2100	88.7338	21.9693	33.3648	5.9156
step 2400	87.1884	21.1333	31.8379	5.52939
step 2700	88.299	21.0705	31.4599	5.5078
step 3000	87.7866	21.5021	30.9039	6.29239
step 3300	84.2971	21.666	29.7455	5.97212
step 3600	83.8983	21.5732	28.6145	6.04748
step 3900	81.8544	22.1087	27.9359	5.99096
step 4200	82.9741	23.392	27.4288	6.4013
step 4500	83.8485	24.2452	27.0575	6.79164
step 4800	81.6052	22.666	26.6918	6.09457
step 5100	77.9661	22.4803	25.1084	6.0098
step 5400	77.2183	21.83	24.656	5.9156
step 5700	76.672	22.1078	24.2606	6.0802
step 6000	76.2712	22.7589	23.9236	6.41485
step 6300	75.7228	23.8737	23.7135	6.78222
step 6600	71.2363	23.177	22.196	6.39601
step 6900	69.8405	22.7125	21.574	6.21703
step 7200	72.9452	23.6679	21.0775	6.6918
step 7500	75.9222	24.7097	20.8999	7.17784
step 7800	67.4975	23.1305	20.6786	6.65034
step 8100	65.2542	23.1305	19.7361	6.49962
step 8400	61.7149	22.3874	18.426	6.12283
step 8700	63.8046	23.6679	18.2166	6.2679
step 9000	64.7059	24.1059	17.9952	6.66918
step 9300	67.5474	24.7097	17.6078	7.16843
step 9600	57.1286	23.3163	17.2385	6.66918
step 9900	58.2752	22.8054	17.1065	6.4431
step 10200	57.7767	24.2917	16.848	6.68802
step 10500	55.2841	25.1277	16.5033	7.12133
step 10800	52.5424	23.8272	15.9566	6.80106
step 11100	55.5334	24.6168	16.2959	6.94235

Framework versions

• Transformers 4.31.0
• Pytorch 2.0.1+cu118
• Datasets 2.14.4
• Tokenizers 0.13.3