ssc-kbd-mms-model-mix-adapt-max

This model is a fine-tuned version of facebook/mms-1b-all on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.2593
• Cer: 0.1091
• Wer: 0.5604

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: 0.001
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED 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: 100
• num_epochs: 10
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer	Wer
0.6422	0.1719	200	0.4864	0.1765	0.8201
0.4943	0.3438	400	0.4324	0.1614	0.7674
0.4538	0.5157	600	0.4279	0.1583	0.7394
0.4292	0.6876	800	0.3844	0.1462	0.7058
0.4018	0.8595	1000	0.3818	0.1411	0.6929
0.3847	1.0309	1200	0.3429	0.1346	0.6741
0.3785	1.2028	1400	0.3468	0.1368	0.6722
0.3708	1.3747	1600	0.3546	0.1335	0.6683
0.3698	1.5466	1800	0.3337	0.1320	0.6632
0.3535	1.7185	2000	0.3305	0.1304	0.6569
0.3683	1.8904	2200	0.3305	0.1285	0.6456
0.3584	2.0619	2400	0.3263	0.1284	0.6398
0.3425	2.2338	2600	0.3225	0.1300	0.6453
0.354	2.4057	2800	0.3208	0.1279	0.6361
0.3379	2.5776	3000	0.3130	0.1256	0.6274
0.3367	2.7495	3200	0.3156	0.1258	0.6284
0.341	2.9214	3400	0.3143	0.1258	0.6303
0.3171	3.0928	3600	0.3143	0.1253	0.6394
0.3106	3.2647	3800	0.3111	0.1225	0.6183
0.3194	3.4366	4000	0.3053	0.1238	0.6266
0.3058	3.6085	4200	0.3048	0.1235	0.6284
0.3188	3.7804	4400	0.3045	0.1219	0.6193
0.3117	3.9523	4600	0.2981	0.1207	0.6137
0.2962	4.1238	4800	0.2977	0.1204	0.6132
0.3019	4.2957	5000	0.2946	0.1190	0.6037
0.2972	4.4676	5200	0.3031	0.1195	0.6071
0.2879	4.6394	5400	0.2928	0.1188	0.6014
0.2934	4.8113	5600	0.2938	0.1174	0.5984
0.2932	4.9832	5800	0.2909	0.1183	0.6051
0.2884	5.1547	6000	0.2854	0.1160	0.5833
0.2864	5.3266	6200	0.2844	0.1158	0.5821
0.2837	5.4985	6400	0.2810	0.1159	0.5901
0.2779	5.6704	6600	0.2746	0.1139	0.5786
0.2739	5.8423	6800	0.2774	0.1147	0.5782
0.2754	6.0138	7000	0.2750	0.1132	0.5745
0.27	6.1856	7200	0.2792	0.1134	0.5756
0.2602	6.3575	7400	0.2775	0.1145	0.5790
0.2611	6.5294	7600	0.2765	0.1140	0.5781
0.2605	6.7013	7800	0.2790	0.1135	0.5782
0.2611	6.8732	8000	0.2741	0.1124	0.5714
0.2637	7.0447	8200	0.2753	0.1126	0.5691
0.251	7.2166	8400	0.2689	0.1123	0.5744
0.2562	7.3885	8600	0.2690	0.1116	0.5686
0.2608	7.5604	8800	0.2667	0.1125	0.5744
0.255	7.7323	9000	0.2708	0.1124	0.5762
0.2466	7.9042	9200	0.2657	0.1113	0.5713
0.2331	8.0756	9400	0.2656	0.1110	0.5666
0.2358	8.2475	9600	0.2670	0.1100	0.5675
0.2447	8.4194	9800	0.2654	0.1108	0.5663
0.2438	8.5913	10000	0.2630	0.1096	0.5620
0.2311	8.7632	10200	0.2601	0.1087	0.5590
0.2429	8.9351	10400	0.2602	0.1095	0.5612
0.2351	9.1066	10600	0.2602	0.1105	0.5666
0.2293	9.2785	10800	0.2593	0.1094	0.5614
0.2342	9.4504	11000	0.2586	0.1092	0.5588
0.2325	9.6223	11200	0.2598	0.1090	0.5595
0.2327	9.7942	11400	0.2598	0.1092	0.5612
0.2289	9.9661	11600	0.2593	0.1091	0.5604

Framework versions

• Transformers 4.57.2
• Pytorch 2.9.1+cu128
• Datasets 3.6.0
• Tokenizers 0.22.0