MistrAND-7B-v1

MistrAND 7B v1 is a fine-tuned version of mistralai/Mistral-7B-v0.1 on the OASST2 dataset converted to a custom Andalusian Spanish orthography.

This project is part of the Master's Degree Final Thesis titled Conservational AI for endangered languages: a preservation strategy for Tsakonian Greek upon the Andalusian Spanish case, aiming to preserve endangered languages by storing them in QLoRA adapters for unlimited use in the future.

Links

• Model repository on GitHub

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.0002
• train_batch_size: 2
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 1
• num_epochs: 1

Training results

Training Loss	Epoch	Step	Validation Loss
0.9639	0.1796	1000	0.8628
0.753	0.3593	2000	0.7898
0.713	0.5389	3000	0.7348
0.6756	0.7185	4000	0.6888
0.6665	0.8981	5000	0.6530

Framework versions

• PEFT 0.10.1.dev0
• Transformers 4.41.0.dev0
• Pytorch 2.3.0+cu121
• Datasets 2.19.1
• Tokenizers 0.19.1