BahiaLLmWorks/Bahia-1

Bahia 1.0 Based on Phi 1.5 tuned using Alpaca_en

This model is a fine-tuned version of microsoft/phi-1_5 on the alpaca_en dataset.

Model description

This is the First LLM coming from the Mediterranean city of Oran (nicknamed El-Bahia), and possibly the first North African LLM trained model, More to come from us soon ;-)

Intended uses & limitations

This comes under an MIT lisence, enjoy!

Training and evaluation data

Trained using Alpaca_en from Llama Factory

Training procedure

Locally trained on baremetal hardware on an AMD GPU under Linux Mint 21

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.0002
• train_batch_size: 2
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 8
• total_train_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• num_epochs: 3.0
• mixed_precision_training: Native AMP

Training results

Please check the files in the files and versions.

Framework versions

• PEFT 0.10.0
• Transformers 4.40.1
• Pytorch 2.3.0+rocm5.7
• Datasets 2.19.1
• Tokenizers 0.19.1