Marialab/finetuned-whisper-small-1000-step

Finetuned Whisper small for darija speech translation

This model is a fine-tuned version of openai/whisper-small on the Darija-C dataset. It achieves the following results on the evaluation set:

• Loss: 0.0000
• Bleu: 0.7440

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: 8
• eval_batch_size: 4
• seed: 42
• optimizer: Use 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: 200
• training_steps: 1000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Bleu
4.1244	0.625	5	4.0913	0.0
4.1401	1.25	10	3.8806	0.0
3.5438	1.875	15	3.0904	0.0
2.7946	2.5	20	2.2453	0.0023
2.1793	3.125	25	1.7106	0.0083
1.6133	3.75	30	1.2200	0.0310
1.1125	4.375	35	0.8124	0.1554
0.8674	5.0	40	0.4519	0.4140
0.4645	5.625	45	0.2318	0.5646
0.2348	6.25	50	0.1173	0.6654
0.1596	6.875	55	0.0513	0.7341
0.0745	7.5	60	0.0323	0.7247
0.0447	8.125	65	0.0136	0.7440
0.014	8.75	70	0.0113	0.7284
0.0185	9.375	75	0.0107	0.7352
0.0638	10.0	80	0.0421	0.7070
0.0472	10.625	85	0.0503	0.6970
0.0681	11.25	90	0.0879	0.6954
0.1465	11.875	95	0.0407	0.6819
0.0483	12.5	100	0.0835	0.6678
0.1844	13.125	105	0.0661	0.6744
0.0737	13.75	110	0.1486	0.6494
0.1454	14.375	115	0.1018	0.6439
0.1203	15.0	120	0.0444	0.7143
0.0858	15.625	125	0.0148	0.7320
0.0463	16.25	130	0.0726	0.6406
0.1464	16.875	135	0.0586	0.6699
0.0938	17.5	140	0.0447	0.6639
0.1116	18.125	145	0.0737	0.6801
0.1031	18.75	150	0.0906	0.6794
0.1601	19.375	155	0.1172	0.6540
0.1957	20.0	160	0.0271	0.7095
0.0043	20.625	165	0.0491	0.6874
0.1013	21.25	170	0.0221	0.7341
0.0506	21.875	175	0.0313	0.6938
0.0545	22.5	180	0.0664	0.6533
0.1434	23.125	185	0.0586	0.6346
0.0891	23.75	190	0.0947	0.6823
0.1784	24.375	195	0.1534	0.6343
0.3143	25.0	200	0.1054	0.6431
0.182	25.625	205	0.0546	0.6610
0.0698	26.25	210	0.0816	0.6662
0.1513	26.875	215	0.0420	0.7162
0.0759	27.5	220	0.0995	0.6411
0.191	28.125	225	0.0334	0.7012
0.0429	28.75	230	0.0748	0.6273
0.1608	29.375	235	0.1665	0.5937
0.2917	30.0	240	0.1436	0.6353
0.2379	30.625	245	0.0348	0.6940
0.0835	31.25	250	0.0238	0.7153
0.0293	31.875	255	0.0581	0.6983
0.0946	32.5	260	0.0471	0.7104
0.1223	33.125	265	0.0660	0.7389
0.1151	33.75	270	0.0598	0.7160
0.1367	34.375	275	0.1139	0.6796
0.1004	35.0	280	0.0553	0.7200
0.0921	35.625	285	0.0396	0.6818
0.0523	36.25	290	0.0691	0.6757
0.0866	36.875	295	0.0505	0.7211
0.1391	37.5	300	0.0480	0.6985
0.0674	38.125	305	0.0701	0.6544
0.058	38.75	310	0.0546	0.7081
0.1008	39.375	315	0.0587	0.6832
0.0989	40.0	320	0.0435	0.6986
0.053	40.625	325	0.0094	0.7107
0.0164	41.25	330	0.0218	0.7248
0.0541	41.875	335	0.0036	0.7274
0.0086	42.5	340	0.0126	0.7213
0.0288	43.125	345	0.0004	0.7440
0.0006	43.75	350	0.0007	0.7440
0.0008	44.375	355	0.0201	0.7187
0.0334	45.0	360	0.0220	0.7380
0.0401	45.625	365	0.0002	0.7440
0.0003	46.25	370	0.0375	0.7178
0.0575	46.875	375	0.0009	0.7440
0.0011	47.5	380	0.0088	0.7250
0.1052	48.125	385	0.0353	0.7248
0.0138	48.75	390	0.0002	0.7440
0.0003	49.375	395	0.0003	0.7440
0.0007	50.0	400	0.0001	0.7440
0.0001	50.625	405	0.0037	0.7415
0.0001	51.25	410	0.0158	0.7415
0.0387	51.875	415	0.0001	0.7440
0.0001	52.5	420	0.0008	0.7440
0.0025	53.125	425	0.0001	0.7440
0.0001	53.75	430	0.0001	0.7440
0.0001	54.375	435	0.0001	0.7440
0.0001	55.0	440	0.0001	0.7440
0.0	55.625	445	0.0000	0.7440
0.0	56.25	450	0.0000	0.7440
0.0	56.875	455	0.0000	0.7440
0.0	57.5	460	0.0000	0.7440
0.0	58.125	465	0.0000	0.7440
0.0	58.75	470	0.0000	0.7440
0.0	59.375	475	0.0000	0.7440
0.0	60.0	480	0.0000	0.7440
0.0	60.625	485	0.0000	0.7440
0.0	61.25	490	0.0000	0.7440
0.0	61.875	495	0.0000	0.7440
0.0	62.5	500	0.0000	0.7440
0.0	63.125	505	0.0000	0.7440
0.0	63.75	510	0.0000	0.7440
0.0	64.375	515	0.0000	0.7440
0.0	65.0	520	0.0000	0.7440
0.0	65.625	525	0.0000	0.7440
0.0	66.25	530	0.0000	0.7440
0.0	66.875	535	0.0000	0.7440
0.0	67.5	540	0.0000	0.7440
0.0	68.125	545	0.0000	0.7440
0.0	68.75	550	0.0000	0.7440
0.0	69.375	555	0.0000	0.7440
0.0	70.0	560	0.0000	0.7440
0.0	70.625	565	0.0000	0.7440
0.0	71.25	570	0.0000	0.7440
0.0	71.875	575	0.0000	0.7440
0.0	72.5	580	0.0000	0.7440
0.0	73.125	585	0.0000	0.7440
0.0	73.75	590	0.0000	0.7440
0.0	74.375	595	0.0000	0.7440
0.0	75.0	600	0.0000	0.7440
0.0	75.625	605	0.0000	0.7440
0.0	76.25	610	0.0000	0.7440
0.0	76.875	615	0.0000	0.7440
0.0	77.5	620	0.0000	0.7440
0.0	78.125	625	0.0000	0.7440
0.0	78.75	630	0.0000	0.7440
0.0	79.375	635	0.0000	0.7440
0.0	80.0	640	0.0000	0.7440
0.0	80.625	645	0.0000	0.7440
0.0	81.25	650	0.0000	0.7440
0.0	81.875	655	0.0000	0.7440
0.0	82.5	660	0.0000	0.7440
0.0	83.125	665	0.0000	0.7440
0.0	83.75	670	0.0000	0.7440
0.0	84.375	675	0.0000	0.7440
0.0	85.0	680	0.0000	0.7440
0.0	85.625	685	0.0000	0.7440
0.0	86.25	690	0.0000	0.7440
0.0	86.875	695	0.0000	0.7440
0.0	87.5	700	0.0000	0.7440
0.0	88.125	705	0.0000	0.7440
0.0	88.75	710	0.0000	0.7440
0.0	89.375	715	0.0000	0.7440
0.0	90.0	720	0.0000	0.7440
0.0	90.625	725	0.0000	0.7440
0.0	91.25	730	0.0000	0.7440
0.0	91.875	735	0.0000	0.7440
0.0	92.5	740	0.0000	0.7440
0.0	93.125	745	0.0000	0.7440
0.0	93.75	750	0.0000	0.7440
0.0	94.375	755	0.0000	0.7440
0.0	95.0	760	0.0000	0.7440
0.0	95.625	765	0.0000	0.7440
0.0	96.25	770	0.0000	0.7440
0.0	96.875	775	0.0000	0.7440
0.0	97.5	780	0.0000	0.7440
0.0	98.125	785	0.0000	0.7440
0.0	98.75	790	0.0000	0.7440
0.0	99.375	795	0.0000	0.7440
0.0	100.0	800	0.0000	0.7440
0.0	100.625	805	0.0000	0.7440
0.0	101.25	810	0.0000	0.7440
0.0	101.875	815	0.0000	0.7440
0.0	102.5	820	0.0000	0.7440
0.0	103.125	825	0.0000	0.7440
0.0	103.75	830	0.0000	0.7440
0.0	104.375	835	0.0000	0.7440
0.0	105.0	840	0.0000	0.7440
0.0	105.625	845	0.0000	0.7440
0.0	106.25	850	0.0000	0.7440
0.0	106.875	855	0.0000	0.7440
0.0	107.5	860	0.0000	0.7440
0.0	108.125	865	0.0000	0.7440
0.0	108.75	870	0.0000	0.7440
0.0	109.375	875	0.0000	0.7440
0.0	110.0	880	0.0000	0.7440
0.0	110.625	885	0.0000	0.7440
0.0	111.25	890	0.0000	0.7440
0.0	111.875	895	0.0000	0.7440
0.0	112.5	900	0.0000	0.7440
0.0	113.125	905	0.0000	0.7440
0.0	113.75	910	0.0000	0.7440
0.0	114.375	915	0.0000	0.7440
0.0	115.0	920	0.0000	0.7440
0.0	115.625	925	0.0000	0.7440
0.0	116.25	930	0.0000	0.7440
0.0	116.875	935	0.0000	0.7440
0.0	117.5	940	0.0000	0.7440
0.0	118.125	945	0.0000	0.7440
0.0	118.75	950	0.0000	0.7440
0.0	119.375	955	0.0000	0.7440
0.0	120.0	960	0.0000	0.7440
0.0	120.625	965	0.0000	0.7440
0.0	121.25	970	0.0000	0.7440
0.0	121.875	975	0.0000	0.7440
0.0	122.5	980	0.0000	0.7440
0.0	123.125	985	0.0000	0.7440
0.0	123.75	990	0.0000	0.7440
0.0	124.375	995	0.0000	0.7440
0.0	125.0	1000	0.0000	0.7440

Framework versions

• Transformers 4.47.0
• Pytorch 2.5.1+cu121
• Datasets 2.19.2
• Tokenizers 0.21.0