SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2503.12559 · 2025

AdaReTaKe: Adaptive Redundancy Reduction to Perceive Longer for Video-language Understanding

arXiv · PDF · Open in the Atlas

Code that ran

We lifted 23 functions out of this paper's own repositories and ran 17 of them in a sandbox. "Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results.

RepositoryRoleRan
sczwangxiao/video-flexreduc canonical 17 of 23
FunctionStatusWhere it lives
repeat_kv Ran sczwangxiao/video-flexreduc/retake/longvideo_cache.py
code served (permissive licence) · get_code("3c76e52815c5401d")
evaluate_mlvu_generation Ran sczwangxiao/video-flexreduc/retake/dataset_utils.py
code served (permissive licence) · get_code("81664204c8525de9")
gather_results Ran sczwangxiao/video-flexreduc/retake/infer_eval.py
code served (permissive licence) · get_code("e1e87b90e35fb630")
load_yaml Ran sczwangxiao/video-flexreduc/retake/infer_eval.py
code served (permissive licence) · get_code("d37f4e5bd2ad847b")
memory_bank_compress_MALLM Ran sczwangxiao/video-flexreduc/retake/visual_compression.py
code served (permissive licence) · get_code("6ad69145ddabafe5")
memory_bank_compress_MALLM_hard Ran sczwangxiao/video-flexreduc/retake/visual_compression.py
code served (permissive licence) · get_code("5f9f5bd51036cfc1")
memory_bank_compress_keyframe Ran sczwangxiao/video-flexreduc/retake/visual_compression.py
code served (permissive licence) · get_code("0366d231d1eb184f")
patch_llava_onevision_config Ran sczwangxiao/video-flexreduc/retake/monkeypatch.py
code served (permissive licence) · get_code("a83c7315b78d5c2c")
patch_qwen2_5_vl_config Ran sczwangxiao/video-flexreduc/retake/monkeypatch.py
code served (permissive licence) · get_code("d7609a472ef2dd0d")
patch_qwen2vl_config Ran sczwangxiao/video-flexreduc/retake/monkeypatch.py
code served (permissive licence) · get_code("8601a846757deb14")
retake_LlavaOnevisionForConditionalGeneration_get_chunk_size Ran sczwangxiao/video-flexreduc/retake/llava_onevision.py
code served (permissive licence) · get_code("8a68c201ee613f88")
retake_LlavaOnevisionForConditionalGeneration_segment_input_ids Ran sczwangxiao/video-flexreduc/retake/llava_onevision.py
code served (permissive licence) · get_code("7ddc25ea1dac67d2")
retake_Qwen2VLForConditionalGeneration_segment_input_ids Ran sczwangxiao/video-flexreduc/retake/qwen2_vl.py
code served (permissive licence) · get_code("eab30ba8c4d44a11")
retake_Qwen2_5_VLForConditionalGeneration_get_chunk_size Ran sczwangxiao/video-flexreduc/retake/qwen2_5_vl.py
code served (permissive licence) · get_code("e5124ec09df7329d")
retake_Qwen2_5_VLForConditionalGeneration_segment_input_ids Ran sczwangxiao/video-flexreduc/retake/qwen2_5_vl.py
code served (permissive licence) · get_code("676b41a7d2ba0916")
rotate_half Ran sczwangxiao/video-flexreduc/retake/longvideo_cache.py
code served (permissive licence) · get_code("e03d53ba9d4f9ae5")
trimm_results Ran sczwangxiao/video-flexreduc/retake/infer_eval.py
code served (permissive licence) · get_code("2a31a1d556220646")
apply_multimodal_rotary_pos_emb Not yet run sczwangxiao/video-flexreduc/retake/longvideo_cache.py
code served (permissive licence) · get_code("8971bec8b941516b")
eval_mlvu_results Not yet run sczwangxiao/video-flexreduc/retake/dataset_utils.py
code served (permissive licence) · get_code("1c4242b450f2a4ac")
eval_videomme_results Not yet run sczwangxiao/video-flexreduc/retake/dataset_utils.py
code served (permissive licence) · get_code("feee0c01f58c3b2e")
retake_Qwen2VLAttention_forward Not yet run sczwangxiao/video-flexreduc/retake/qwen2_vl.py
code served (permissive licence) · get_code("b9e6d7a085769799")
retake_Qwen2VLForConditionalGeneration_compress_video_tokens Not yet run sczwangxiao/video-flexreduc/retake/qwen2_vl.py
code served (permissive licence) · get_code("508f605be1850961")
retake_Qwen2_5_VLAttention_forward Not yet run sczwangxiao/video-flexreduc/retake/qwen2_5_vl.py
code served (permissive licence) · get_code("8104fe56ba8a1b10")

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

Multimodal Large Language Models (MLLMs) have revolutionized video understanding, yet are still limited by context length when processing long videos. Recent methods compress videos by leveraging visual redundancy uniformly, yielding promising results. Nevertheless, our quantitative analysis shows that redundancy varies significantly across time and model layers, necessitating a more flexible compression strategy. We propose AdaReTaKe, a training-free method that flexibly reduces visual redundancy by allocating compression ratios among time and layers with theoretical guarantees. Integrated into state-of-the-art MLLMs, AdaReTaKe improves processing capacity from 256 to 2048 frames while preserving critical information. Experiments on VideoMME, MLVU, LongVideoBench, and LVBench datasets demonstrate that AdaReTaKe outperforms existing methods by 2.3% and 2.8% for 7B and 72B models, respectively, with even greater improvements of 5.9% and 6.0% on the longest LVBench. Our code is available at https://github.com/SCZwangxiao/video-FlexReduc.git.

For agents

The same record, over MCP at https://syntology.ai/mcp:

get_harvested_code_for_paper("2503.12559")
get_code_for_paper("2503.12559")
have("2503.12559")

Connect an agent — have() is free.