RepoJournal
Interpretability Interpretability
14 wires and counting

$ follow Interpretability

Keep up with Interpretability in about 3 minutes: what actually shipped — the commits, pull requests, releases, and security advisories that matter.

or

fair warning: these emails are deeply technical. diffs, version numbers, CVEs, benchmark deltas. if that's not your idea of a good read, this isn't your newsletter.

Folds into your digest — weekly by default, monthly if you prefer. Unsubscribe in one click.

$ status

wire 2026-10-19
stories 4
RSS Archive Reviews

© 2026 RepoJournal Home Showcase How it works Privacy

$ the-wire · showcase

SAELens 6.54.6 forwards model kwargs to reconstruction evals

By RepoJournal · Filed · About Interpretability · Composed from the cited sources · methodology

AaravMehta-07
1 person shipped this
AaravMehta-07 @AaravMehta-07 2 cited

SAELens shipped 6.54.6, closing a gap where reconstruction metrics ignored the model options you passed to run_evals while sparsity and variance evaluation honored them.

6.54.6 jbloomAus/SAELens

by semantic-release

Version 6.54.6 carries the fix through to a released package, so pulling it is now the way to get consistent evaluation behavior rather than pinning a branch.

fix: forward model kwargs to reconstruction evaluation jbloomAus/SAELens

by AaravMehta-07

Previously run_evals(model_kwargs=...) forwarded your options to sparsity and variance evaluation but dropped them from the CE/KL reconstruction path, so an attention_mask changed the sparsity forward pass while reconstruction metrics silently fell back to the model's default mask. The mapping now flows through get_downstream_reconstruction_metrics to get_recons_loss, which already passed it to...

fix: forward model kwargs to reconstruction evaluation (#765) jbloomAus/SAELens

by Aarav

The merge landed as commit 83c7b9bd and is the change behind the release, if you track the repo's history rather than its tags.

$ ls interpretability/reviews/ # the step-back reads

the week in review SAELens 6.54.5 fixes silent feature-density skips
Full archive → Weekly reviews →

Quick answers

What shipped in Interpretability on October 19, 2026?
SAELens shipped 6.54.6, closing a gap where reconstruction metrics ignored the model options you passed to run_evals while sparsity and variance evaluation honored them. In total, 2 commits, 1 pull requests, and 1 releases landed.
Who contributed to Interpretability on October 19, 2026?
1 developer shipped this update, including AaravMehta-07.
What were the notable Interpretability updates?
6.54.6, fix: forward model kwargs to reconstruction evaluation, and fix: forward model kwargs to reconstruction evaluation (#765).
♥

$ log Interpretability --recent

TransformerLens 4.3.0 lands with fixes for J-space projections, ablation, and bridge block discovery 10 shipped Remote runs keep their attention implementation, and nnsight keys now carry trust_remote_code 51 shipped nnsight 0.8.0 cuts import time from 7.6s to 1.6s 3 shipped nnsight 0.8.0 cuts import time, adds a left-padding fix 7 shipped

$ log --all # elsewhere on the wire

Alignment & Post-training Open-Instruct moves to Miles v0.1.2; verl fixes teacher top-k packing under CP Agent Security Windows push keys survive uninstall, and managed installs can skip absent agents

$ ls showcases/

Safety Evals AI Red Teaming Agent Security Alignment & Post-training

$ init your-desk

RepoJournal can write one of these for any repo you care about. Sign in with GitHub, pick your repos and sources, and your desk gets a wire like this one every morning.