shinigamiRaj/IndicVedas
shinigamiRaj/IndicVedas is a specialized 14 billion parameter large language model based on Qwen2.5-14B-Instruct, continuously pre-trained and fine-tuned on a comprehensive corpus of ancient Indian scriptures including the four Vedas and foundational Ayurvedic texts. Optimized for deep understanding and scholarly analysis, this model excels at interactive exploration of classical Indian texts. It supports a context length of 4096 tokens during training and up to 16,384 during inference, making it suitable for detailed textual analysis.
Loading preview...
VedaGPT (IndicVedas): Specialized LLM for Ancient Indian Texts
VedaGPT (IndicVedas) is a 14 billion parameter language model, built upon Qwen2.5-14B-Instruct, specifically designed for scholarly analysis and interactive exploration of ancient Indian scriptures. It has undergone continuous pre-training and fine-tuning on a custom-scraped corpus.
Key Capabilities & Training:
- Specialized Knowledge: Deep understanding of the Rig Veda, Sama Veda, Yajur Veda, Atharva Veda, and foundational Ayurvedic texts like Charaka Samhita, Sushruta Samhita, and Rasa Jala Nidhi.
- Comprehensive Corpus: Trained on complete texts including Griffith Translations and bilingual Sanskrit/Hindi texts for Vedas, and structural chapters for Ayurvedic works.
- Optimized Fine-Tuning: Utilizes Unsloth for efficient fine-tuning, employing an aggressive knowledge overriding strategy over 2.0 epochs to ensure deep familiarity with the subject matter.
- Context Length: Trained with a context length of 4096 tokens, supporting up to 16,384 tokens during inference.
- Deployment Flexibility: Available in merged 16-bit
bfloat16for cloud/high-end GPU and optimized GGUF (Q4_K_M) quantizations for local CPU/GPU deployment (e.g., Ollama, llama.cpp).
Ideal Use Cases:
- Scholarly Research: Analyzing and interpreting ancient Indian scriptures and classical texts.
- Educational Tools: Developing interactive platforms for learning about Vedic and Ayurvedic knowledge.
- Content Generation: Creating content that requires deep contextual understanding of these specific texts.
- Local Deployment: Running on consumer hardware via GGUF quantizations for accessibility.