release: 0.33.0 — ship against SKaiNET engine 0.33.0

CHANGELOG.md

@@ -7,6 +7,20 @@ version line is kept in lock-step with the underlying SKaiNET engine
 The format roughly follows [Keep a Changelog](https://keepachangelog.com/en/1.1.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [0.33.0] — 2026-06-29
+
+Ships against **SKaiNET engine 0.33.0**. No transformers API changes — this release adopts the new
+engine line and routine dependency updates.
+
+### Changed
+
+- **Engine → 0.33.0.** Transformer models authored with this layer inherit the engine's 0.33.0 work;
+  most relevant here, `layerNorm` / `rmsNorm` now lower to real `stablehlo.reduce`, so transformer
+  exports compile and run on stock IREE (engine #769). The engine also fixes a silent autodiff
+  gradient-drop (`elu`/`leakyRelu`/`permute`) and adds new differentiable ops (`cos`/`sin`/`gather`/…),
+  available to model authors. (engine 0.33.0)
+- **Dependencies:** Ktor client `3.5.1` (#198), Logback `1.5.36` (#199).
+
 ## [0.32.1] — 2026-06-26
 
 Fixes streaming detokenization — generated text no longer runs words together

README.md

@@ -103,10 +103,12 @@ Honest status — see the project-status note at the top of this README.
 
 ## Current release
 
-The current release is **0.32.1** (against **SKaiNET 0.32.4**). It fixes streaming
-detokenization — per-token decode now keeps each word's leading space, so generated text
-no longer runs together (`"the process"` not `"theprocess"`). On top of the **0.32.0**
-real-GGUF **Llama** eager + StableHLO/IREE export work:
+The current release is **0.33.0** (against **SKaiNET 0.33.0**). It adopts the new engine line —
+no transformers API changes — so transformer models inherit the engine's 0.33.0 work: `layerNorm`/
+`rmsNorm` now lower to real `stablehlo.reduce` (exports compile on stock IREE), a silent autodiff
+gradient-drop is fixed, and new differentiable ops (`cos`/`sin`/`gather`/…) are available. On top of
+the **0.32.1** streaming-detokenization fix and the **0.32.0** real-GGUF **Llama** eager +
+StableHLO/IREE export work:
 
 - The eager **`NATIVE_OPTIMIZED` path now works for Llama** (`Q4_K`/`Q6_K`): weights stay
   packed and `LlamaNetworkLoader.fromGguf(NATIVE_OPTIMIZED) + OptimizedLLMRuntime` decodes

gradle.properties

@@ -1,5 +1,5 @@
 GROUP=sk.ainet.transformers
-VERSION_NAME=0.32.1
+VERSION_NAME=0.33.0
 
 POM_DESCRIPTION=SKaiNET-transformers
 

gradle/libs.versions.toml

@@ -1,5 +1,5 @@
 [versions]
-skainet = "0.32.4"
+skainet = "0.33.0"
 agp = "9.2.1"
 jacksonDatabind = "2.22.0"
 jsonSchemaValidator = "3.0.5"