|
--- |
|
library_name: peft |
|
license: llama2 |
|
inference: true |
|
datasets: |
|
- Enno-Ai/fr-instructs |
|
language: |
|
- fr |
|
- en |
|
tags: |
|
- code |
|
- peft |
|
- llama |
|
- llama2 |
|
- codellama |
|
pipeline_tag: text-generation |
|
--- |
|
## ARIA CODE FR Instruct is a finetuned model based on LLAMA CODE 34B INSTRUCT |
|
|
|
This model has been trained over 10 Millions tokens on a french language dataset structured with Alpaca style. |
|
|
|
GPU used for training : NVIDIA A100 |
|
|
|
Timing of training: 24H |
|
## 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 0.6.0.dev0 |
