gemini_chakma_ai4indic

This model is a fine-tuned version of ai4bharat/indic-bert on the None dataset. It achieves the following results on the evaluation set:

• Loss: 3.5834

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-05
• train_batch_size: 4
• eval_batch_size: 4
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 8
• optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• num_epochs: 20
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss
4.1266	1.0	162	4.5210
3.653	2.0	324	4.2779
3.4327	3.0	486	4.0695
3.317	4.0	648	3.9998
3.2046	5.0	810	3.9530
3.1096	6.0	972	3.8994
3.0451	7.0	1134	3.8628
3.0244	8.0	1296	3.7406
2.9736	9.0	1458	3.6823
2.9345	10.0	1620	3.7661
2.8932	11.0	1782	3.6943
2.8789	12.0	1944	3.6660
2.8254	13.0	2106	3.6488
2.8338	14.0	2268	3.6219
2.7854	15.0	2430	3.5199
2.7878	16.0	2592	3.6030
2.8031	17.0	2754	3.5799
2.7622	18.0	2916	3.5498
2.7232	19.0	3078	3.5592
2.7826	20.0	3240	3.5834

Framework versions

• Transformers 4.56.1
• Pytorch 2.8.0+cu126
• Datasets 4.0.0
• Tokenizers 0.22.0