Nalandadata

Live Demos

Indian STEM reasoning models — solver, table recognition, and multimodal QA. GPU inference runs serverless on Modal; the first call after idle wakes the GPU (~1–2 min), then it's fast.

JEE/NEET Solver

Nalanda Qwen 7B (GRPO / RLVR-tuned) solves a physics, chemistry, or maths problem with a worked solution.

Table Structure Recognition

DrishtiTable (Qwen2.5-VL-7B + LoRA) converts a table image into clean HTML.

Click to upload a table image

Multimodal Image QA

nalanda-image-vl answers a question about a scientific diagram or figure.

Click to upload a diagram / figure

Note: this model builds on Meta's gated Llama-3.2-Vision. If access is still pending on the backend token, this tab will return an error until it's granted.

DrishtiTable — Table Structure Recognition (TEDS)

Image → HTML table recognition on 135 held-out Indian-textbook tables. Ranked by TEDS (the field-standard TSR metric). Our fine-tuned 7B model beats every zero-shot frontier model.

ModelMethodTEDSS-TEDS
DrishtiTable-Qwen2.5-VL-7BFine-tuned (SFT)83.2%89.7%
Claude Sonnet 4.6Zero-shot77.3%89.2%
Claude Opus 4.8Zero-shot75.5%88.2%
GPT-4oZero-shot71.1%84.3%
GPT-5.1Zero-shot69.9%83.3%
Gemini 3.1 ProZero-shot65.7%73.6%
Fine-tuned 7B vs best zero-shot+5.9 pp

135-table held-out test set, frozen v1.0 (SHA-256), TEDS scored against private ground truth. Full 10-model board, methodology, and submission guide: live leaderboard · benchmark dataset.

NalandaBench — JEE/NEET MCQ accuracy

Held-out multiple-choice questions. GRPO / RLVR post-training vs the open base model.

ModelMethodAccuracy
Nalanda Qwen 7BGRPO / RLVR66.8%
Qwen2.5-7B-InstructBase (no RLVR)60.5%
Improvement from RLVR post-training+6.3 pp

Evaluated on 800 held-out MCQs from NalandaJEENEETBench. Figures are indicative and are refreshed as the benchmark split is finalized.