ShivRamSaud/gemma4-e4b-nepali-denoise-merged
ShivRamSaud/gemma4-e4b-nepali-denoise-merged is a 7.9 billion parameter Gemma 4 E4B-it model, fine-tuned by ShivRamSaud, specifically for Nepali text denoising tasks. This model leverages a 32768 token context length and is optimized to correct and clean noisy Nepali text. Its primary application is in improving the quality of Nepali language data through specialized denoising capabilities.
Loading preview...
Model Overview
ShivRamSaud/gemma4-e4b-nepali-denoise-merged is a specialized fine-tuned version of the google/gemma-4-E4B-it model, developed by ShivRamSaud. This 7.9 billion parameter model is designed for Nepali text denoising, focusing on correcting and cleaning noisy Nepali language inputs. It operates with a substantial context length of 32768 tokens.
Key Capabilities
- Nepali Text Denoising: The model is explicitly fine-tuned on a multi-task Nepali corpus to excel at identifying and correcting errors in Nepali text.
- Standalone Merged FP16: Provided as a fully merged FP16 weight model, suitable for direct deployment and inference.
Benchmark Performance
Benchmarked against a base Gemma 4 model on 5,450 test samples, this fine-tuned model shows significant changes in performance metrics, particularly in character error rate (CER) reduction, indicating its specialized focus on denoising.
Use Cases
- Data Preprocessing: Ideal for cleaning user-generated content, OCR outputs, or other noisy Nepali text data before further processing.
- Language Quality Improvement: Can be integrated into applications requiring high-quality Nepali text, such as translation systems or content moderation tools.
Usage
The model can be easily integrated using the transformers library, with a clear example provided for text denoising tasks, where it's instructed to output only the clean/corrected text.