Turkish Finance Summarizer

This model is a Turkish financial news abstractive summarization model trained with a custom tokenizer and a custom training pipeline.

Model Architecture

  • Transformer encoder-decoder
  • mBART-style architecture
  • Trained from scratch
  • No weights initialized from Meta/Facebook mBART checkpoints

Model Details

  • Language: Turkish
  • Task: Abstractive Summarization
  • Domain: Finance / Economy / Markets
  • Training: Custom pretraining + fine-tuning

Usage

Load the model and tokenizer using Hugging Face Transformers and use generate() for abstractive summarization.

Notes

  • Optimized for Turkish financial news
  • Suitable for research and production inference
  • Scratch-trained with domain-specific data
Downloads last month
26
Safetensors
Model size
71.3M params
Tensor type
F32
·
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support