Skip to content

boyiwei/alignment-attribution-code

View on GitHub ↗Related repositories →

[ICML 2024] Assessing the Brittleness of Safety Alignment via Pruning and Low-Rank Modifications

active 2024-03-072026-03-12 (UTC)

Partial coverage16,802 / 21,336 hourly files (79%) · 2 absent upstream · 4,533 failed, retryable2024-03-062026-08-12 (UTC)— sampled evenly across the window, so rankings and trends hold; absolute counts scale up.
Events
93
Pushes
4
Pull requests
0
Issues
14
Stars
36
Forks
9

Activity over time

Daily event counts in the loaded window

Line chart, 736 days from 2024-03-07 to 2026-03-12. Pushes: 4 total, peak 2 in a day. Pull requests: 0 total, peak 0 in a day. Issues: 14 total, peak 3 in a day. Comments: 30 total, peak 12 in a day. Stars: 36 total, peak 2 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

ContributorContributionsPushesPRsComments
boyiwei284016
zzxxxl8007
jeffreyzhanghc4004
ybwang1192001
YichenBC1001
MqLeet1001
Zongmin7411000
orgadhadas1000
SJSoJSooJ1000
slardar941000

Recent activity

Latest issues, pull requests and releases

  • Issue#16Zongmin7412026-03-12 00:47
    Code to calculate "Rank-level subspace similarity" for Fig. 4b
  • Issue comment#12boyiwei2025-03-30 12:43
    In prune_wandg_set_difference, there is No such file or directory
  • Issue#12boyiwei2025-03-30 12:43
    In prune_wandg_set_difference, there is No such file or directory
  • Issue#12SJSoJSooJ2025-03-30 09:27
    In prune_wandg_set_difference, there is No such file or directory
  • Issue comment#4boyiwei2025-03-14 07:48
    Great work. Can you release the code fine-tuning the models while the alignment region is frozen?
  • Issue comment#11boyiwei2025-02-23 21:27
    Why did you take the absolute value in the SNIP score?
  • Issue#11boyiwei2025-02-23 21:27
    Why did you take the absolute value in the SNIP score?
  • Issue#11orgadhadas2025-02-23 15:31
    Why did you take the absolute value in the SNIP score?
  • Issue comment#10MqLeet2025-02-19 02:28
    Code about drawing Fig.4 in paper
  • Issue#8boyiwei2025-01-18 22:04
    Clarification on the loop for wandg
  • Issue comment#9boyiwei2025-01-02 14:08
    error in zero-shot evaluation for safety-related ranks
  • Issue comment#9ybwang1192025-01-02 12:26
    error in zero-shot evaluation for safety-related ranks
  • Issue#9ybwang1192025-01-02 12:22
    error in zero-shot evaluation for safety-related ranks
  • Issue#5boyiwei2024-09-10 15:29
    Perplexity is way off for Llama2 models
  • Issue comment#5boyiwei2024-09-09 01:59
    Perplexity is way off for Llama2 models
  • Issue#5slardar942024-09-08 02:05
    Perplexity is way off for Llama2 models
  • Issue comment#2jeffreyzhanghc2024-08-28 03:30
    set_difference problem
  • Issue comment#2boyiwei2024-08-28 03:12
    set_difference problem
  • Issue comment#2jeffreyzhanghc2024-08-28 03:08
    set_difference problem
  • Issue comment#2boyiwei2024-08-28 02:59
    set_difference problem
  • Issue comment#2jeffreyzhanghc2024-08-28 02:56
    set_difference problem
  • Issue comment#2boyiwei2024-08-28 02:47
    set_difference problem
  • Issue comment#2jeffreyzhanghc2024-08-28 02:45
    set_difference problem
  • Issue#4boyiwei2024-07-20 08:15
    Great work. Can you release the code fine-tuning the models while the alignment region is frozen?
  • Issue comment#4YichenBC2024-07-20 07:59
    Great work. Can you release the code fine-tuning the models while the alignment region is frozen?

Totals cover only the window loaded into ClickHouse and count events, not GitHub's lifetime totals — 36 stars here means stars gained during the window, not the repo's star count.