|
--- |
|
language: |
|
- ko |
|
pipeline_tag: text-generation |
|
tags: |
|
- finetune |
|
--- |
|
# Model Card for mistral-ko-7b-wiki-neft |
|
It is a fine-tuned model using Korean and NEFT in the mistral-7b model. |
|
|
|
## Model Details |
|
|
|
* **Model Developers** : shleeeee(Seunghyeon Lee) |
|
* **Repository** : To be added |
|
* **Model Architecture** : The mistral-ko-7b-wiki-neft is is a fine-tuned version of the Mistral-7B-v0.1. |
|
* **Lora target modules** : q_proj, k_proj, v_proj, o_proj,gate_proj |
|
* **train_batch** : 4 |
|
* **neftune_noise_alpha** : 5 |
|
* **Max_step** : 1000 |
|
|
|
## Dataset |
|
Korean Custom Dataset |
|
|
|
## Prompt template: Mistral |
|
``` |
|
<s>[INST]{['instruction']}[/INST]{['output']}</s> |
|
``` |
|
|
|
## Usage |
|
``` |
|
# Load model directly |
|
from transformers import AutoTokenizer, AutoModelForCausalLM |
|
|
|
tokenizer = AutoTokenizer.from_pretrained("shleeeee/mistral-7b-wiki") |
|
model = AutoModelForCausalLM.from_pretrained("shleeeee/mistral-7b-wiki") |
|
|
|
# Use a pipeline as a high-level helper |
|
from transformers import pipeline |
|
|
|
pipe = pipeline("text-generation", model="shleeeee/mistral-7b-wiki") |
|
``` |
|
|
|
## Evaluation |
|
 |
