PVIS2027-JT-6597/LLaVA-OneVision-2-8B-SciVisCap

LLaVA-OneVision-2-8B-SciVisCap

This repository contains a LoRA adapter for lmms-lab-encoder/LLaVA-OneVision-2-8B-Instruct, trained on SciVisCap for scientific visualization (SciVis) figure captioning.

Training Configuration

• —Epochs: 3 (final checkpoint)
• —LoRA rank: 8
• —LoRA alpha: 32
• —Batch size: 8
• —Learning rate: 1e-4 (cosine schedule, warmup ratio 0.1)
• —Max sequence length: 4,096 tokens