Edit model card

Model Card for Model ID

Model Details

Model Description

  • Language(s) (NLP): Khmer
  • Finetuned from model: Qwen/Qwen2-1.5B-Instruct

Usage Notebook on Kaggle

https://www.kaggle.com/code/mlao01/km-typing-assist

Run Validation Set Notebook on Kaggle

https://www.kaggle.com/code/mlao01/benchmark

Bias, Risks, and Limitations

Be cautious, we did not proof read the dataset used to tune this model.

Training Data

mlao01/km-news

Training Procedure

QLoRa for 20% of the training data:

  • Rank = 128
  • Learing rate = 2e-4
  • Warmup = 25% of the training data
  • Gradient Clipping = 5.0

Number of Parameters

trainable params: 147,718,144 || all params: 1,691,432,448 || trainable%: 8.7333

Evaluation

Last logged training loos reached ~0.5 cross entropy

Downloads last month
8
Safetensors
Model size
909M params
Tensor type
F32
FP16
U8
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.