Edit model card

結合兩個數據庫來做微調模型來達到知識問答和聊天的機器人

  • wikimedia/wikipedia
  • stingning/ultrachat

1.效率:透過使用GPU加速、LoRA、梯度累積和混合精度訓練(FP16),最大化運算資源和訓練速度。

2.適應性:透過LoRA對模型的特定組件進行微調,它可以以減少參數達到(30%)以更新更有效地適應目標任務的預訓練模型。

api使用方法:

from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("j40pl7lly/fine-tuning-chat-liu")
model = AutoModelForCausalLM.from_pretrained("j40pl7lly/fine-tuning-chat-liu")

Reference

If you use this model and love it, use this to cite it 🤗

Citation

@misc{privacy_faceemotionrecognition_system,
      title={Fine-tuned LLM model based on open source mistral-7B},
      author={Liu Hsin Kuo},
      year={2024},
}
Downloads last month
4
Safetensors
Model size
7.24B params
Tensor type
BF16
·
Inference Examples
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social visibility and check back later, or deploy to Inference Endpoints (dedicated) instead.

Datasets used to train j40pl7lly/fine-tuning-chat-liu