Fauna-v3.3 - Rootflo

This model is a fine-tuned version of openai/whisper-large-v3 on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.0704
• Wer: 26.2284

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: 2e-06
• train_batch_size: 96
• eval_batch_size: 96
• seed: 42
• distributed_type: multi-GPU
• num_devices: 4
• gradient_accumulation_steps: 2
• total_train_batch_size: 768
• total_eval_batch_size: 384
• optimizer: Use adamw_bnb_8bit with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 50
• training_steps: 0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.2281	1.0	460	0.1114	35.7178
0.1647	2.0	920	0.0913	31.4522
0.1418	3.0	1380	0.0819	29.2863
0.1222	4.0	1840	0.0766	28.2008
0.1136	5.0	2300	0.0728	26.6919
0.1077	6.0	2760	0.0712	26.3193
0.104	7.0	3220	0.0705	129.6359
0.1028	8.0	3680	0.0704	26.2284

Framework versions

• Transformers 4.46.1
• Pytorch 2.5.1+cu124
• Datasets 3.0.2
• Tokenizers 0.20.3