Indic-gemma-7b-finetuned-sft-Navarasa-2.0
| Property | Value |
|---|---|
| Base Model | google/gemma-7b |
| Training Data | 650K instruction samples |
| Languages Supported | 15 Indian languages + English |
| Training Infrastructure | 1x A100 80GB GPU |
| Training Duration | 45 Hours |
| Model Hub | Hugging Face |
What is Indic-gemma-7b-finetuned-sft-Navarasa-2.0?
This is a multilingual language model based on Google's Gemma-7B architecture, specifically fine-tuned for Indian languages using LoRA techniques. The model supports 15 Indian languages including Hindi, Telugu, Tamil, Malayalam, and more, along with English, making it a comprehensive solution for Indian language processing tasks.
Implementation Details
The model leverages the unsloth library for optimization and offers two inference methods: using unsloth for 2x faster inference or traditional HuggingFace implementation. It was trained on high-quality instruction datasets from various sources for each supported language, ensuring robust multilingual capabilities.
- Fine-tuned using LoRA on carefully curated instruction datasets
- Supports both unsloth and HuggingFace inference pipelines
- Optimized for production deployment with flexible dtype support
- Maximum sequence length of 2048 tokens
Core Capabilities
- Multilingual instruction following across 15+ languages
- High-performance inference with unsloth optimization
- Support for both 4-bit quantization and full precision inference
- Comprehensive coverage of Indian language processing tasks
Frequently Asked Questions
Q: What makes this model unique?
The model's primary strength lies in its comprehensive coverage of Indian languages and optimization for practical deployment using the unsloth library. It's trained on a diverse set of high-quality instruction datasets, making it particularly effective for Indian language processing tasks.
Q: What are the recommended use cases?
The model is ideal for multilingual applications requiring understanding and generation across Indian languages, including translation, content generation, and instruction following. It's particularly suitable for applications requiring efficient inference through unsloth optimization.
