Skip to content

[ICML'25] Official implementation of paper "SparseVLM: Visual Token Sparsification for Efficient Vision-Language Model Inference".

active 2024-10-042026-05-25 (UTC)

Partial coverage20,537 / 26,293 hourly files (78%) · 2 absent upstream · 5,753 failed, retryable2023-08-152026-08-14 (UTC)— sampled evenly across the window, so rankings and trends hold; absolute counts scale up.
Events
213
Pushes
22
Pull requests
3
Issues
24
Stars
125
Forks
11

Activity over time

Daily event counts in the loaded window

Line chart, 599 days from 2024-10-04 to 2026-05-25. Pushes: 22 total, peak 7 in a day. Pull requests: 3 total, peak 2 in a day. Issues: 24 total, peak 3 in a day. Comments: 21 total, peak 3 in a day. Stars: 125 total, peak 10 in a day.

  • Pushes
  • Pull requests
  • Issues
  • Comments
  • Stars

Stars, PRs, issues and forks are under-captured in the later part of this window. GH Archive progressively stopped capturing non-push events during 2026 — −95% or worse by the end of the window. Every series here except Pushes fades for that reason, so a decline above reflects the archive, not this repository. Pushes stay reliable throughout, so read them, and the contributor counts derived from them, as the real signal. Data health has the measurements.

Top contributors

Pushes, PRs, issues, reviews and comments — stars and forks excluded, so this is contribution rather than popularity

Recent activity

Latest issues, pull requests and releases

  • Issue#34Dayoung-Kil2026-02-12 03:49
    Question about FLOPs and Latency Measurement (Table 1)
  • Issue comment#8fode-design2026-01-08 08:54
    Could you please provide the code for efficiency analysis in Table 4?
  • Issue#32fode-design2026-01-07 04:06
    What is the specific calculation formula for latency (ms/token)? Is it only the decoding time that needs to be calculated, or does the pre-filling time also need to be taken into account?
  • Issue#31Sukidat2025-12-21 07:15
    VideoLlava环境
  • Issue comment#26oPresent2025-12-14 11:56
    Compatibility with FlashAttention
  • ReleaseGumpest2025-12-12 02:38
    SparseVLM+
  • Issue#28panxuyan2025-11-08 06:34
    P100 support?
  • Issue#22Gumpest2025-09-26 02:48
    How to obtain visualization results in Figure 6?
  • Issue comment#24Gumpest2025-09-26 02:46
    Could you please clarify the detailed meaning and computing method of "equal token num until now: 192.03125 ,total_layers_cuda_time:56175.52429199219,TFLOPs_avg_sample:1.8177142649287112"?
  • Issue#24Gumpest2025-09-26 02:46
    Could you please clarify the detailed meaning and computing method of "equal token num until now: 192.03125 ,total_layers_cuda_time:56175.52429199219,TFLOPs_avg_sample:1.8177142649287112"?
  • Issue#12Gumpest2025-09-26 02:45
    Correction of the accuracy of FastV
  • Issue comment#16xiaonengmiao2025-09-25 08:19
    other benchmark eval code
  • Issue comment#12Man-PaperRejected2025-08-15 08:20
    Correction of the accuracy of FastV
  • Issue#259249732922025-07-12 06:50
    Why the inference speed is extremely slower than the normal state? e.g. normal 5 mins on MME but SparseVLMs takes about 1 h?
  • Issue#249249732922025-07-11 03:46
    Could you please clarify the detailed meaning and computing method of "equal token num until now: 192.03125 ,total_layers_cuda_time:56175.52429199219,TFLOPs_avg_sample:1.8177142649287112"?
  • Issue#2Gumpest2025-06-04 08:04
    About Video Understanding
  • Issue comment#23Gumpest2025-06-04 07:57
    Upload SparseVideoLLaVA
  • Pull request#23Gumpest2025-06-04 07:57
  • Pull request#23leofan902025-06-04 07:35
  • Issue comment#2kinredon2025-05-24 04:37
    About Video Understanding
  • Issue#20zhuyglx2025-04-27 09:41
    SparseVLMs 能用在 qwen2.5-vl系列吗
  • Issue comment#18Gumpest2025-04-15 04:03
    I wonder how I can evaluate a model with lmms_eval
  • Issue comment#13Gumpest2025-04-15 04:00
    use the SparseVLMs in another MLLM
  • Pull request#19ylnxxts2025-03-02 10:55
  • Issue#17KaKa-1012024-12-24 08:04
    How to calculate the true used token numbers?

Totals cover only the window loaded into ClickHouse and count events, not GitHub's lifetime totals — 125 stars here means stars gained during the window, not the repo's star count.