Skip to content

Pull requests: EleutherAI/sparsify

New pull request New
Clear current search query, filters, and sorts
2 Open 32 Closed
2 Open 32 Closed
Author
Filter by author
Loading
Label
Filter by label
Loading
Use alt + click/return to exclude labels
or ⇧ + click/return for logical OR
Projects
Filter by project
Loading
Milestones
Filter by milestone
Loading
Reviews
Filter by reviews
No reviews Review required Approved review Changes requested
Assignee
Filter by who’s assigned
Sort
Sort by
Newest Oldest Most commented Least commented Recently updated Least recently updated Best match
Most reactions
👍 👎 😄 🎉 😕 ❤️ 🚀 👀

Pull requests list

Add k decay schedule
#53 by luciaquirke was merged Feb 20, 2025 Loading…
Use rank 0 if not distributed
#52 by luciaquirke was merged Feb 19, 2025 Loading…
1
Fix illegal memory access in Xformers kernel
#51 by neverix was merged Feb 17, 2025 Loading…
1
Fix resume - num_tokens_since_fired is rank dependent
#50 by luciaquirke was merged Feb 14, 2025 Loading…
1
Fix SAE-centric naming, fix decoder bias for transcoders, add GroupMax activation
#49 by norabelrose was merged Feb 12, 2025 Loading…
5
auxk loss: sign flip for correct prediction of the residual
#47 by fel-thomas was merged Feb 12, 2025 Loading…
2
Use modified xformers Triton kernel
#46 by norabelrose was merged Feb 6, 2025 Loading…
Support using multiple random seeds for init
#39 by norabelrose was merged Dec 18, 2024 Loading…
Skip connections for transcoders
#38 by norabelrose was merged Dec 18, 2024 Loading…
[WIP] Enable training SAEs on vision models
#37 by luciaquirke was closed Feb 12, 2025 Loading…
1 of 3 tasks
2
Update README
#36 by luciaquirke was merged Dec 4, 2024 Loading…
1
Transcoder MVP
#35 by norabelrose was merged Nov 13, 2024 Loading…
3
Update README.md
#32 by pminervini was merged Oct 22, 2024 Loading…
Enable finetuning SAEs
#31 by luciaquirke was merged Oct 22, 2024 Loading…
1
Use the same spelling for cfg everywhere
#28 by luciaquirke was closed Sep 9, 2024 Loading…
Added back in "signed" property of SaeConfig to support loading older weights
#25 by jkminder was closed Aug 26, 2024 Loading…
2
Multi-TopK support
#23 by norabelrose was merged Aug 16, 2024 Loading…
Support resuming training after interruption
#22 by norabelrose was merged Aug 15, 2024 Loading…
Make loss more like standard MSE loss
#20 by norabelrose was merged Aug 8, 2024 Loading…
Support mmap datasets
#19 by norabelrose was merged Aug 2, 2024 Loading…
stop dropping samples every batch
#15 by Lewington-pitsos was closed Feb 12, 2025 Loading…
3
remove line
#14 by Lewington-pitsos was merged Jul 25, 2024 Loading…
fix bug in load many and support load multiple selected layers
#13 by HongchuanZeng was merged Jul 26, 2024 Loading…
4
Custom hookpoints
#12 by norabelrose was merged Jul 16, 2024 Loading…
add installation and testing instructions
#10 by Lewington-pitsos was closed Jul 15, 2024 Loading…
1
ProTip! Exclude everything labeled bug with -label:bug.