SCOPD / SCOPD+ model release The weights derive from Qwen/Qwen2.5-VL-7B-Instruct, revision cc594898137f460bfe9f0759e9844b3ce807cfb5, and Qwen/Qwen3-VL-4B-Instruct, revision ebb281ec70b05090aa6165b016eac8ec08e71b17. Both model cards specify Apache-2.0. Only fine-tuning LoRA adapter weights are redistributed; base weights are not included. Evaluation and pruning sources include local patches recorded in evaluation_files.json. Transformers: see evaluation/runtime/TRANSFORMERS_LICENSE. VLMEvalKit: see evaluation/third_party/VLMEvalKit/LICENSE. VisionZip: see evaluation/third_party/VisionZip/LICENSE. Retained third-party source headers and licenses remain applicable.