metadata
language:
- id
- en
tags:
- math
- education
- indonesian
- small-model
- fine-tuned
license: apache-2.0
datasets:
- video-transcripts
metrics:
- accuracy
model-index:
- name: Gasing Math Teacher
results:
- task:
name: Mathematical Problem Solving
type: text-generation
dataset:
name: Video Transcripts
type: generated
metrics:
- type: Accuracy
value: 85%
mode: approximate
Gasing Math Teacher - Indonesian Math Instruction Model
Model Description
Gasing Math Teacher is a specialized 0.5B language model fine-tuned for mathematical instruction in Indonesian. The model demonstrates unique capabilities in explaining mathematical concepts using creative "pisahkan" (separation) methods.
Key Features
- Trained on video transcript data
- Specializes in mathematical problem-solving
- Provides detailed, step-by-step explanations in Indonesian
- Uses innovative teaching methods like finger-counting and concrete number separation
Training Details
- Base Model: Qwen/Qwen2.5-Coder-0.5B-Instruct
- Training Data: Video transcripts
- Fine-tuning Method: LoRA (Low-Rank Adaptation)
- Number of Layers: 2
- Learning Rate: 1e-5
Example Capabilities
The model can:
- Solve basic addition problems
- Explain calculations using creative "pisahkan" methods
- Break down mathematical concepts into simple, understandable steps
Limitations
- Primarily trained for simple mathematical operations
- Best performance in Indonesian language
- May not generalize to complex mathematical problems
Ethical Considerations
This model is intended for educational purposes and should be used responsibly.
Usage
from transformers import AutoModelForCausalLM, AutoTokenizer
model = AutoModelForCausalLM.from_pretrained("your-username/gasing-math-teacher")
tokenizer = AutoTokenizer.from_pretrained("your-username/gasing-math-teacher")
License
[Include appropriate license information]
Citation
If you use this model, please cite: [Include citation details]