learning_rate: 1e-05
train_batch_size: 2
eval_batch_size: 8
seed: 42
distributed_type: multi-GPU
gradient_accumulation_steps: 8
total_train_batch_size: 16
optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
lr_scheduler_type: linear
lr_scheduler_warmup_steps: 500
num_epochs: 100
mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
9.9472	0.9814	33	10.1025	1.0079
9.9673	1.9814	66	10.0618	1.0090
9.8936	2.9814	99	9.9765	1.0157
9.7701	3.9814	132	9.8477	1.0202
9.6077	4.9814	165	9.6755	1.0337
9.4201	5.9814	198	9.4574	1.0472
9.1832	6.9814	231	9.1933	1.0664
8.8801	7.9814	264	8.8763	1.0787
8.5414	8.9814	297	8.5257	1.0877
8.1854	9.9814	330	8.1284	1.0709
7.763	10.9814	363	7.6892	1.0529
7.2828	11.9814	396	7.1843	1.0304
6.7643	12.9814	429	6.6300	1.0124
6.2062	13.9814	462	6.0169	1.0022
5.5846	14.9814	495	5.3558	1.0
4.92	15.9814	528	4.6667	1.0
4.3568	16.9814	561	4.1007	1.0
3.9133	17.9814	594	3.7354	1.0
3.6595	18.9814	627	3.5499	1.0
3.532	19.9814	660	3.4643	1.0
3.4538	20.9814	693	3.4138	1.0
3.4068	21.9814	726	3.3794	1.0
3.3777	22.9814	759	3.3551	1.0
3.3522	23.9814	792	3.3338	1.0
3.3334	24.9814	825	3.3149	1.0
3.313	25.9814	858	3.2934	1.0
3.2863	26.9814	891	3.2699	1.0
3.2705	27.9814	924	3.2419	1.0
3.239	28.9814	957	3.2087	1.0
3.2112	29.9814	990	3.1722	0.9989
3.1728	30.9814	1023	3.1235	0.9966
3.1277	31.9814	1056	3.0639	0.9966
3.0739	32.9814	1089	2.9944	0.9978
3.0063	33.9814	1122	2.9176	0.9978
2.9492	34.9814	1155	2.8354	0.9978
2.8745	35.9814	1188	2.7467	0.9989
2.8018	36.9814	1221	2.6584	0.9978
2.7188	37.9814	1254	2.5763	0.9989
2.6582	38.9814	1287	2.4999	0.9933
2.5816	39.9814	1320	2.4275	0.9910
2.5207	40.9814	1353	2.3672	0.9831
2.4655	41.9814	1386	2.3115	0.9831
2.4185	42.9814	1419	2.2616	0.9764
2.383	43.9814	1452	2.2175	0.9764
2.3566	44.9814	1485	2.1813	0.9719
2.3175	45.9814	1518	2.1477	0.9685
2.2717	46.9814	1551	2.1162	0.9663
2.2673	47.9814	1584	2.0888	0.9629
2.2369	48.9814	1617	2.0654	0.9606
2.2121	49.9814	1650	2.0425	0.9606
2.197	50.9814	1683	2.0228	0.9573
2.1756	51.9814	1716	2.0032	0.9550
2.1686	52.9814	1749	1.9858	0.9561
2.1429	53.9814	1782	1.9690	0.9516
2.1337	54.9814	1815	1.9542	0.9516
2.1211	55.9814	1848	1.9408	0.9505
2.1063	56.9814	1881	1.9288	0.9483
2.092	57.9814	1914	1.9173	0.9471
2.0887	58.9814	1947	1.9069	0.9471
2.0763	59.9814	1980	1.8977	0.9438
2.0762	60.9814	2013	1.8890	0.9460
2.0668	61.9814	2046	1.8809	0.9415
2.0594	62.9814	2079	1.8732	0.9404
2.0534	63.9814	2112	1.8659	0.9426
2.0405	64.9814	2145	1.8594	0.9426
2.0395	65.9814	2178	1.8530	0.9393
2.0177	66.9814	2211	1.8471	0.9449
2.033	67.9814	2244	1.8416	0.9404
2.0289	68.9814	2277	1.8361	0.9393
2.0164	69.9814	2310	1.8312	0.9404
2.0012	70.9814	2343	1.8253	0.9393
1.9994	71.9814	2376	1.8225	0.9393
2.0064	72.9814	2409	1.8193	0.9404
2.0064	73.9814	2442	1.8147	0.9404
1.9899	74.9814	2475	1.8115	0.9404
1.9993	75.9814	2508	1.8082	0.9381
1.9953	76.9814	2541	1.8052	0.9393
1.99	77.9814	2574	1.8025	0.9393
1.9757	78.9814	2607	1.7995	0.9393
1.9755	79.9814	2640	1.7962	0.9404
1.9813	80.9814	2673	1.7936	0.9393
1.9851	81.9814	2706	1.7917	0.9393
1.975	82.9814	2739	1.7895	0.9404
1.9693	83.9814	2772	1.7875	0.9370
1.9684	84.9814	2805	1.7854	0.9370
1.9619	85.9814	2838	1.7836	0.9370
1.963	86.9814	2871	1.7819	0.9359
1.9678	87.9814	2904	1.7804	0.9359
1.9611	88.9814	2937	1.7791	0.9359
1.9659	89.9814	2970	1.7778	0.9359
1.9627	90.9814	3003	1.7767	0.9348
1.9517	91.9814	3036	1.7758	0.9348
1.9563	92.9814	3069	1.7749	0.9348
1.9574	93.9814	3102	1.7739	0.9348
1.9665	94.9814	3135	1.7733	0.9348
1.9416	95.9814	3168	1.7728	0.9336
1.9613	96.9814	3201	1.7723	0.9336
1.955	97.9814	3234	1.7720	0.9336
1.9579	98.9814	3267	1.7720	0.9336
1.9555	99.9814	3300	1.7716	0.9348

Framework versions

Transformers 4.51.3
Pytorch 2.4.1
Datasets 2.16.1
Tokenizers 0.21.1

sqrk/mms-1b-allFT-Dahnon-ara

mms-1b-allFT-Dahnon-ara

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

Training results

Framework versions

sqrk/
mms-1b-allFT-Dahnon-ara