File size: 961 Bytes
442eca3 68e912e 0329018 68e912e 7cd8664 68e912e 7cd8664 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 |
---
library_name: peft
---
⚠️⚠️⚠️ Only for research purpose.
Do not use it for medical purpose. ⚠️⚠️⚠️
## 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: bfloat16
### Framework versions
- PEFT 0.4.0
### How to cite
本データを利用する場合は以下の文献の引用をご検討ください.
```
@article{sukeda2023jmedlora,
title={{JMedLoRA: Medical Domain Adaptation on Japanese Large Language Models using Instruction-tuning}},
author={Sukeda, Issey and Suzuki, Masahiro and Sakaji, Hiroki and Kodera, Satoshi},
journal={arXiv preprint arXiv:2310.10083},
year={2023}
}
``` |