How ganesh-dark/tamil-news-classifier works
This repository presents a Tamil-English, multi-label news classifier built by fine-tuning Qwen 2.5 7B. The documented user workflow sends a headline or article to the quantized model through Ollama and receives one or more topic labels. A separate training workflow uses the labeled TSV dataset and Colab notebook to fine-tune and export model weights. The Python package currently exposes only a placeholder greeting, so it is not part of the documented inference path. Training and inference links are based on README descriptions; the model and Modelfile are not present in the tree.
Main entry points: train_news_unsloth_colab.ipynb Colab fine-tuning workflow · Training_news_dataset.tsv Labeled news dataset · README.md 83-topic taxonomy · __init__.py Python package entry
Components 10
Classification inference
- Ollama runtime
- Fine-tuned Qwen model
- 83-topic taxonomy
README.md - Predicted topic labels
Model training
- Labeled news dataset
Training_news_dataset.tsv - Colab fine-tuning workflow
train_news_unsloth_colab.ipynb - Qwen base model
- Unsloth framework
Python package
- Python package entry
src/tamil_news_classifier/__init__.py
Other components
- News user
Connections 9
- News user → Ollama runtime: submits headline
README.md - Ollama runtime → Fine-tuned Qwen model: runs inference
README.md - Fine-tuned Qwen model → Predicted topic labels: generates labels
README.md - Fine-tuned Qwen model → 83-topic taxonomy: classifies into
README.md - Labeled news dataset → Colab fine-tuning workflow: provides examples
README.md - Qwen base model → Colab fine-tuning workflow: provides starting model
README.md - Unsloth framework → Colab fine-tuning workflow: supports fine-tuning
README.md - Colab fine-tuning workflow → Fine-tuned Qwen model: exports quantized model
README.md - News user → Python package entry: invokes CLI