Cabra: A portuguese finetuned instruction commercial model

LoRA adapter created with the procedures detailed at the GitHub repository: https://github.com/gustrd/cabra .

This training was done at 1 epoch using P100 at Kaggle, by around 11 hours, at a random slice of the dataset.

This LoRA adapter was created following the procedure:

Training procedure

The following bitsandbytes quantization config was used during training:

• quant_method: bitsandbytes
• load_in_8bit: False
• load_in_4bit: True
• llm_int8_threshold: 6.0
• llm_int8_skip_modules: None
• llm_int8_enable_fp32_cpu_offload: False
• llm_int8_has_fp16_weight: False
• bnb_4bit_quant_type: fp4
• bnb_4bit_use_double_quant: False
• bnb_4bit_compute_dtype: float32

Framework versions

• PEFT 0.5.0.dev0