---
license: other
tags: [medical-imaging, chest-xray, visual-question-answering, diff-vqa]
---

# LAP-VQA — GPT-2 Medium (visual prefix)

Best checkpoint from Diff-VQA online multi-objective (reg-steps) training.

**Weights only** — this file contains `state_dict` + architecture config
(`vis_dim`, `d_model`, `num_layers`, ...). It does **not** contain optimizer or
LR-scheduler state, so it cannot be used to exactly resume an interrupted
training run (fine for eval / deployment / fine-tuning from weights).

⚠️ Gated: medical-imaging model weights, access granted on request.