ChipGPT-Llama2-SFT-7B/verilogrepair_peft/README.md

library_name, base_model

library_name	base_model
peft	shakechen/Llama-2-7b-hf

ChipGPT-FT-Repair

This repository contains the code for the fine-tuning of the ChipGPT model using the PEFT library. The fine-tuning process was performed using the bitsandbytes quantization method.

The foundation model is Llama 2.0-7B

Training procedure

The following bitsandbytes quantization config was used during training:

• quant_method: bitsandbytes
• load_in_8bit: True
• load_in_4bit: False
• 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 0.7.2.dev0

• PEFT 0.6.0.dev0