bump: version 2.0.0b27 → 2.0.0b28

Author: github-actions[bot]

CHANGELOG.md

@@ -1,3 +1,21 @@
+## v2.0.0b28 (2026-04-14)
+
+
+- Merge pull request #212 from OpenMOSS/dev
+- Release: fix and refactor QK tracing
+- fix(serialize): migrate decorator must target staticmethod's underlying func
+- fix(circuits): drop k_side transpose in slot_pair_attribution, guard empty side
+- refactor(circuits): extract attribution_scores helper, add flat_topk, rewrite qk pair dedup via zero-out
+- chore(server): bf16 dtype and explicit device_mesh in model/SAE loaders
+- fix(circuits): correct QK tracing scale, target post-LN/rotary Q/K, dedup pairs
+- feat(circuits): QKTracingResult bundle with Q/K marginals and merged top-k picks
+- feat(circuits): role-labeled QK pair attribution via VJP with K as cotangent
+- chore: add .playwright-mcp to .gitignore
+- fix(circuits): split bias leaves around apply_saes so SAE sees correct tensor
+- refactor(circuits): reduce intermediate dimension variables in attribution
+- fix(backend): align interfaces of model.qk_trace and qk_trace
+- refactor(circuits): replace NodeInfoRef with collection type NodeRefs
+
 ## v2.0.0b27 (2026-04-12)
 
 

README.md

@@ -35,13 +35,13 @@
 Use [pip](https://pypi.org/project/pip/) to install Language-Model-SAEs:
 
 ```bash
-pip install lm-saes==2.0.0b27
+pip install lm-saes==2.0.0b28
 ```
 
 We also highly recommend using [uv](https://docs.astral.sh/uv/) to manage your own project dependencies. You can use
 
 ```bash
-uv add lm-saes==2.0.0b27
+uv add lm-saes==2.0.0b28
 ```
 
 to add Language-Model-SAEs as your project dependency.

docs/index.md

@@ -24,7 +24,7 @@ This library provides:
     To add our library as a project dependency, run:
 
     ```bash
-    uv add lm-saes==2.0.0b27
+    uv add lm-saes==2.0.0b28
     ```
 
     We also support [Ascend NPU](https://github.com/Ascend/pytorch) as an accelerator backend. To add our library as a project dependency with NPU dependency constraints, run:
@@ -38,7 +38,7 @@ This library provides:
     Of course, you can also directly use [pip](https://pypi.org/project/pip/) to install our library. To install our library with pip, run:
 
     ```bash
-    pip install lm-saes==2.0.0b27
+    pip install lm-saes==2.0.0b28
     ```
 
     We also support [Ascend NPU](https://github.com/Ascend/pytorch) as an accelerator backend. To install our library with NPU dependency constraints, run:

pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "lm-saes"
-version = "2.0.0b27"
+version = "2.0.0b28"
 description = "For OpenMOSS Mechanistic Interpretability Team's Sparse Autoencoder (SAE) research. Open-sourced and constantly updated."
 dependencies = [
     "transformer-lens>=2.16.2",