wav2vec2-large-xlsr-53-AL-1000

This model is a fine-tuned version of facebook/wav2vec2-large-xlsr-53 on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.3865
• Wer: 0.2321

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: 16
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• 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: 10
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
6.0156	0.1063	200	3.0290	1.0
2.9588	0.2126	400	2.6180	1.0
1.1877	0.3189	600	0.6898	0.6574
0.6517	0.4252	800	0.5704	0.5222
0.5701	0.5315	1000	0.5273	0.5166
0.5243	0.6378	1200	0.4915	0.4742
0.4977	0.7441	1400	0.4686	0.4341
0.4642	0.8504	1600	0.4500	0.4045
0.4541	0.9567	1800	0.4374	0.4073
0.4264	1.0627	2000	0.4095	0.3677
0.4004	1.1690	2200	0.4050	0.3632
0.3904	1.2753	2400	0.3954	0.3399
0.391	1.3816	2600	0.4028	0.3688
0.3721	1.4879	2800	0.3784	0.3173
0.3776	1.5942	3000	0.3721	0.3143
0.3614	1.7005	3200	0.3769	0.3214
0.3621	1.8068	3400	0.3712	0.3190
0.3564	1.9131	3600	0.3675	0.3122
0.3501	2.0191	3800	0.3578	0.2831
0.322	2.1254	4000	0.3491	0.2820
0.3087	2.2317	4200	0.3517	0.2860
0.3135	2.3380	4400	0.3478	0.2734
0.3135	2.4443	4600	0.3524	0.2747
0.3021	2.5506	4800	0.3418	0.2573
0.3015	2.6569	5000	0.3439	0.2627
0.3037	2.7632	5200	0.3428	0.2652
0.2971	2.8695	5400	0.3411	0.2576
0.3002	2.9758	5600	0.3354	0.2492
0.2659	3.0818	5800	0.3399	0.2512
0.2597	3.1881	6000	0.3355	0.2489
0.275	3.2944	6200	0.3332	0.2599
0.2703	3.4007	6400	0.3340	0.2489
0.2678	3.5070	6600	0.3296	0.2497
0.2646	3.6133	6800	0.3333	0.2495
0.2573	3.7196	7000	0.3266	0.2445
0.2604	3.8259	7200	0.3288	0.2400
0.2603	3.9322	7400	0.3243	0.2476
0.2582	4.0383	7600	0.3236	0.2468
0.2322	4.1446	7800	0.3292	0.2602
0.2296	4.2509	8000	0.3232	0.2397
0.2289	4.3572	8200	0.3246	0.2485
0.2267	4.4635	8400	0.3192	0.2382
0.2284	4.5698	8600	0.3193	0.2395
0.2336	4.6761	8800	0.3203	0.2385
0.2346	4.7824	9000	0.3164	0.2385
0.2279	4.8887	9200	0.3199	0.2365
0.2293	4.9950	9400	0.3223	0.2377
0.1982	5.1010	9600	0.3386	0.2415
0.198	5.2073	9800	0.3437	0.2444
0.2092	5.3136	10000	0.3343	0.2387
0.1964	5.4199	10200	0.3318	0.2381
0.1957	5.5262	10400	0.3307	0.2314
0.1951	5.6325	10600	0.3371	0.2343
0.1969	5.7388	10800	0.3328	0.2306
0.2021	5.8451	11000	0.3288	0.2327
0.2036	5.9514	11200	0.3278	0.2337
0.1828	6.0574	11400	0.3498	0.2297
0.1759	6.1637	11600	0.3464	0.2302
0.1702	6.2700	11800	0.3438	0.2311
0.1735	6.3763	12000	0.3469	0.2285
0.1689	6.4826	12200	0.3529	0.2288
0.1735	6.5889	12400	0.3380	0.2284
0.1687	6.6952	12600	0.3429	0.2261
0.1671	6.8015	12800	0.3426	0.2293
0.1752	6.9078	13000	0.3462	0.2263
0.1677	7.0138	13200	0.3379	0.2369
0.1486	7.1201	13400	0.3437	0.2328
0.1489	7.2264	13600	0.3440	0.2331
0.1484	7.3327	13800	0.3469	0.2375
0.1474	7.4390	14000	0.3455	0.2337
0.1463	7.5453	14200	0.3477	0.2310
0.1516	7.6516	14400	0.3450	0.2328
0.1489	7.7579	14600	0.3496	0.2318
0.1501	7.8642	14800	0.3421	0.2319
0.1478	7.9705	15000	0.3510	0.2383
0.1297	8.0765	15200	0.3693	0.2310
0.1288	8.1828	15400	0.3776	0.2321
0.1289	8.2891	15600	0.3683	0.2344
0.1384	8.3954	15800	0.3683	0.2295
0.1304	8.5017	16000	0.3712	0.2318
0.1328	8.6080	16200	0.3659	0.2306
0.1296	8.7143	16400	0.3624	0.2305
0.1292	8.8206	16600	0.3607	0.2312
0.1302	8.9269	16800	0.3659	0.2295
0.1266	9.0330	17000	0.3809	0.2304
0.1202	9.1393	17200	0.3856	0.2291
0.1191	9.2455	17400	0.3837	0.2317
0.1149	9.3518	17600	0.3847	0.2309
0.1113	9.4581	17800	0.3865	0.2330
0.1193	9.5644	18000	0.3867	0.2344
0.1162	9.6707	18200	0.3855	0.2351
0.1181	9.7770	18400	0.3853	0.2328
0.1186	9.8833	18600	0.3861	0.2321
0.1166	9.9896	18800	0.3865	0.2321

Framework versions

• Transformers 4.49.0.dev0
• Pytorch 2.5.1+cu121
• Datasets 3.2.0
• Tokenizers 0.21.0