|
--- |
|
library_name: peft |
|
base_model: scb10x/typhoon-7b |
|
datasets: |
|
- Thaweewat/alpaca-cleaned-52k-th |
|
language: |
|
- en |
|
- th |
|
--- |
|
|
|
|
|
|
|
## Model Description |
|
[Typhoon-7b](scb10x/typhoon-7b) QLoRA Finetune by unsloth with [Thai Alpaca](Thaweewat/alpaca-cleaned-52k-th) dataset. |
|
|
|
#### Training Hyperparameters |
|
The following `bitsandbytes` quantization config was used during training: |
|
- r = 64 |
|
- lora_alpha = 16 |
|
- lora_dropout = 0.05 |
|
- 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 |
|
|
|
### Loss |
|
|
|
Step Training Loss |
|
2500 1.241900 |
|
5000 1.123600 |
|
7500 1.014600 |
|
10000 0.902200 |
|
12500 0.906500 |
|
15000 0.683900 |
|
17500 0.650900 |
|
20000 0.584800 |
|
22500 0.385100 |
|
25000 0.384100 |
|
|
|
### Framework versions |
|
|
|
- PEFT 0.7.0 |