| --- |
| pretty_name: Vision Encoder Evaluation Reproduction Data |
| language: |
| - en |
| task_categories: |
| - image-classification |
| - other |
| license: other |
| license_name: upstream-data-and-model-terms |
| license_link: https://huggingface.co/datasets/336labs/VisionEncoder-Eval-ReproData/blob/main/LICENSES.md |
| size_categories: |
| - 100K<n<1M |
| source_datasets: |
| - liuhaotian/LLaVA-Pretrain |
| - ILSVRC/imagenet-1k |
| tags: |
| - vision-encoder-evaluation |
| - multimodal |
| - feature-extraction |
| - ravel |
| - numpy |
| - arxiv:2610.05413 |
| configs: |
| - config_name: lcs-1k |
| default: true |
| data_files: |
| - split: train |
| path: lcs-1k/samples/samples.jsonl |
| - config_name: lcs-10k |
| data_files: |
| - split: train |
| path: lcs-10k/samples/samples.jsonl |
| - config_name: imagenet-200k |
| data_files: |
| - split: train |
| path: imagenet-200k/samples/samples.jsonl |
| --- |
| |
| <h1 align="center">A Strong Baseline for Evaluating Vision Encoders<br>in Multimodal Large Language Models</h1> |
|
|
| <p align="center"> |
| Yilin Yang<sup>1,*</sup> · |
| Jun-Tao Tang<sup>2,*</sup> · |
| Kengyi Wang<sup>3</sup> · |
| Siyuan Su<sup>3</sup> · |
| Gaoyong Luo<sup>4</sup> · |
| Mingda Chen<sup>1,†</sup> |
| </p> |
|
|
| <p align="center"> |
| <sup>1</sup>School of Artificial Intelligence, Shanghai Jiao Tong University<br> |
| <sup>2</sup>Nanjing University · |
| <sup>3</sup>Fudan University · |
| <sup>4</sup>Independent Researcher<br> |
| <sup>*</sup>Equal contribution. <sup>†</sup>Corresponding author. |
| </p> |
| |
| <p align="center"> |
| <a href="https://arxiv.org/abs/2610.05413"><img src="https://img.shields.io/badge/arXiv-2610.05413-b31b1b" alt="Paper: arXiv 2610.05413"></a> |
| <a href="https://arxiv.org/pdf/2610.05413"><img src="https://img.shields.io/badge/Paper-PDF-red" alt="Paper PDF"></a> |
| <a href="https://github.com/JuntaoTang/MLLM-VisionEncoder-Eval"><img src="https://img.shields.io/badge/GitHub-Code-181717?logo=github" alt="GitHub code"></a> |
| <a href="https://huggingface.co/336labs/VisionEncoder-to-MLLM-ModelZoo/tree/main"><img src="https://img.shields.io/badge/Model_Zoo-Checkpoints-FFD21E" alt="Model zoo checkpoints"></a> |
| <a href="#citation"><img src="https://img.shields.io/badge/Citation-BibTeX-blue" alt="BibTeX citation"></a> |
| </p> |
| |
| ## Dataset Summary |
| |
| Reproduction data, including precomputed **vision and text features**, for the evaluations in [A Strong Baseline for Evaluating Vision Encoders in Multimodal Large Language Models](https://arxiv.org/abs/2610.05413). The release covers **70 vision encoders** (43 language-supervised, 22 self-supervised, and 5 discrete), text features from **three language models**, and shared sample metadata. |
| |
| ## Subsets |
| |
| | Subset | Samples | Contents | Size | |
| |---|---:|---|---:| |
| | [`lcs-1k`](https://huggingface.co/datasets/336labs/VisionEncoder-Eval-ReproData/tree/main/lcs-1k) | 1,000 | LCS-558K image patch features, text features, and captions | 55.3 GiB | |
| | [`lcs-10k`](https://huggingface.co/datasets/336labs/VisionEncoder-Eval-ReproData/tree/main/lcs-10k) | 10,000 | LCS-558K image patch features, text features, and captions | 553.0 GiB | |
| | [`imagenet-200k`](https://huggingface.co/datasets/336labs/VisionEncoder-Eval-ReproData/tree/main/imagenet-200k) | 200,000 | ImageNet-1K pooled features, class labels, and kNN split | 59.9 GiB | |
| |
| Each subset includes features for all 70 vision encoders. The LCS subsets also include text features from [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). The 10k arrays are stored in two 5k blocks per encoder/model. |
| |
| Keep the shared sample row order when loading features. Shapes, extraction settings, and sampling details are recorded in the per-file metadata and sample manifests. |
| |
| |
| ## Encoder Catalog |
| |
| [encoders.json](encoders.json) lists feature paths, shapes, metadata, and upstream encoder sources for all 70 encoders. |
| |
| | Tokenizer ID | Family | LCS `[T, D]` | ImageNet `D` | Feature files | |
| | --- | --- | --- | ---: | --- | |
| | [`toklip_l_384`](https://huggingface.co/TencentARC/TokLIP) | DISCRETE | `[576, 1152]` | 1152 | [patches](lcs-1k/vision/toklip_l_384_patch_n1000_seed42.npy) · [pooled](imagenet-200k/features/011_toklip_l_384.npy) · [10k blocks](https://huggingface.co/datasets/336labs/VisionEncoder-Eval-ReproData/tree/main/lcs-10k/vision/toklip_l_384) | |
| | [`toklip_s_256`](https://huggingface.co/TencentARC/TokLIP) | DISCRETE | `[256, 1152]` | 1152 | [patches](lcs-1k/vision/toklip_s_256_patch_n1000_seed42.npy) · [pooled](imagenet-200k/features/012_toklip_s_256.npy) · [10k blocks](https://huggingface.co/datasets/336labs/VisionEncoder-Eval-ReproData/tree/main/lcs-10k/vision/toklip_s_256) | |
| | [`uniar_bsq`](https://huggingface.co/ShareLab-SII/UniAR-SFT/tree/main/bsq_encoder) | DISCRETE | `[1024, 4608]` | 4096 | [patches](lcs-1k/vision/uniar_bsq_patch_n1000_seed42.npy) · [pooled](imagenet-200k/features/013_uniar_bsq.npy) · [10k blocks](https://huggingface.co/datasets/336labs/VisionEncoder-Eval-ReproData/tree/main/lcs-10k/vision/uniar_bsq) | |
| | [`unitok_attn`](https://huggingface.co/FoundationVision/unitok_tokenizer) | DISCRETE | `[256, 1024]` | 1024 | [patches](lcs-1k/vision/unitok_attn_patch_n1000_seed42.npy) · [pooled](imagenet-200k/features/014_unitok_attn.npy) · [10k blocks](https://huggingface.co/datasets/336labs/VisionEncoder-Eval-ReproData/tree/main/lcs-10k/vision/unitok_attn) | |
| | [`vilau_256`](https://huggingface.co/mit-han-lab/vila-u-7b-256) | DISCRETE | `[256, 1024]` | 1024 | [patches](lcs-1k/vision/vilau_256_patch_n1000_seed42.npy) · [pooled](imagenet-200k/features/015_vilau_256.npy) · [10k blocks](https://huggingface.co/datasets/336labs/VisionEncoder-Eval-ReproData/tree/main/lcs-10k/vision/vilau_256) | |
| | [`clip_openai__l14`](https://huggingface.co/openai/clip-vit-large-patch14) | LANG | `[256, 1024]` | 1024 | [patches](lcs-1k/vision/clip_openai__l14_patch_n1000_seed42.npy) · [pooled](imagenet-200k/features/001_clip_openai__l14.npy) · [10k blocks](https://huggingface.co/datasets/336labs/VisionEncoder-Eval-ReproData/tree/main/lcs-10k/vision/clip_openai__l14) | |
| | [`mc1_b16_224_2.5b`](https://github.com/facebookresearch/MetaCLIP#pre-trained-models) | LANG | `[196, 768]` | 768 | [patches](lcs-1k/vision/mc1_b16_224_2.5b_patch_n1000_seed42.npy) · [pooled](imagenet-200k/features/021_mc1_b16_224_2.5b.npy) · [10k blocks](https://huggingface.co/datasets/336labs/VisionEncoder-Eval-ReproData/tree/main/lcs-10k/vision/mc1_b16_224_2.5b) | |
| | [`mc1_b16_224_400m`](https://github.com/facebookresearch/MetaCLIP#pre-trained-models) | LANG | `[196, 768]` | 768 | [patches](lcs-1k/vision/mc1_b16_224_400m_patch_n1000_seed42.npy) · [pooled](imagenet-200k/features/022_mc1_b16_224_400m.npy) · [10k blocks](https://huggingface.co/datasets/336labs/VisionEncoder-Eval-ReproData/tree/main/lcs-10k/vision/mc1_b16_224_400m) | |
| | [`mc1_b32_224_2.5b`](https://github.com/facebookresearch/MetaCLIP#pre-trained-models) | LANG | `[49, 768]` | 768 | [patches](lcs-1k/vision/mc1_b32_224_2.5b_patch_n1000_seed42.npy) · [pooled](imagenet-200k/features/023_mc1_b32_224_2.5b.npy) · [10k blocks](https://huggingface.co/datasets/336labs/VisionEncoder-Eval-ReproData/tree/main/lcs-10k/vision/mc1_b32_224_2.5b) | |
| | [`mc1_b32_224_400m`](https://github.com/facebookresearch/MetaCLIP#pre-trained-models) | LANG | `[49, 768]` | 768 | [patches](lcs-1k/vision/mc1_b32_224_400m_patch_n1000_seed42.npy) · [pooled](imagenet-200k/features/024_mc1_b32_224_400m.npy) · [10k blocks](https://huggingface.co/datasets/336labs/VisionEncoder-Eval-ReproData/tree/main/lcs-10k/vision/mc1_b32_224_400m) | |
| | [`mc1_g14_224_2.5b`](https://github.com/facebookresearch/MetaCLIP#pre-trained-models) | LANG | `[256, 1664]` | 1664 | [patches](lcs-1k/vision/mc1_g14_224_2.5b_patch_n1000_seed42.npy) · [pooled](imagenet-200k/features/025_mc1_g14_224_2.5b.npy) · [10k blocks](https://huggingface.co/datasets/336labs/VisionEncoder-Eval-ReproData/tree/main/lcs-10k/vision/mc1_g14_224_2.5b) | |
| | [`mc1_h14_224_2.5b`](https://github.com/facebookresearch/MetaCLIP#pre-trained-models) | LANG | `[256, 1280]` | 1280 | [patches](lcs-1k/vision/mc1_h14_224_2.5b_patch_n1000_seed42.npy) · [pooled](imagenet-200k/features/026_mc1_h14_224_2.5b.npy) · [10k blocks](https://huggingface.co/datasets/336labs/VisionEncoder-Eval-ReproData/tree/main/lcs-10k/vision/mc1_h14_224_2.5b) | |
| | [`mc1_h14_224_v1.2`](https://github.com/facebookresearch/MetaCLIP#pre-trained-models) | LANG | `[256, 1280]` | 1280 | [patches](lcs-1k/vision/mc1_h14_224_v1.2_patch_n1000_seed42.npy) · [pooled](imagenet-200k/features/027_mc1_h14_224_v1.2.npy) · [10k blocks](https://huggingface.co/datasets/336labs/VisionEncoder-Eval-ReproData/tree/main/lcs-10k/vision/mc1_h14_224_v1.2) | |
| | [`mc1_l14_224_2.5b`](https://github.com/facebookresearch/MetaCLIP#pre-trained-models) | LANG | `[256, 1024]` | 1024 | [patches](lcs-1k/vision/mc1_l14_224_2.5b_patch_n1000_seed42.npy) · [pooled](imagenet-200k/features/028_mc1_l14_224_2.5b.npy) · [10k blocks](https://huggingface.co/datasets/336labs/VisionEncoder-Eval-ReproData/tree/main/lcs-10k/vision/mc1_l14_224_2.5b) | |
| | [`mc1_l14_224_400m`](https://github.com/facebookresearch/MetaCLIP#pre-trained-models) | LANG | `[256, 1024]` | 1024 | [patches](lcs-1k/vision/mc1_l14_224_400m_patch_n1000_seed42.npy) · [pooled](imagenet-200k/features/029_mc1_l14_224_400m.npy) · [10k blocks](https://huggingface.co/datasets/336labs/VisionEncoder-Eval-ReproData/tree/main/lcs-10k/vision/mc1_l14_224_400m) | |
| | [`mc2_b16_224`](https://github.com/facebookresearch/MetaCLIP#pre-trained-models) | LANG | `[196, 768]` | 768 | [patches](lcs-1k/vision/mc2_b16_224_patch_n1000_seed42.npy) · [pooled](imagenet-200k/features/030_mc2_b16_224.npy) · [10k blocks](https://huggingface.co/datasets/336labs/VisionEncoder-Eval-ReproData/tree/main/lcs-10k/vision/mc2_b16_224) | |
| | [`mc2_b16_384`](https://github.com/facebookresearch/MetaCLIP#pre-trained-models) | LANG | `[576, 768]` | 768 | [patches](lcs-1k/vision/mc2_b16_384_patch_n1000_seed42.npy) · [pooled](imagenet-200k/features/031_mc2_b16_384.npy) · [10k blocks](https://huggingface.co/datasets/336labs/VisionEncoder-Eval-ReproData/tree/main/lcs-10k/vision/mc2_b16_384) | |
| | [`mc2_b32_224`](https://github.com/facebookresearch/MetaCLIP#pre-trained-models) | LANG | `[49, 768]` | 768 | [patches](lcs-1k/vision/mc2_b32_224_patch_n1000_seed42.npy) · [pooled](imagenet-200k/features/032_mc2_b32_224.npy) · [10k blocks](https://huggingface.co/datasets/336labs/VisionEncoder-Eval-ReproData/tree/main/lcs-10k/vision/mc2_b32_224) | |
| | [`mc2_b32_224_mt5`](https://github.com/facebookresearch/MetaCLIP#pre-trained-models) | LANG | `[49, 768]` | 768 | [patches](lcs-1k/vision/mc2_b32_224_mt5_patch_n1000_seed42.npy) · [pooled](imagenet-200k/features/033_mc2_b32_224_mt5.npy) · [10k blocks](https://huggingface.co/datasets/336labs/VisionEncoder-Eval-ReproData/tree/main/lcs-10k/vision/mc2_b32_224_mt5) | |
| | [`mc2_b32_384`](https://github.com/facebookresearch/MetaCLIP#pre-trained-models) | LANG | `[144, 768]` | 768 | [patches](lcs-1k/vision/mc2_b32_384_patch_n1000_seed42.npy) · [pooled](imagenet-200k/features/034_mc2_b32_384.npy) · [10k blocks](https://huggingface.co/datasets/336labs/VisionEncoder-Eval-ReproData/tree/main/lcs-10k/vision/mc2_b32_384) | |
| | [`mc2_g14_224`](https://github.com/facebookresearch/MetaCLIP#pre-trained-models) | LANG | `[256, 1664]` | 1664 | [patches](lcs-1k/vision/mc2_g14_224_patch_n1000_seed42.npy) · [pooled](imagenet-200k/features/035_mc2_g14_224.npy) · [10k blocks](https://huggingface.co/datasets/336labs/VisionEncoder-Eval-ReproData/tree/main/lcs-10k/vision/mc2_g14_224) | |
| | [`mc2_g14_378`](https://github.com/facebookresearch/MetaCLIP#pre-trained-models) | LANG | `[729, 1664]` | 1664 | [patches](lcs-1k/vision/mc2_g14_378_patch_n1000_seed42.npy) · [pooled](imagenet-200k/features/036_mc2_g14_378.npy) · [10k blocks](https://huggingface.co/datasets/336labs/VisionEncoder-Eval-ReproData/tree/main/lcs-10k/vision/mc2_g14_378) | |
| | [`mc2_h14_378`](https://github.com/facebookresearch/MetaCLIP#pre-trained-models) | LANG | `[729, 1280]` | 1280 | [patches](lcs-1k/vision/mc2_h14_378_patch_n1000_seed42.npy) · [pooled](imagenet-200k/features/037_mc2_h14_378.npy) · [10k blocks](https://huggingface.co/datasets/336labs/VisionEncoder-Eval-ReproData/tree/main/lcs-10k/vision/mc2_h14_378) | |
| | [`mc2_l14_224`](https://github.com/facebookresearch/MetaCLIP#pre-trained-models) | LANG | `[256, 1024]` | 1024 | [patches](lcs-1k/vision/mc2_l14_224_patch_n1000_seed42.npy) · [pooled](imagenet-200k/features/038_mc2_l14_224.npy) · [10k blocks](https://huggingface.co/datasets/336labs/VisionEncoder-Eval-ReproData/tree/main/lcs-10k/vision/mc2_l14_224) | |
| | [`mc2_m16_224`](https://github.com/facebookresearch/MetaCLIP#pre-trained-models) | LANG | `[196, 512]` | 512 | [patches](lcs-1k/vision/mc2_m16_224_patch_n1000_seed42.npy) · [pooled](imagenet-200k/features/039_mc2_m16_224.npy) · [10k blocks](https://huggingface.co/datasets/336labs/VisionEncoder-Eval-ReproData/tree/main/lcs-10k/vision/mc2_m16_224) | |
| | [`mc2_m16_224_mt5`](https://github.com/facebookresearch/MetaCLIP#pre-trained-models) | LANG | `[196, 512]` | 512 | [patches](lcs-1k/vision/mc2_m16_224_mt5_patch_n1000_seed42.npy) · [pooled](imagenet-200k/features/040_mc2_m16_224_mt5.npy) · [10k blocks](https://huggingface.co/datasets/336labs/VisionEncoder-Eval-ReproData/tree/main/lcs-10k/vision/mc2_m16_224_mt5) | |
| | [`mc2_m16_384`](https://github.com/facebookresearch/MetaCLIP#pre-trained-models) | LANG | `[576, 512]` | 512 | [patches](lcs-1k/vision/mc2_m16_384_patch_n1000_seed42.npy) · [pooled](imagenet-200k/features/041_mc2_m16_384.npy) · [10k blocks](https://huggingface.co/datasets/336labs/VisionEncoder-Eval-ReproData/tree/main/lcs-10k/vision/mc2_m16_384) | |
| | [`mc2_s16_224`](https://github.com/facebookresearch/MetaCLIP#pre-trained-models) | LANG | `[196, 384]` | 384 | [patches](lcs-1k/vision/mc2_s16_224_patch_n1000_seed42.npy) · [pooled](imagenet-200k/features/042_mc2_s16_224.npy) · [10k blocks](https://huggingface.co/datasets/336labs/VisionEncoder-Eval-ReproData/tree/main/lcs-10k/vision/mc2_s16_224) | |
| | [`mc2_s16_224_mt5`](https://github.com/facebookresearch/MetaCLIP#pre-trained-models) | LANG | `[196, 384]` | 384 | [patches](lcs-1k/vision/mc2_s16_224_mt5_patch_n1000_seed42.npy) · [pooled](imagenet-200k/features/043_mc2_s16_224_mt5.npy) · [10k blocks](https://huggingface.co/datasets/336labs/VisionEncoder-Eval-ReproData/tree/main/lcs-10k/vision/mc2_s16_224_mt5) | |
| | [`mc2_s16_384`](https://github.com/facebookresearch/MetaCLIP#pre-trained-models) | LANG | `[576, 384]` | 384 | [patches](lcs-1k/vision/mc2_s16_384_patch_n1000_seed42.npy) · [pooled](imagenet-200k/features/044_mc2_s16_384.npy) · [10k blocks](https://huggingface.co/datasets/336labs/VisionEncoder-Eval-ReproData/tree/main/lcs-10k/vision/mc2_s16_384) | |
| | [`pe_core_b16_224`](https://huggingface.co/facebook/PE-Core-B16-224) | LANG | `[196, 768]` | 768 | [patches](lcs-1k/vision/pe_core_b16_224_patch_n1000_seed42.npy) · [pooled](imagenet-200k/features/045_pe_core_b16_224.npy) · [10k blocks](https://huggingface.co/datasets/336labs/VisionEncoder-Eval-ReproData/tree/main/lcs-10k/vision/pe_core_b16_224) | |
| | [`pe_core_g14_448`](https://huggingface.co/facebook/PE-Core-G14-448) | LANG | `[1024, 1536]` | 1536 | [patches](lcs-1k/vision/pe_core_g14_448_patch_n1000_seed42.npy) · [pooled](imagenet-200k/features/046_pe_core_g14_448.npy) · [10k blocks](https://huggingface.co/datasets/336labs/VisionEncoder-Eval-ReproData/tree/main/lcs-10k/vision/pe_core_g14_448) | |
| | [`pe_lang_l14_448`](https://huggingface.co/facebook/PE-Lang-L14-448) | LANG | `[1024, 1024]` | 1024 | [patches](lcs-1k/vision/pe_lang_l14_448_patch_n1000_seed42.npy) · [pooled](imagenet-200k/features/047_pe_lang_l14_448.npy) · [10k blocks](https://huggingface.co/datasets/336labs/VisionEncoder-Eval-ReproData/tree/main/lcs-10k/vision/pe_lang_l14_448) | |
| | [`siglip2_b16_224`](https://github.com/google-research/big_vision/blob/main/big_vision/configs/proj/image_text/README_siglip2.md) | LANG | `[196, 768]` | 768 | [patches](lcs-1k/vision/siglip2_b16_224_patch_n1000_seed42.npy) · [pooled](imagenet-200k/features/052_siglip2_b16_224.npy) · [10k blocks](https://huggingface.co/datasets/336labs/VisionEncoder-Eval-ReproData/tree/main/lcs-10k/vision/siglip2_b16_224) | |
| | [`siglip2_b16_256`](https://github.com/google-research/big_vision/blob/main/big_vision/configs/proj/image_text/README_siglip2.md) | LANG | `[256, 768]` | 768 | [patches](lcs-1k/vision/siglip2_b16_256_patch_n1000_seed42.npy) · [pooled](imagenet-200k/features/053_siglip2_b16_256.npy) · [10k blocks](https://huggingface.co/datasets/336labs/VisionEncoder-Eval-ReproData/tree/main/lcs-10k/vision/siglip2_b16_256) | |
| | [`siglip2_b16_384`](https://github.com/google-research/big_vision/blob/main/big_vision/configs/proj/image_text/README_siglip2.md) | LANG | `[576, 768]` | 768 | [patches](lcs-1k/vision/siglip2_b16_384_patch_n1000_seed42.npy) · [pooled](imagenet-200k/features/054_siglip2_b16_384.npy) · [10k blocks](https://huggingface.co/datasets/336labs/VisionEncoder-Eval-ReproData/tree/main/lcs-10k/vision/siglip2_b16_384) | |
| | [`siglip2_b16_512`](https://github.com/google-research/big_vision/blob/main/big_vision/configs/proj/image_text/README_siglip2.md) | LANG | `[1024, 768]` | 768 | [patches](lcs-1k/vision/siglip2_b16_512_patch_n1000_seed42.npy) · [pooled](imagenet-200k/features/055_siglip2_b16_512.npy) · [10k blocks](https://huggingface.co/datasets/336labs/VisionEncoder-Eval-ReproData/tree/main/lcs-10k/vision/siglip2_b16_512) | |
| | [`siglip2_b32_256`](https://github.com/google-research/big_vision/blob/main/big_vision/configs/proj/image_text/README_siglip2.md) | LANG | `[64, 768]` | 768 | [patches](lcs-1k/vision/siglip2_b32_256_patch_n1000_seed42.npy) · [pooled](imagenet-200k/features/056_siglip2_b32_256.npy) · [10k blocks](https://huggingface.co/datasets/336labs/VisionEncoder-Eval-ReproData/tree/main/lcs-10k/vision/siglip2_b32_256) | |
| | [`siglip2_g16_256`](https://github.com/google-research/big_vision/blob/main/big_vision/configs/proj/image_text/README_siglip2.md) | LANG | `[256, 1536]` | 1536 | [patches](lcs-1k/vision/siglip2_g16_256_patch_n1000_seed42.npy) · [pooled](imagenet-200k/features/057_siglip2_g16_256.npy) · [10k blocks](https://huggingface.co/datasets/336labs/VisionEncoder-Eval-ReproData/tree/main/lcs-10k/vision/siglip2_g16_256) | |
| | [`siglip2_g16_384`](https://github.com/google-research/big_vision/blob/main/big_vision/configs/proj/image_text/README_siglip2.md) | LANG | `[576, 1536]` | 1536 | [patches](lcs-1k/vision/siglip2_g16_384_patch_n1000_seed42.npy) · [pooled](imagenet-200k/features/058_siglip2_g16_384.npy) · [10k blocks](https://huggingface.co/datasets/336labs/VisionEncoder-Eval-ReproData/tree/main/lcs-10k/vision/siglip2_g16_384) | |
| | [`siglip2_l16_256`](https://github.com/google-research/big_vision/blob/main/big_vision/configs/proj/image_text/README_siglip2.md) | LANG | `[256, 1024]` | 1024 | [patches](lcs-1k/vision/siglip2_l16_256_patch_n1000_seed42.npy) · [pooled](imagenet-200k/features/059_siglip2_l16_256.npy) · [10k blocks](https://huggingface.co/datasets/336labs/VisionEncoder-Eval-ReproData/tree/main/lcs-10k/vision/siglip2_l16_256) | |
| | [`siglip2_l16_384`](https://github.com/google-research/big_vision/blob/main/big_vision/configs/proj/image_text/README_siglip2.md) | LANG | `[576, 1024]` | 1024 | [patches](lcs-1k/vision/siglip2_l16_384_patch_n1000_seed42.npy) · [pooled](imagenet-200k/features/060_siglip2_l16_384.npy) · [10k blocks](https://huggingface.co/datasets/336labs/VisionEncoder-Eval-ReproData/tree/main/lcs-10k/vision/siglip2_l16_384) | |
| | [`siglip2_l16_512`](https://github.com/google-research/big_vision/blob/main/big_vision/configs/proj/image_text/README_siglip2.md) | LANG | `[1024, 1024]` | 1024 | [patches](lcs-1k/vision/siglip2_l16_512_patch_n1000_seed42.npy) · [pooled](imagenet-200k/features/061_siglip2_l16_512.npy) · [10k blocks](https://huggingface.co/datasets/336labs/VisionEncoder-Eval-ReproData/tree/main/lcs-10k/vision/siglip2_l16_512) | |
| | [`siglip2_sm14_224`](https://github.com/google-research/big_vision/blob/main/big_vision/configs/proj/image_text/README_siglip2.md) | LANG | `[256, 1152]` | 1152 | [patches](lcs-1k/vision/siglip2_sm14_224_patch_n1000_seed42.npy) · [pooled](imagenet-200k/features/062_siglip2_sm14_224.npy) · [10k blocks](https://huggingface.co/datasets/336labs/VisionEncoder-Eval-ReproData/tree/main/lcs-10k/vision/siglip2_sm14_224) | |
| | [`siglip2_sm14_384`](https://github.com/google-research/big_vision/blob/main/big_vision/configs/proj/image_text/README_siglip2.md) | LANG | `[729, 1152]` | 1152 | [patches](lcs-1k/vision/siglip2_sm14_384_patch_n1000_seed42.npy) · [pooled](imagenet-200k/features/063_siglip2_sm14_384.npy) · [10k blocks](https://huggingface.co/datasets/336labs/VisionEncoder-Eval-ReproData/tree/main/lcs-10k/vision/siglip2_sm14_384) | |
| | [`siglip2_sm16_256`](https://github.com/google-research/big_vision/blob/main/big_vision/configs/proj/image_text/README_siglip2.md) | LANG | `[256, 1152]` | 1152 | [patches](lcs-1k/vision/siglip2_sm16_256_patch_n1000_seed42.npy) · [pooled](imagenet-200k/features/064_siglip2_sm16_256.npy) · [10k blocks](https://huggingface.co/datasets/336labs/VisionEncoder-Eval-ReproData/tree/main/lcs-10k/vision/siglip2_sm16_256) | |
| | [`siglip2_sm16_384`](https://github.com/google-research/big_vision/blob/main/big_vision/configs/proj/image_text/README_siglip2.md) | LANG | `[576, 1152]` | 1152 | [patches](lcs-1k/vision/siglip2_sm16_384_patch_n1000_seed42.npy) · [pooled](imagenet-200k/features/065_siglip2_sm16_384.npy) · [10k blocks](https://huggingface.co/datasets/336labs/VisionEncoder-Eval-ReproData/tree/main/lcs-10k/vision/siglip2_sm16_384) | |
| | [`siglip2_sm16_512`](https://github.com/google-research/big_vision/blob/main/big_vision/configs/proj/image_text/README_siglip2.md) | LANG | `[1024, 1152]` | 1152 | [patches](lcs-1k/vision/siglip2_sm16_512_patch_n1000_seed42.npy) · [pooled](imagenet-200k/features/066_siglip2_sm16_512.npy) · [10k blocks](https://huggingface.co/datasets/336labs/VisionEncoder-Eval-ReproData/tree/main/lcs-10k/vision/siglip2_sm16_512) | |
| | [`dino_vitb16`](https://huggingface.co/facebook/dino-vitb16) | SSL | `[196, 768]` | 1536 | [patches](lcs-1k/vision/dino_vitb16_patch_n1000_seed42.npy) · [pooled](imagenet-200k/features/002_dino_vitb16.npy) · [10k blocks](https://huggingface.co/datasets/336labs/VisionEncoder-Eval-ReproData/tree/main/lcs-10k/vision/dino_vitb16) | |
| | [`dino_vitb8`](https://huggingface.co/facebook/dino-vitb8) | SSL | `[784, 768]` | 1536 | [patches](lcs-1k/vision/dino_vitb8_patch_n1000_seed42.npy) · [pooled](imagenet-200k/features/003_dino_vitb8.npy) · [10k blocks](https://huggingface.co/datasets/336labs/VisionEncoder-Eval-ReproData/tree/main/lcs-10k/vision/dino_vitb8) | |
| | [`dino_vits16`](https://huggingface.co/facebook/dino-vits16) | SSL | `[196, 384]` | 768 | [patches](lcs-1k/vision/dino_vits16_patch_n1000_seed42.npy) · [pooled](imagenet-200k/features/004_dino_vits16.npy) · [10k blocks](https://huggingface.co/datasets/336labs/VisionEncoder-Eval-ReproData/tree/main/lcs-10k/vision/dino_vits16) | |
| | [`dino_vits8`](https://huggingface.co/facebook/dino-vits8) | SSL | `[784, 384]` | 768 | [patches](lcs-1k/vision/dino_vits8_patch_n1000_seed42.npy) · [pooled](imagenet-200k/features/005_dino_vits8.npy) · [10k blocks](https://huggingface.co/datasets/336labs/VisionEncoder-Eval-ReproData/tree/main/lcs-10k/vision/dino_vits8) | |
| | [`dinov2_base`](https://huggingface.co/facebook/dinov2-base) | SSL | `[256, 768]` | 1536 | [patches](lcs-1k/vision/dinov2_base_patch_n1000_seed42.npy) · [pooled](imagenet-200k/features/006_dinov2_base.npy) · [10k blocks](https://huggingface.co/datasets/336labs/VisionEncoder-Eval-ReproData/tree/main/lcs-10k/vision/dinov2_base) | |
| | [`dinov2_giant`](https://huggingface.co/facebook/dinov2-giant) | SSL | `[256, 1536]` | 3072 | [patches](lcs-1k/vision/dinov2_giant_patch_n1000_seed42.npy) · [pooled](imagenet-200k/features/007_dinov2_giant.npy) · [10k blocks](https://huggingface.co/datasets/336labs/VisionEncoder-Eval-ReproData/tree/main/lcs-10k/vision/dinov2_giant) | |
| | [`dinov2_large`](https://huggingface.co/facebook/dinov2-large) | SSL | `[256, 1024]` | 2048 | [patches](lcs-1k/vision/dinov2_large_patch_n1000_seed42.npy) · [pooled](imagenet-200k/features/008_dinov2_large.npy) · [10k blocks](https://huggingface.co/datasets/336labs/VisionEncoder-Eval-ReproData/tree/main/lcs-10k/vision/dinov2_large) | |
| | [`dinov2_small`](https://huggingface.co/facebook/dinov2-small) | SSL | `[256, 384]` | 768 | [patches](lcs-1k/vision/dinov2_small_patch_n1000_seed42.npy) · [pooled](imagenet-200k/features/009_dinov2_small.npy) · [10k blocks](https://huggingface.co/datasets/336labs/VisionEncoder-Eval-ReproData/tree/main/lcs-10k/vision/dinov2_small) | |
| | [`dinov3_vitl16`](https://huggingface.co/facebook/dinov3-vitl16-pretrain-lvd1689m) | SSL | `[256, 1024]` | 2048 | [patches](lcs-1k/vision/dinov3_vitl16_patch_n1000_seed42.npy) · [pooled](imagenet-200k/features/010_dinov3_vitl16.npy) · [10k blocks](https://huggingface.co/datasets/336labs/VisionEncoder-Eval-ReproData/tree/main/lcs-10k/vision/dinov3_vitl16) | |
| | [`eupe_convnext_b`](https://huggingface.co/facebook/EUPE-ConvNeXt-B) | SSL | `[64, 1024]` | 1024 | [patches](lcs-1k/vision/eupe_convnext_b_patch_n1000_seed42.npy) · [pooled](imagenet-200k/features/016_eupe_convnext_b.npy) · [10k blocks](https://huggingface.co/datasets/336labs/VisionEncoder-Eval-ReproData/tree/main/lcs-10k/vision/eupe_convnext_b) | |
| | [`eupe_vit_b`](https://huggingface.co/facebook/EUPE-ViT-B) | SSL | `[256, 768]` | 1536 | [patches](lcs-1k/vision/eupe_vit_b_patch_n1000_seed42.npy) · [pooled](imagenet-200k/features/017_eupe_vit_b.npy) · [10k blocks](https://huggingface.co/datasets/336labs/VisionEncoder-Eval-ReproData/tree/main/lcs-10k/vision/eupe_vit_b) | |
| | [`eupe_vit_s`](https://huggingface.co/facebook/EUPE-ViT-S) | SSL | `[256, 384]` | 768 | [patches](lcs-1k/vision/eupe_vit_s_patch_n1000_seed42.npy) · [pooled](imagenet-200k/features/018_eupe_vit_s.npy) · [10k blocks](https://huggingface.co/datasets/336labs/VisionEncoder-Eval-ReproData/tree/main/lcs-10k/vision/eupe_vit_s) | |
| | [`eupe_vit_t`](https://huggingface.co/facebook/EUPE-ViT-T) | SSL | `[256, 192]` | 384 | [patches](lcs-1k/vision/eupe_vit_t_patch_n1000_seed42.npy) · [pooled](imagenet-200k/features/019_eupe_vit_t.npy) · [10k blocks](https://huggingface.co/datasets/336labs/VisionEncoder-Eval-ReproData/tree/main/lcs-10k/vision/eupe_vit_t) | |
| | [`ijepa_vith14`](https://github.com/facebookresearch/ijepa) | SSL | `[256, 1280]` | 1280 | [patches](lcs-1k/vision/ijepa_vith14_patch_n1000_seed42.npy) · [pooled](imagenet-200k/features/020_ijepa_vith14.npy) · [10k blocks](https://huggingface.co/datasets/336labs/VisionEncoder-Eval-ReproData/tree/main/lcs-10k/vision/ijepa_vith14) | |
| | [`pixio_vitb16`](https://huggingface.co/facebook/pixio-vitb16) | SSL | `[256, 768]` | 768 | [patches](lcs-1k/vision/pixio_vitb16_patch_n1000_seed42.npy) · [pooled](imagenet-200k/features/048_pixio_vitb16.npy) · [10k blocks](https://huggingface.co/datasets/336labs/VisionEncoder-Eval-ReproData/tree/main/lcs-10k/vision/pixio_vitb16) | |
| | [`pixio_vith16`](https://huggingface.co/facebook/pixio-vith16) | SSL | `[256, 1280]` | 1280 | [patches](lcs-1k/vision/pixio_vith16_patch_n1000_seed42.npy) · [pooled](imagenet-200k/features/049_pixio_vith16.npy) · [10k blocks](https://huggingface.co/datasets/336labs/VisionEncoder-Eval-ReproData/tree/main/lcs-10k/vision/pixio_vith16) | |
| | [`pixio_vitl16`](https://huggingface.co/facebook/pixio-vitl16) | SSL | `[256, 1024]` | 1024 | [patches](lcs-1k/vision/pixio_vitl16_patch_n1000_seed42.npy) · [pooled](imagenet-200k/features/050_pixio_vitl16.npy) · [10k blocks](https://huggingface.co/datasets/336labs/VisionEncoder-Eval-ReproData/tree/main/lcs-10k/vision/pixio_vitl16) | |
| | [`raev2_dinov3l_k7`](https://huggingface.co/nyu-visionx/RAEv2-models) | SSL | `[256, 1024]` | 1024 | [patches](lcs-1k/vision/raev2_dinov3l_k7_patch_n1000_seed42.npy) · [pooled](imagenet-200k/features/051_raev2_dinov3l_k7.npy) · [10k blocks](https://huggingface.co/datasets/336labs/VisionEncoder-Eval-ReproData/tree/main/lcs-10k/vision/raev2_dinov3l_k7) | |
| | [`webssl_dino1b_full2b_224`](https://huggingface.co/facebook/webssl-dino1b-full2b-224) | SSL | `[256, 1536]` | 1536 | [patches](lcs-1k/vision/webssl_dino1b_full2b_224_patch_n1000_seed42.npy) · [pooled](imagenet-200k/features/067_webssl_dino1b_full2b_224.npy) · [10k blocks](https://huggingface.co/datasets/336labs/VisionEncoder-Eval-ReproData/tree/main/lcs-10k/vision/webssl_dino1b_full2b_224) | |
| | [`webssl_mae1b_full2b_224`](https://huggingface.co/facebook/webssl-mae1b-full2b-224) | SSL | `[256, 1536]` | 1536 | [patches](lcs-1k/vision/webssl_mae1b_full2b_224_patch_n1000_seed42.npy) · [pooled](imagenet-200k/features/068_webssl_mae1b_full2b_224.npy) · [10k blocks](https://huggingface.co/datasets/336labs/VisionEncoder-Eval-ReproData/tree/main/lcs-10k/vision/webssl_mae1b_full2b_224) | |
| | [`webssl_mae300m_full2b_224`](https://huggingface.co/facebook/webssl-mae300m-full2b-224) | SSL | `[196, 1024]` | 1024 | [patches](lcs-1k/vision/webssl_mae300m_full2b_224_patch_n1000_seed42.npy) · [pooled](imagenet-200k/features/069_webssl_mae300m_full2b_224.npy) · [10k blocks](https://huggingface.co/datasets/336labs/VisionEncoder-Eval-ReproData/tree/main/lcs-10k/vision/webssl_mae300m_full2b_224) | |
| | [`webssl_mae3b_full2b_224`](https://huggingface.co/facebook/webssl-mae3b-full2b-224) | SSL | `[256, 3072]` | 3072 | [patches](lcs-1k/vision/webssl_mae3b_full2b_224_patch_n1000_seed42.npy) · [pooled](imagenet-200k/features/070_webssl_mae3b_full2b_224.npy) · [10k blocks](https://huggingface.co/datasets/336labs/VisionEncoder-Eval-ReproData/tree/main/lcs-10k/vision/webssl_mae3b_full2b_224) | |
| |
| |
| <a id="datasets-required-by-the-evaluation-pipelines"></a> |
| |
| ## Data Sources |
| |
| This repository provides the three cached feature subsets listed below. Raw images, MLLM training data, evaluation benchmarks, and other baseline datasets are obtained separately. |
| |
| ### Data used by each evaluation methods |
| |
| | Method | Data needed | Available here / source | |
| |---|---|---| |
| | RAVEL | Paired LCS-558K image patches and text features | `lcs-1k/` and `lcs-10k/` | |
| | RSA, CCA, GW, MutualNN | Paired LCS-558K global image vectors and text features | Shared samples and text are included; prepare the required global image vectors separately | |
| | kNN | ImageNet training features, labels, and support/query split | `imagenet-200k/` | |
| | Linear probing | ImageNet train and validation images | [ImageNet](https://www.image-net.org/download.php); the released 200k cache contains training images only | |
| | Alignment probing | COCO Karpathy captions; DreamLIP CC3M for the cross-dataset variants | [COCO / Karpathy splits](https://cs.stanford.edu/people/karpathy/deepimagesent/) and [DreamLIP captions](https://huggingface.co/datasets/qidouxiong619/dreamlip_long_captions) | |
| | AC Policy / Law of Vision Representation | LCS-558K, Stage-1 projectors, and SPair-71k | [LLaVA-Pretrain](https://huggingface.co/datasets/liuhaotian/LLaVA-Pretrain), [Model Zoo](https://huggingface.co/336labs/VisionEncoder-to-MLLM-ModelZoo), and [SPair-71k](https://cvlab.postech.ac.kr/research/SPair-71k/index.html) | |
| | Mid-training loss | LCS-558K images/captions and projector-training logs | [LLaVA-Pretrain](https://huggingface.co/datasets/liuhaotian/LLaVA-Pretrain); training instructions in the [code repository](https://github.com/JuntaoTang/MLLM-VisionEncoder-Eval#running) | |
| | TokBench | Original images/annotations and matching tokenizer reconstructions | [TokBench](https://huggingface.co/datasets/Junfeng5/TokBench) | |
| | MLLM training / evaluation | Training: LCS-558K and filtered mix665k; evaluation: the 11 benchmarks below | [Model Zoo](https://huggingface.co/336labs/VisionEncoder-to-MLLM-ModelZoo) and the linked dataset sources | |
| |
| ### Downstream MLLM benchmarks used in this paper |
| |
| | Benchmark / required TSV | Source | |
| |---|---| |
| | `MMMU_TEST.tsv` | [MMMU](https://huggingface.co/datasets/MMMU/MMMU), test split | |
| | `MMBench_TEST_EN_V11.tsv` | [MMBench](https://github.com/open-compass/MMBench), English test v1.1 | |
| | `VQAv2_VAL.tsv` | [VQA v2](https://visualqa.org/download.html), validation split | |
| | `ScienceQA_VAL.tsv` | [ScienceQA](https://github.com/lupantech/ScienceQA), validation split | |
| | `ChartQA_TEST.tsv` | [ChartQA](https://github.com/vis-nlp/ChartQA), test split | |
| | `DocVQA_VAL.tsv` | [DocVQA](https://www.docvqa.org/datasets/docvqa), validation split | |
| | `TextVQA_VAL.tsv` | [TextVQA](https://textvqa.org/), validation split | |
| | `POPE.tsv` | [POPE](https://github.com/RUCAIBox/POPE) | |
| | `GQA_TestDev_Balanced.tsv` | [GQA](https://cs.stanford.edu/people/dorarad/gqa/download.html), balanced test-dev split | |
| | `MSCOCO_KARPATHY_TEST.tsv` | [Karpathy splits](https://cs.stanford.edu/people/karpathy/deepimagesent/) and [COCO images](https://cocodataset.org/#download) | |
| | `FLICKR30K_KARPATHY_TEST.tsv` | [Karpathy splits](https://cs.stanford.edu/people/karpathy/deepimagesent/) and [Flickr30k images](https://shannon.cs.illinois.edu/DenotationGraph/) | |
| |
| Use the [VLMEvalKit dataset loaders](https://github.com/open-compass/VLMEvalKit/tree/main/vlmeval/dataset) for benchmark preparation. |
| Configuration used in our paper are in the code repository's [Setup](https://github.com/JuntaoTang/MLLM-VisionEncoder-Eval#setup) and [Running](https://github.com/JuntaoTang/MLLM-VisionEncoder-Eval#running) sections. |
| |
| ## Metadata and Integrity |
| |
| The released arrays retain their original storage precision. Per-file metadata and sample manifests record available extraction and sampling information. [FILES.json](FILES.json) and [checksums.sha256](checksums.sha256) provide file sizes and SHA-256 checksums. |
| |
| ## Licensing and Attribution |
| |
| See [LICENSES.md](LICENSES.md). Use of the artifacts is subject to the applicable source-dataset and encoder terms. Source license information should be checked at the upstream links before reuse or redistribution; this card does not grant new rights over upstream images, captions, or model-derived artifacts. |
| |
| ## Citation |
| |
| If you use these feature caches or RAVEL, please cite the paper and the relevant source datasets and encoders: |
| |
| ```bibtex |
| @misc{yang2026strong, |
| title={A Strong Baseline for Evaluating Vision Encoders in Multimodal Large Language Models}, |
| author={Yang, Yilin and Tang, Jun-Tao and Wang, Kengyi and Su, Siyuan and Luo, Gaoyong and Chen, Mingda}, |
| year={2026}, |
| eprint={2610.05413}, |
| archivePrefix={arXiv}, |
| primaryClass={cs.CV}, |
| url={https://arxiv.org/abs/2610.05413} |
| } |
| ``` |
| |
