20 lines
576 B
Markdown
20 lines
576 B
Markdown
---
|
|
license: apache-2.0
|
|
language:
|
|
- fr
|
|
pipeline_tag: text-generation
|
|
---
|
|
|
|
- Model : https://huggingface.co/GeneZC/MiniChat-1.5-3B
|
|
- FT : @teilomillet
|
|
|
|
- Instruction tune using QLoRA on a french dataset for 1 epoch.
|
|
The aim was to test and try the dataset. Implementing a customization via a dataset and fine-tuning on it.
|
|
The way to respond is also important to see if it's taken from the dataset and add to the customization.
|
|
|
|
This is the first of a long serie of multiple models. Aimed to be minuscule as possible.
|
|
|
|
- Batch : 6
|
|
- Gradient step : 1
|
|
- Epoch : 1
|
|
- Lr : 0.0002 |