release: 0.60.0

CHANGELOG.md

@@ -11,6 +11,16 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+## [0.60.0] — 2026-10-15
+
+**Exact GELU from the engine.** Against engine **0.59.0**, which makes `erf` a first-class op and
+gives `gelu` an `approximate` flag carried as op metadata. Laya's decision models now run their
+exact GELU through `gelu(approximate = false)` instead of a host-side loop per FFN per layer, so the
+activation is a regular op on the compiled path (graph executor, backward rule, StableHLO `chlo.erf`
+or `skainet.gelu` composite) and the multilingual predict fixtures still reproduce (55 values,
+max |diff| 4.99e-5). The BOM re-exports `skainet-bom:0.59.0`. The SKEEP-006 deprecation schedule
+moves by one more minor (see *Deprecated*).
+
 ### Changed
 
 - **Engine 0.59.0.** The version catalog moves `sk.ainet.core:*` to 0.59.0.
@@ -21,6 +31,17 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
   the 22-layer encoder no longer pays 44 host round trips per decision. `LayaGelu.geluErf` /
   `erf` stay as the double-precision reference the engine kernel is checked against.
 
+### Deprecated
+
+- **Schedule moved by one more minor.** The SKEEP-006 legacy surface (`GGUFTokenizer` and its
+  strategies, `TokenizerType`, `SentencePieceSpecialTokens`, `TekkenTokenizerAdapter`, `ModelRegistry`,
+  the `sk.ainet.apps.kllama.GGUFTokenizer` alias) stays at WARNING in 0.60, goes to ERROR in 0.61
+  and leaves in 0.62; `TokenizerFactory` and the `sk.ainet.apps.llm.Tokenizer` sub-interface are
+  deprecated in 0.61 and removed in 0.62. The runtimes, CLIs and benchmarks still reach the engine
+  through the factory (about thirty main-source files); that migration is a release of its own and
+  did not fit next to the engine bump. This matches the engine, whose io-core aliases also stay at
+  WARNING in 0.59.
+
 ## [0.59.0] — 2026-10-13
 
 **Typed decisions.** Three new modules on the JVM path, against engine **0.58.0**: the

README.md

@@ -113,7 +113,11 @@ Honest status — see the project-status note at the top of this README.
 
 ## Current release
 
-The current release is **0.59.0**, against **SKaiNET 0.58.0**.
+The current release is **0.60.0**, against **SKaiNET 0.59.0**.
+
+**Exact GELU from the engine.** SKaiNET 0.59.0 makes `erf` a first-class op and gives `gelu` an `approximate` flag that travels as op metadata through the graph executor, the backward rule and the StableHLO export. Laya's decision models now run their exact GELU through `gelu(approximate = false)` instead of a host-side loop per FFN per layer, and the multilingual predict fixtures still reproduce (55 values, max |diff| 4.99e-5). The SKEEP-006 legacy tokenizer surface stays at WARNING in 0.60; the schedule moves by one more minor (ERROR in 0.61, gone in 0.62).
+
+It builds on **0.59.0**, against **SKaiNET 0.58.0**:
 
 **Typed decisions, not text.** Three new modules bring decision models to the JVM path. `skainet-transformers-decisions-api` (`sk.ainet.decisions`, no engine dependency) is the contract: a `DecisionRequest` carries a state and up to 64 questions of three primitives — `choice` (one option out of a list), `score` (an ordinal scale), `boolean probability` — and every question ends in `Accepted`, `Abstained(reason, candidate)` or `Failed(error)`, never a silent guess; the caller owns the `AcceptancePolicy`. `skainet-transformers-decisions-runtime` validates a request against the adapter's capabilities and applies the policy. `skainet-transformers-inference-laya` is the first adapter: the Laya multilingual checkpoint (mmBERT-base encoder, decision head, Apache 2.0, Convai Innovations) on the FP32 CPU path with exact GELU, verified against the fixtures shipped with the checkpoint (26/26 packed sequences, a 40-text tokenizer corpus, predictions within 5e-5). The Kotlin DSL — `decisionSchema { }` declares questions, policy and model once and stamps out a request per state; `decisionRequest { }`, `acceptancePolicy { }`; `choice(id)` / `score(id)` / `boolean(id)` readback — follows the NN and data pipeline DSLs. Eager FP32 is seconds per decision on a laptop; the compiled and quantized lanes, batching across questions (#490) and an exact GELU op in the engine (SKaiNET#1342) are the open items. `HuggingFaceBPETokenizer` is removed; the remaining SKEEP-006 deprecations stay at WARNING and leave in 0.61.
 
@@ -287,7 +291,7 @@ The recommended way to consume is via the BOM. It pins every published `skainet-
 
 ```kotlin
 dependencies {
-    implementation(platform("sk.ainet.transformers:skainet-transformers-bom:0.59.0"))
+    implementation(platform("sk.ainet.transformers:skainet-transformers-bom:0.60.0"))
 
     // Versions resolved from the BOM:
     implementation("sk.ainet.transformers:skainet-transformers-core")

docs/modules/ROOT/pages/how-to/export-moonshine-v2-streaming.adoc

@@ -32,7 +32,7 @@ From your own build, resolve the published module and run its main class:
 ----
 val exportTool by configurations.creating
 dependencies {
-    exportTool(platform("sk.ainet.transformers:skainet-transformers-bom:0.59.0"))
+    exportTool(platform("sk.ainet.transformers:skainet-transformers-bom:0.60.0"))
     exportTool("sk.ainet.transformers:skainet-transformers-inference-moonshine")
 }
 tasks.register<JavaExec>("exportMoonshine") {

docs/modules/ROOT/pages/reference/moonshine-encoder.adoc

@@ -24,7 +24,7 @@ a DSL decoder is future work.
 [source,kotlin]
 ----
 dependencies {
-    implementation(platform("sk.ainet.transformers:skainet-transformers-bom:0.59.0"))
+    implementation(platform("sk.ainet.transformers:skainet-transformers-bom:0.60.0"))
     implementation("sk.ainet.transformers:skainet-transformers-inference-moonshine")
 }
 ----

docs/modules/ROOT/pages/reference/tokenizer-api.adoc

@@ -87,6 +87,7 @@ descriptor with `+`. `GGUFModelInfo.identity` is the string identity
 |===
 
 `HuggingFaceBPETokenizer` (no callers since 0.58) was removed in 0.59.0. The schedule for the
-rest moved by one minor in 0.59.0: the legacy classes stay at WARNING through 0.59, go to ERROR
-in 0.60 and leave in 0.61, because the runtimes, CLIs and benchmarks still reach the engine
-through `TokenizerFactory` and the sub-interface, and moving them is a release of its own.
+rest moved by one minor in 0.59.0 and by one more in 0.60.0: the legacy classes stay at WARNING
+through 0.60, go to ERROR in 0.61 and leave in 0.62, because the runtimes, CLIs and benchmarks
+still reach the engine through `TokenizerFactory` and the sub-interface, and moving them is a
+release of its own.

docs/modules/ROOT/pages/reference/typed-decisions.adoc

@@ -12,7 +12,7 @@ below put that on the JVM path with the same contract for every model.
 [source,kotlin]
 ----
 dependencies {
-    implementation(platform("sk.ainet.transformers:skainet-transformers-bom:0.59.0"))
+    implementation(platform("sk.ainet.transformers:skainet-transformers-bom:0.60.0"))
     implementation("sk.ainet.transformers:skainet-transformers-decisions-api")      // the contract, no engine dependency
     implementation("sk.ainet.transformers:skainet-transformers-decisions-runtime")  // validation + acceptance policy
     implementation("sk.ainet.transformers:skainet-transformers-inference-laya")     // Laya as a DecisionAdapter

docs/modules/ROOT/pages/tutorials/android-getting-started.adoc

@@ -55,7 +55,7 @@ dependencies {
     // self-registers it on ART at process start — nothing to call.
     runtimeOnly("sk.ainet.core:skainet-backend-jni-cpu")
 
-    implementation(platform("sk.ainet.transformers:skainet-transformers-bom:0.59.0"))
+    implementation(platform("sk.ainet.transformers:skainet-transformers-bom:0.60.0"))
     implementation("sk.ainet.transformers:skainet-transformers-core")
     implementation("sk.ainet.transformers:skainet-transformers-runtime-kllama")
     implementation("sk.ainet.transformers:skainet-transformers-inference-llama")

docs/modules/ROOT/pages/tutorials/getting-started-java.adoc

@@ -25,7 +25,7 @@ In your `build.gradle.kts`:
 [source,kotlin]
 ----
 dependencies {
-    implementation(platform("sk.ainet.transformers:skainet-transformers-bom:0.59.0"))
+    implementation(platform("sk.ainet.transformers:skainet-transformers-bom:0.60.0"))
 
     implementation("sk.ainet.transformers:skainet-transformers-runtime-kllama")
     implementation("sk.ainet.transformers:skainet-transformers-agent")
@@ -41,7 +41,7 @@ Or in Maven (Maven needs the `-jvm` classifier suffix on platform artifacts):
     <dependency>
       <groupId>sk.ainet.transformers</groupId>
       <artifactId>skainet-transformers-bom</artifactId>
-      <version>0.59.0</version>
+      <version>0.60.0</version>
       <type>pom</type>
       <scope>import</scope>
     </dependency>

docs/modules/ROOT/pages/tutorials/getting-started-leaf.adoc

@@ -34,7 +34,7 @@ encoder output to the advertised dimensionality. The runtime applies it automati
 [source,kotlin]
 ----
 dependencies {
-    implementation(platform("sk.ainet.transformers:skainet-transformers-bom:0.59.0"))
+    implementation(platform("sk.ainet.transformers:skainet-transformers-bom:0.60.0"))
     implementation("sk.ainet.transformers:skainet-transformers-providers")
 }
 ----

docs/modules/ROOT/pages/tutorials/llama3-tool-calling.adoc

@@ -52,7 +52,7 @@ The pieces you need live in three modules:
 [source,kotlin]
 ----
 dependencies {
-    implementation(platform("sk.ainet.transformers:skainet-transformers-bom:0.59.0"))
+    implementation(platform("sk.ainet.transformers:skainet-transformers-bom:0.60.0"))
 
     implementation("sk.ainet.transformers:skainet-transformers-runtime-kllama")
     implementation("sk.ainet.transformers:skainet-transformers-agent")

gradle.properties

@@ -1,5 +1,5 @@
 GROUP=sk.ainet.transformers
-VERSION_NAME=0.59.0
+VERSION_NAME=0.60.0
 
 POM_DESCRIPTION=SKaiNET-transformers