gemma-7b-spanishbillionwords

This model is a fine-tuned version of google/gemma-7b on Spanish Billion Words. This is the base Gemma model fine-tuned to perform better on spanish language. It achieves the following results on the evaluation set:

• Loss: 12.1686

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: 0.0002
• train_batch_size: 1
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 4
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 2
• training_steps: 30
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss
3.7924	0.0	1	12.2361
4.4499	0.0	2	12.1524
3.9217	0.0	3	12.0710
4.3292	0.0	4	12.1710
6.6619	0.0	5	12.1710
4.4917	0.0	6	12.2628
4.8346	0.0	7	12.3997
3.6987	0.0	8	12.4212
6.0457	0.0	9	12.4049
3.7882	0.0	10	12.4228
3.9878	0.0	11	12.4168
5.1707	0.0	12	12.3961
3.7024	0.0	13	12.3430
5.8496	0.0	14	12.3009
5.1708	0.0	15	12.2863
4.9796	0.0	16	12.2789
4.3754	0.0	17	12.2600
4.8339	0.0	18	12.2371
4.0352	0.0	19	12.2284
3.9643	0.0	20	12.2266
3.6923	0.0	21	12.2103
4.8213	0.0	22	12.2015
3.8048	0.0	23	12.1901
4.3145	0.0	24	12.1837
3.6633	0.0	25	12.1811
4.2401	0.0	26	12.1775
3.3954	0.0	27	12.1757
5.631	0.0	28	12.1720
3.8886	0.0	29	12.1714
4.3891	0.0	30	12.1686

Framework versions

• PEFT 0.8.2
• Transformers 4.38.0
• Pytorch 2.2.1+cu121
• Datasets 2.17.0
• Tokenizers 0.15.2