This is a Finetuning of GPT-J-6B using LoRa - https://huggingface.co/EleutherAI/gpt-j-6B
The dataset is the cleaned version of the Alpaca dataset - https://github.com/gururise/AlpacaDataCleaned
A model similar to this has been talked about
The performance is good but not as good as the orginal Alpaca trained from a base model of LLaMa
This is mostly due to the LLaMa 7B model being pretrained on 1T tokens and GPT-J-6B being trained on 300-400M tokens
You will need a 3090 or A100 to run it, unfortunately this current version won't work on a T4.
library_name: peft license: apache-2.0 language: - en tags: - Text Generation
Training procedure
The following bitsandbytes quantization config was used during training:
- 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
The following bitsandbytes quantization config was used during training:
- 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
The following bitsandbytes quantization config was used during training:
- 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 0.4.0.dev0
PEFT 0.4.0.dev0
PEFT 0.4.0.dev0