llama-2-7b-chat-tagalog-v0.1d

Taga-llama-v0.1d

• Test Tagalog model: safe or accurate outputs not guaranteed (not for production use)!
• Observing effects of fine-tuning on a small Tagalog-focused dataset (14 manually edited items); will work with larger soon
• Trained on llama 7b chat for 30 epochs (compensating for small dataset, further trained and cutoff before overfit)
• Tends to still generate English or Taglish, for certain subjects (science-related)
• GGMLs
• QLoras (hf and GGML)