| license: gemma | |
| base_model: google/gemma-2-2b-it | |
| tags: | |
| - gemma2 | |
| - instruction-tuning | |
| - nirf | |
| - india | |
| pipeline_tag: text-generation | |
| # Gemma-2B (IT) — NIRF Lookup 2025 (Merged FP16) | |
| Base: google/gemma-2-2b-it | |
| This repository contains merged full weights (LoRA baked into base). | |
| Intended use: | |
| Short factual lookup answers about NIRF 2025 (Indian institutes). | |
| How to use (summary): | |
| Load with Transformers AutoTokenizer and AutoModelForCausalLM from this repo id. | |
| Use bfloat16 on NVIDIA L4. Provide an instruction (and optional context), then generate. | |
| Training summary: | |
| QLoRA (4-bit) on Gemma-2-2b-it. LoRA r=16, alpha=64, dropout=0.1. | |
| Target modules: q_proj, k_proj, v_proj, o_proj, gate_proj, up_proj, down_proj. | |
| bf16 on NVIDIA L4. Data: 100 NIRF 2025 lookup samples. | |
| License & notice: | |
| This model is a Model Derivative of google/gemma-2-2b-it and is distributed under Google’s Gemma Terms of Use. | |
| See the NOTICE file in this repo. | |