harpomaxx/Llama-2-13b-hf-juris-adapter

Description:

Training Details

The complete training script used for finetuning the model is available here

Training Data

[harpomaxx/jurisgpt/] (https://huggingface.co/datasets/harpomaxx/jurisgpt)

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: nf4
• bnb_4bit_use_double_quant: True
• bnb_4bit_compute_dtype: float16

Framework versions

• PEFT 0.6.2