|
--- |
|
library_name: peft |
|
--- |
|
## Training procedure |
|
|
|
We decided to release an ARIA 7B model trained with mistral 7B instruct as base model. We adressed the language challenge with a dataset focused on french language. |
|
|
|
The finetuning has been done with Nvidia GPUs. |
|
|
|
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 0.6.0.dev0 |
|
|
