publish 0.3.1 (#39)

CHANGELOG

@@ -2,6 +2,17 @@
 
 ## Unreleased
 
+## 0.3.1 - 2024-07-01
+
+* Add `method="pca_center"` option for training vectors with a more accurate method. (#29 / #34)
+  * Thanks to @r3ndd ! :tada:
+  * (The old method, `"pca_diff"`, remains the default.)
+  * Also adds the undocumented / experimental `method="umap"`.
+* Adds `ControlVector.import_gguf(filename)` as a peer to `export_gguf`. (#34)
+* Fixes a bug with training vectors on right-padded models, such as llama-3-*. (#38)
+  * Thanks to @ohxh ! :tada:
+* (0.3.0 was a botched release, has been yanked)
+
 ## 0.2.2 - 2024-03-09
 
 * Fix a bug in control.py (#18)

README.md

@@ -65,6 +65,12 @@ for strength in (-2.2, 1, 2.2):
 
 For a more detailed explanation of how the library works and what it can do, see [the blog post](https://vgel.me/posts/representation-engineering).
 
+## Notes
+
+* For a list of changes by version, see the [CHANGELOG](https://github.com/vgel/repeng/blob/main/CHANGELOG).
+* For quantized use, you may be interested in [llama.cpp#5970](https://github.com/ggerganov/llama.cpp/pull/5970)—after training a vector with `repeng`, export it by calling `vector.export_gguf(filename)` and then use it in `llama.cpp` with any quant!
+* Vector training *currently does not work* with MoE models (such as Mixtral). (This is theoretically fixable with some work, let me know if you're interested.)
+
 ## Notice
 
 Some of the code in this repository derives from [andyzoujm/representation-engineering](https://github.com/andyzoujm/representation-engineering) (MIT license).

pyproject.toml

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "repeng"
-version = "0.2.2"
+version = "0.3.1"
 description = "representation engineering / control vectors"
 authors = ["Theia Vogel <[email protected]>"]
 readme = "README.md"