Views
No views yet
| Parameter / Metric | Value / Setting |
|---|---|
| Base Model | Qwen/Qwen2.5-7B-Instruct (7.61 Billion Parameters) |
| Language Target | 100% Pure English (Zero Sanskrit shlokas or transliteration leakage) |
| Fine-Tuning Method | Supervised Fine-Tuning (SFT) + PEFT LoRA ($r=16, \alpha=32$) |
| Hardware | 2× NVIDIA H200 NVL GPUs (288 GB VRAM) |
| Dataset Mix | ~200,000 Pure English Instruction Records (JDhruv14/Bhagavad-Gita-QA, mahabharata_great_india_epic, Ramayana English, Mahabharata English) |
| Entity Memory Retention | 100% Coherent Character Lineage & Factual Accuracy |
rājña, vīryeṣu), preventing dataset template leakage.Qwen2.5-7B-Instruct ($d_{model}=3584$) possesses the capacity to retain complex character lineages (e.g. Yudhishthira, Karna, Lakshmana, Abhimanyu) without mixing entities across epics.adapter/: PEFT LoRA adapter model weights (adapter_model.safetensors, adapter_config.json, chat_template.jinja).QWEN-7B_FINE_TUNING_FOR_RAMAYANA/: Complete codebase, English dataset consolidator, evaluation scripts, docs, and metrics.Foundational_Model_Scratch_Training_on_Ramayana_Mahabharata.pptx: 10-slide widescreen executive PowerPoint deck.