ss-multilang-asr-st-finetuned-decoder

This model is a fine-tuned version of facebook/hf-seamless-m4t-medium on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 1.7983
• Model Preparation Time: 0.0149
• Wer: 0.8542
• Bleu: 0.2768
• Precisions: [0.4340828862884621, 0.29685577775870453, 0.23779068053877214, 0.1914906549156663]
• Brevity Penalty: 1.0
• Length Ratio: 1.0410
• Translation Length: 35977
• Reference Length: 34560

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: 5e-05
• train_batch_size: 4
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 6
• total_train_batch_size: 24
• 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_ratio: 0.05
• num_epochs: 5
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Model Preparation Time	Wer	Bleu	Precisions	Brevity Penalty	Length Ratio	Translation Length	Reference Length
12.7608	0.2784	20000	2.8066	0.0149	1.4463	0.1254	[0.22588904694167852, 0.13874601564763836, 0.10379760850310009, 0.07596674691543785]	1.0	1.6282	56240	34541
10.9289	0.5568	40000	2.4744	0.0149	1.2093	0.1609	[0.27369429217696084, 0.17579970446835522, 0.13528973950026582, 0.10304123823395143]	1.0	1.4202	49073	34553
10.0457	0.8352	60000	2.3169	0.0149	0.9066	0.2330	[0.3883495145631068, 0.2515326975476839, 0.1967376052385407, 0.15332690453230471]	1.0	1.0493	36256	34554
9.6721	1.1136	80000	2.2281	0.0149	0.9359	0.2336	[0.3850319243514103, 0.2532483723507411, 0.19802685979869408, 0.15413986548007166]	1.0	1.0745	37119	34546
9.4745	1.3920	100000	2.1353	0.0149	1.0404	0.2084	[0.34192439862542956, 0.2250012059234962, 0.17676442949700777, 0.13873243010097935]	1.0	1.2291	42486	34566
9.1014	1.6704	120000	2.0704	0.0149	0.8681	0.2590	[0.42204688781425864, 0.2793488074089175, 0.2198240926965449, 0.17358852860107157]	1.0	1.0231	35361	34561
7.4958	1.9488	140000	2.0211	0.0149	0.8963	0.2492	[0.4072540531820504, 0.26832370261826427, 0.21118224498506188, 0.1670232408194367]	1.0	1.0761	37193	34563
8.5763	2.2272	160000	1.9877	0.0149	0.8438	0.2714	[0.443215012835266, 0.2983539720692488, 0.23637848192695413, 0.1875141957883124]	0.9808	0.9810	33891	34549
8.54	2.5056	180000	1.9506	0.0149	0.8279	0.2736	[0.4470650201907883, 0.29728506787330317, 0.23560356097864657, 0.1869333161854543]	0.9892	0.9893	34174	34544
8.3762	2.7841	200000	1.9113	0.0149	0.8634	0.2661	[0.42729698182424547, 0.2857714972252417, 0.2268226557435021, 0.18094804010938925]	1.0	1.0414	35982	34551
8.2387	3.0624	220000	1.8902	0.0149	0.8407	0.2751	[0.44139749879477075, 0.2962411285376325, 0.23447237693813036, 0.1867913391941847]	1.0	1.0204	35263	34559
8.2291	3.3409	240000	1.8642	0.0149	0.9300	0.2459	[0.39067699733265593, 0.2642862731801466, 0.21062786397620387, 0.1679938280109112]	1.0	1.1397	39365	34541
8.0448	3.6193	260000	1.8468	0.0149	0.8495	0.2760	[0.4382000448028674, 0.2957506918819188, 0.23618702471482889, 0.18958333333333333]	1.0	1.0335	35712	34555
8.0853	3.8977	280000	1.8246	0.0149	0.8923	0.2616	[0.41458885941644563, 0.2807285418257171, 0.22391450690003367, 0.17962342612914403]	1.0	1.0906	37700	34567
8.099	4.1761	300000	1.8173	0.0149	0.8456	0.2798	[0.4427621861152142, 0.30020479812755996, 0.23983592713234406, 0.1923938752645338]	1.0	1.0191	35204	34545
8.0542	4.4545	320000	1.8117	0.0149	0.8436	0.2832	[0.4455193482688391, 0.30397925891400607, 0.24324405476819602, 0.19538413878562577]	1.0	1.0234	35352	34544
8.027	4.7329	340000	1.7983	0.0149	0.8542	0.2768	[0.4340828862884621, 0.29685577775870453, 0.23779068053877214, 0.1914906549156663]	1.0	1.0410	35977	34560

Framework versions

• Transformers 4.57.6
• Pytorch 2.7.1+cu118
• Datasets 4.0.0
• Tokenizers 0.22.2