ikema-asr-youtube-romaji

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 5.4360
• Cer: 0.5728

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: 0.0003
• 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: 100
• num_epochs: 150
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer
5.7654	1.1117	100	2.9458	1.0
3.1287	2.2235	200	2.8488	1.0
2.9631	3.3352	300	2.9340	1.0
2.8705	4.4469	400	2.8182	1.0
2.7162	5.5587	500	2.7401	1.0
2.4733	6.6704	600	2.5031	0.7435
2.1943	7.7821	700	2.3506	0.7736
1.9286	8.8939	800	2.2760	0.6852
1.719	10.0	900	2.3650	0.6970
1.5783	11.1117	1000	2.3247	0.6237
1.4361	12.2235	1100	2.1219	0.5678
1.338	13.3352	1200	2.2438	0.5707
1.2396	14.4469	1300	2.5065	0.5927
1.2052	15.5587	1400	2.3047	0.6268
1.0853	16.6704	1500	2.4457	0.5839
1.0515	17.7821	1600	2.3201	0.5521
1.0278	18.8939	1700	2.4907	0.5751
0.9595	20.0	1800	2.7885	0.5627
0.8868	21.1117	1900	2.7341	0.5844
0.8572	22.2235	2000	2.8150	0.5741
0.8335	23.3352	2100	2.6458	0.5686
0.8002	24.4469	2200	2.7114	0.5580
0.7593	25.5587	2300	2.6934	0.5555
0.7298	26.6704	2400	2.8385	0.5753
0.6951	27.7821	2500	2.8914	0.5651
0.6568	28.8939	2600	2.8913	0.5712
0.6522	30.0	2700	2.9266	0.5818
0.6165	31.1117	2800	2.9721	0.5626
0.6044	32.2235	2900	3.1186	0.6141
0.5791	33.3352	3000	3.1073	0.5619
0.5613	34.4469	3100	3.1489	0.5626
0.5616	35.5587	3200	3.3946	0.5865
0.528	36.6704	3300	3.2624	0.5566
0.5308	37.7821	3400	3.0666	0.5800
0.4913	38.8939	3500	3.3816	0.5830
0.4989	40.0	3600	3.6244	0.5663
0.4657	41.1117	3700	3.5415	0.5849
0.4693	42.2235	3800	3.6088	0.5834
0.4479	43.3352	3900	3.8216	0.5726
0.4331	44.4469	4000	3.5104	0.5866
0.4364	45.5587	4100	3.2249	0.5602
0.4195	46.6704	4200	3.3988	0.5659
0.4215	47.7821	4300	3.5439	0.5613
0.3899	48.8939	4400	3.7876	0.5835
0.3906	50.0	4500	3.6664	0.5647
0.3777	51.1117	4600	3.8201	0.5887
0.3875	52.2235	4700	3.8447	0.5933
0.3542	53.3352	4800	4.0995	0.5845
0.3513	54.4469	4900	4.2251	0.5892
0.3471	55.5587	5000	4.1187	0.5852
0.339	56.6704	5100	4.2435	0.5868
0.3299	57.7821	5200	4.1356	0.5750
0.3242	58.8939	5300	4.0832	0.5985
0.3143	60.0	5400	4.4107	0.5636
0.3025	61.1117	5500	4.3838	0.5915
0.3101	62.2235	5600	4.4010	0.5827
0.2909	63.3352	5700	4.1933	0.5727
0.2942	64.4469	5800	4.7073	0.6168
0.2805	65.5587	5900	4.3165	0.5717
0.2784	66.6704	6000	4.3823	0.5760
0.2667	67.7821	6100	4.4065	0.5885
0.2612	68.8939	6200	4.5967	0.6013
0.2634	70.0	6300	4.5820	0.5947
0.2494	71.1117	6400	4.7173	0.5738
0.2538	72.2235	6500	4.6189	0.5717
0.2483	73.3352	6600	4.5211	0.5829
0.2295	74.4469	6700	4.4063	0.5692
0.2272	75.5587	6800	4.5388	0.5769
0.2364	76.6704	6900	4.1843	0.5412
0.2322	77.7821	7000	4.2108	0.5678
0.2144	78.8939	7100	4.3772	0.5618
0.216	80.0	7200	4.5920	0.6020
0.2116	81.1117	7300	4.4980	0.5811
0.2088	82.2235	7400	4.5054	0.5804
0.1999	83.3352	7500	4.3327	0.5696
0.1913	84.4469	7600	4.5404	0.5639
0.1909	85.5587	7700	4.5547	0.5722
0.1854	86.6704	7800	4.6510	0.5619
0.187	87.7821	7900	4.4342	0.5746
0.1785	88.8939	8000	4.8637	0.5746
0.1725	90.0	8100	4.7284	0.5605
0.1744	91.1117	8200	4.6714	0.5536
0.1654	92.2235	8300	4.6927	0.5663
0.1703	93.3352	8400	4.7706	0.5655
0.1544	94.4469	8500	4.7921	0.5631
0.1536	95.5587	8600	4.6269	0.5744
0.1558	96.6704	8700	4.4396	0.5716
0.151	97.7821	8800	4.5563	0.5618
0.1455	98.8939	8900	4.6194	0.5549
0.1463	100.0	9000	4.5304	0.5543
0.139	101.1117	9100	4.7211	0.5653
0.1313	102.2235	9200	4.9493	0.5656
0.1362	103.3352	9300	4.6552	0.5554
0.1307	104.4469	9400	5.0708	0.5656
0.1271	105.5587	9500	5.0151	0.5683
0.1248	106.6704	9600	4.9161	0.5758
0.1182	107.7821	9700	5.3568	0.5768
0.1182	108.8939	9800	4.9611	0.5646
0.1197	110.0	9900	5.1136	0.5580
0.1121	111.1117	10000	4.9628	0.5701
0.1106	112.2235	10100	5.1576	0.5856
0.1102	113.3352	10200	5.1091	0.5738
0.1036	114.4469	10300	5.1654	0.5721
0.1048	115.5587	10400	4.9932	0.5665
0.1019	116.6704	10500	5.0801	0.5611
0.1034	117.7821	10600	4.8809	0.5650
0.1004	118.8939	10700	5.0913	0.5752
0.0936	120.0	10800	5.0903	0.5691
0.095	121.1117	10900	5.0648	0.5760
0.0893	122.2235	11000	5.1733	0.5827
0.0896	123.3352	11100	5.2183	0.5715
0.086	124.4469	11200	5.1838	0.5751
0.0869	125.5587	11300	5.1150	0.5744
0.0791	126.6704	11400	5.1499	0.5656
0.0832	127.7821	11500	5.2312	0.5695
0.0827	128.8939	11600	5.2175	0.5732
0.0799	130.0	11700	5.2239	0.5777
0.0747	131.1117	11800	5.2310	0.5690
0.0729	132.2235	11900	5.2735	0.5764
0.073	133.3352	12000	5.2856	0.5721
0.0755	134.4469	12100	5.2211	0.5716
0.0748	135.5587	12200	5.3144	0.5639
0.0695	136.6704	12300	5.3409	0.5734
0.0673	137.7821	12400	5.4097	0.5758
0.0674	138.8939	12500	5.3630	0.5738
0.0657	140.0	12600	5.4513	0.5713
0.0669	141.1117	12700	5.4735	0.5784
0.0627	142.2235	12800	5.4461	0.5727
0.0665	143.3352	12900	5.4985	0.5752
0.0626	144.4469	13000	5.4834	0.5734
0.0616	145.5587	13100	5.4716	0.5728
0.062	146.6704	13200	5.4455	0.5735
0.0601	147.7821	13300	5.4347	0.5721

Framework versions

• Transformers 4.51.2
• Pytorch 2.6.0+cu124
• Datasets 3.5.0
• Tokenizers 0.21.1