LICENSES.md

# Upstream data and model terms

This repository assembles precomputed representations and sample metadata from multiple sources. The artifacts do not have a single newly assigned blanket license. The applicable upstream dataset and encoder terms continue to govern reuse.

| Component | Source and terms to consult |
| --- | --- |
| LCS-558K sample captions and image identifiers | [LLaVA-Pretrain dataset card](https://huggingface.co/datasets/liuhaotian/LLaVA-Pretrain) and [LLaVA project](https://github.com/haotian-liu/LLaVA) |
| ImageNet-1K sample labels, identifiers, and derived features | [ImageNet access and source terms](https://www.image-net.org/download.php) and [official ImageNet-1K dataset card](https://huggingface.co/datasets/ILSVRC/imagenet-1k) |
| Visual features | Source-data terms and the individual encoder/model terms linked in [encoders.json](encoders.json) and the [Model Zoo](https://huggingface.co/336labs/VisionEncoder-to-MLLM-ModelZoo) |
| Text features | Source-data terms and [Qwen2.5-1.5B-Instruct](https://huggingface.co/Qwen/Qwen2.5-1.5B-Instruct), [Qwen3-1.7B](https://huggingface.co/Qwen/Qwen3-1.7B), and [SmolLM2-1.7B-Instruct](https://huggingface.co/HuggingFaceTB/SmolLM2-1.7B-Instruct) model cards |
| Evaluation software | [MLLM-VisionEncoder-Eval source repository](https://github.com/JuntaoTang/MLLM-VisionEncoder-Eval) and its applicable license |

Raw source images and encoder weights are obtained from their respective upstream releases. Listing upstream download links does not override access conditions or grant new redistribution rights. Cite the paper, source datasets, and the encoders used in your work.