nb-asr-north-saami-parakeet-v2-lr5e-4

Fine-tuned NeMo Parakeet TDT model for North Saami ASR.

Run Info

• Run name: parakeet_v5_full_lr5e-4_20260316_200220
• Model artifact: /home/javierr/git/nb-asr-parakeet/outputs/parakeet_v5_full_lr5e-4_20260316_200220/artifacts/model.nemo
• Weights & Biases run: https://wandb.ai/versae/nbailab/runs/2026-03-16_20-02-23

Validation Metrics During Training

step	val_loss	val_wer
500	1.425318	0.600886
1172	1.033970	0.538678
1844	0.775067	0.441631
2516	0.695292	0.400844
3188	0.730898	0.365682
3860	0.716703	0.358650
4532	0.784751	0.348805
5204	0.843597	0.345992
5876	0.909719	0.372714
6548	0.941020	0.341772
7220	0.942366	0.323488
7892	0.989544	0.329114
8564	1.047186	0.322082
9236	1.098891	0.323488
9908	1.101070	0.315049
10580	1.109831	0.309423
11252	1.131614	0.313643
11924	1.132424	0.320675
12596	1.174766	0.315049
13268	1.156192	0.320675

• Best during training: step 10580, val_loss 1.109831, val_wer 0.309423
• Final during training: step 13268, val_loss 1.156192, val_wer 0.320675

Final Evaluation Metrics

split	loss	wer
validation	1.149497	0.305813
test	1.971457	0.459347

Notes

• Validation/test metrics above are computed post-training from the exported .nemo model.
• Training validation table is parsed from logged Validation metrics logged to wandb entries.