You need to agree to share your contact information to access this model

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

mms-1b-all-lwazi-gcp

This model is a fine-tuned version of facebook/mms-1b-all on the Lwazi dataset. It achieves the following results on the evaluation set:

More information needed

More information needed

More information needed

The following hyperparameters were used during training:

learning_rate: 0.0001
train_batch_size: 16
eval_batch_size: 16
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: 200
training_steps: 2000
mixed_precision_training: Native AMP

Training Loss	Epoch	Step	Validation Loss	Wer
1.2757	0.1696	200	1.5263	0.8641
1.0044	0.3393	400	0.7391	0.6981
0.8441	0.5089	600	0.6358	0.6503
0.8042	0.6785	800	0.5251	0.5572
0.7301	0.8482	1000	0.4920	0.5228
0.6085	1.0178	1200	0.4222	0.4242
0.6014	1.1874	1400	0.3945	0.4101
0.5723	1.3571	1600	0.3659	0.3880
0.525	1.5267	1800	0.3385	0.3639
0.5201	1.6964	2000	0.3278	0.3548

Safetensors

Model size

1.0B params

Tensor type

F32

Base model

Finetuned

(367)

this model