eleutherai/sae issues and pull requests

#45 - Does this framework support multi-machine multi-GPU distributed training?

Issue - State: closed - Opened by likewei-rookie 21 days ago - 1 comment

#44 - How to use decoder?

Issue - State: closed - Opened by henselman-petrusek about 1 month ago - 1 comment

#43 - Nested directories prevent methods from being visible to top-level package

Issue - State: open - Opened by kewiechecki about 1 month ago

#41 - Error loading models

Issue - State: open - Opened by henselman-petrusek about 2 months ago

#40 - Different Encoder value for same input in CPU and CUDA

Issue - State: closed - Opened by Tomertech about 2 months ago

#39 - Support using multiple random seeds for init

Pull Request - State: closed - Opened by norabelrose about 2 months ago

#38 - Skip connections for transcoders

Pull Request - State: closed - Opened by norabelrose about 2 months ago

#37 - [WIP] Enable training SAEs on vision models

Pull Request - State: open - Opened by luciaquirke 2 months ago

#36 - Update README

Pull Request - State: closed - Opened by luciaquirke 2 months ago

#35 - Transcoder MVP

Pull Request - State: closed - Opened by norabelrose 3 months ago

#34 - Missing Normalization of Input/Output in SAE Implementation

Issue - State: open - Opened by tatsuropfgt 3 months ago

#33 - Plans to release additional SAEs for Llama-3.1-8B

Issue - State: open - Opened by jacknewsom 3 months ago

#32 - Update README.md

Pull Request - State: closed - Opened by pminervini 4 months ago

#31 - Enable finetuning SAEs

Pull Request - State: closed - Opened by luciaquirke 4 months ago

#30 - Bibtex citation

Issue - State: closed - Opened by amakelov 4 months ago - 3 comments

#29 - AttributeError: module 'triton.language' has no attribute 'device_assert' when calling `Sae.decode()`

Issue - State: closed - Opened by zlkqz 5 months ago - 2 comments

#28 - Use the same spelling for cfg everywhere

Pull Request - State: closed - Opened by luciaquirke 5 months ago

#27 - Training process stuck with --distribute_modules flag

Issue - State: open - Opened by Hambaobao 6 months ago - 2 comments

#26 - A requirements.txt file is needed

Issue - State: closed - Opened by SyedShaQutub 6 months ago - 1 comment

#25 - Added back in "signed" property of SaeConfig to support loading older weights

Pull Request - State: closed - Opened by jkminder 6 months ago - 2 comments

#24 - Minor typos

Issue - State: closed - Opened by regressorjournal 6 months ago - 2 comments

#23 - Multi-TopK support

Pull Request - State: closed - Opened by norabelrose 6 months ago

#22 - Support resuming training after interruption

Pull Request - State: closed - Opened by norabelrose 6 months ago

#21 - Training SAEs for Chat Models

Issue - State: closed - Opened by Hambaobao 6 months ago - 2 comments

#20 - Make loss more like standard MSE loss

Pull Request - State: closed - Opened by norabelrose 6 months ago

#19 - Support mmap datasets

Pull Request - State: closed - Opened by norabelrose 6 months ago

#18 - Analyze the features

Issue - State: closed - Opened by ypw-lbj 7 months ago - 1 comment

#17 - Can SAE support the training of larger models?

Issue - State: closed - Opened by ypw-lbj 7 months ago - 1 comment

#16 - SAE layer lndex question

Issue - State: closed - Opened by yangjingyuan 7 months ago - 1 comment

#15 - stop dropping samples every batch

Pull Request - State: open - Opened by Lewington-pitsos 7 months ago - 3 comments

#14 - remove line

Pull Request - State: closed - Opened by Lewington-pitsos 7 months ago

#13 - fix bug in load many and support load multiple selected layers

Pull Request - State: closed - Opened by HongchuanZeng 7 months ago - 1 comment

#12 - Custom hookpoints

Pull Request - State: closed - Opened by norabelrose 7 months ago

#12 - Custom hookpoints

Pull Request - State: closed - Opened by norabelrose 7 months ago

#11 - Questions on output_hidden_states to get activations

Issue - State: closed - Opened by HPLQAQ 7 months ago - 1 comment

#10 - add installation and testing instructions

Pull Request - State: closed - Opened by Lewington-pitsos 7 months ago - 1 comment

#9 - Fix for file name change

Pull Request - State: closed - Opened by vdlad 7 months ago - 2 comments

#8 - Fix model download issue

Pull Request - State: closed - Opened by HPLQAQ 7 months ago - 2 comments

#7 - Missing cfg.json File

Issue - State: closed - Opened by Hambaobao 7 months ago - 5 comments

#6 - add requirements and update readme

Pull Request - State: closed - Opened by Lewington-pitsos 7 months ago - 6 comments

#5 - Fix hang after calling `SaeTrainer.save()` w/ DDP

Pull Request - State: closed - Opened by vgel 7 months ago - 1 comment

#4 - Signed SAEs, better encoder init, etc.

Pull Request - State: closed - Opened by norabelrose 7 months ago

#3 - Flattening inputs in forward

Pull Request - State: closed - Opened by SrGonao 8 months ago

#2 - Fix immediate error with default command

Pull Request - State: closed - Opened by rhaps0dy 8 months ago - 1 comment

#1 - Default command errors because SDPA not supported for Pythia

Issue - State: closed - Opened by rhaps0dy 8 months ago - 4 comments

GitHub / eleutherai/sae issues and pull requests