SKaiNET-developers · michalharakal · Sep 24, 2026 · Sep 24, 2026 · Sep 24, 2026 · Sep 24, 2026
diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -2,6 +2,16 @@
 
 ## [Unreleased]
 
+## [0.57.0] - 2026-09-24
+
+Headline: **grouped-query KV-cache graphs export with an attention mask, and the toolchain moves to
+Kotlin 2.4.20.** The StableHLO attention converter now emits valid IR for an explicit mask under
+grouped-query attention and onto a dynamic key length, the shape of a chunked-prefill graph over a
+KV cache. Static graphs emit exactly what they did in 0.56.0.
+
+Version note: SKaiNET-transformers 0.57.0 follows on this engine release, keeping the two on one
+version line.
+
 ### Fixed
 
 - **SDPA StableHLO export: explicit masks under grouped-query attention and onto a dynamic key
@@ -14,6 +24,25 @@
   `known_nonexpanding_dimensions`, the form IREE 3.11 lowers. Static graphs are unchanged. Note:
   IREE 3.11 does not lower `dynamic_reshape`, so for IREE prefer a head-shared `[b, 1, Sq, ?]` mask
   when the key length is dynamic.
+- **`skainet-data-source` exports kotlinx-io as `api`** (#1295). `DataSourceArtifact.openSource()` /
+  `copyTo()`, `DataSourceRemoteContent` and `DataSourceArtifactStore` expose `kotlinx.io.Source` and
+  `Sink`, but kotlinx-io was an implementation dependency (runtime scope in the published POM), so
+  consumers failed with "Cannot access class kotlinx.io.Source" unless they declared it themselves.
+
+### Changed
+
+- **Kotlin 2.4.20** (#1304; supersedes #1271, #1272, #1273), with the Kotlin/JS `yarn.lock`
+  regenerated for the newer JS toolchain.
+- **Security pins**: the `webpack` pin is dropped. Kotlin 2.4.20's Gradle plugin requests webpack
+  5.108.1, already past 5.104.1, the first release clearing GHSA-8fgc-7cc6-rx7x and
+  GHSA-38r7-794h-5758, so the pin only held the bundler behind the version the plugin is tested
+  with. All other npm pins are unchanged.
+- **Dependencies**: AGP 9.4.1 (#1299), Ktor client 3.6.0 (#1296), kctfork 0.14.0 (#1300).
+- **CI**: the publish workflow gains a manual dry run (`workflow_dispatch`): the full release
+  pipeline into Maven Local, never given the Maven Central credentials (#1294). `GITFLOW.adoc`
+  documents the dry-run and green-before-tag gates. Action bumps: `actions/deploy-pages` 5.0.1
+  (#1263), `android-actions/setup-android` 4.0.4 (#1298), `github/codeql-action/upload-sarif`
+  4.38.1 (#1297).
 
 ## [0.56.0] - 2026-09-20
 

diff --git a/README.md b/README.md
@@ -51,7 +51,7 @@ Add the core dependencies (Gradle Kotlin DSL):
 ```kotlin
 dependencies {
     // Recommended: import the umbrella BOM and drop versions on the engine modules.
-    implementation(platform("sk.ainet:skainet-bom:0.56.0"))
+    implementation(platform("sk.ainet:skainet-bom:0.57.0"))
 
     implementation("sk.ainet.core:skainet-lang-core")
     implementation("sk.ainet.core:skainet-backend-cpu")
@@ -308,24 +308,18 @@ val withoutLabel = dataPipeline<RawDataset>()
 
 ---
 
-## What's New in 0.56.0
+## What's New in 0.57.0
 
-Grouped-query attention becomes native to the engine, and the compiled leg of SKEEP-005 lands:
+Grouped-query KV-cache graphs export with an attention mask, and the toolchain moves to Kotlin 2.4.20:
 
-- **Grouped-query `scaledDotProductAttention`** — K/V arrive with their own head count
-  (`[b, nKV, Sk, hd]`); query head `h` reads K/V head `h / (H / nKV)`. Bit-identical to before when
-  `nKV == H`, and to the old tiled form otherwise. Model code no longer expands K/V heads before
-  attention — not eagerly, not on the tape.
-- **Leaner StableHLO for GQA models** — attention lowers with the head groups as a batching
-  dimension, so exports carry no broadcast or concatenate of K/V.
-- **Structure at compile time, cores at run time** — `ScheduleAnnotationPass` stamps structural
-  defaults (`parallel_dims = [batch, heads]` on every attention); an explicit `parallelism` is
-  advisory and a default can never carry a core count.
-- **Graph contexts honour the schedule of their ops** — the new `ScheduledOps` seam lets
-  `DefaultGraphExecutionContext` answer `schedule`/`withSchedule` from the ops it wraps, so the JVM
-  `ComputeGraphExecutor` runs under the caller's schedule.
-
-0.55.0 is skipped so the engine and SKaiNET-transformers share a version line again.
+- **Masked grouped-query attention exports to valid StableHLO** — a per-head mask under GQA is split
+  into the K/V head groups instead of broadcast onto them, and a mask onto a dynamic key length (the
+  chunked-prefill shape of a KV-cache graph) uses `dynamic_broadcast_in_dim` with the expansion hints
+  IREE 3.11 needs. Static graphs emit exactly what they did before.
+- **kotlinx-io is part of `skainet-data-source`'s API** — callers of `openSource()` / `copyTo()` no
+  longer need to declare it themselves.
+- **Kotlin 2.4.20**, AGP 9.4.1, Ktor client 3.6.0; the `webpack` security pin is retired now that
+  the Kotlin toolchain's own webpack is past the fix.
 
 See [CHANGELOG.md](CHANGELOG.md) for full release notes, including every prior release.
 
@@ -351,6 +345,12 @@ We love contributions! Whether it's a new operator, documentation, or a bug fix:
 
 Browse the full codebase documentation on [DeepWiki](https://deepwiki.com/SKaiNET-developers/SKaiNET).
 
+### Contributors (0.57.0)
+
+- **Michal Harakal** ([@michalharakal](https://github.com/michalharakal)) — masked grouped-query
+  attention in the StableHLO export, the kotlinx-io API fix, the Kotlin 2.4.20 upgrade, and the
+  publish dry run
+
 ### Contributors (0.56.0)
 
 - **Michal Harakal** ([@michalharakal](https://github.com/michalharakal)) — grouped-query-native

diff --git a/docs/antora.yml b/docs/antora.yml
@@ -15,7 +15,7 @@ asciidoc:
     framework_name: SKaiNET
     # Current SKaiNET release — bump once per release; referenced as
     # {skainet_version} in dependency snippets (blocks need subs="attributes+").
-    skainet_version: 0.56.0
+    skainet_version: 0.57.0
     ksp_version: 2.2.21-2.0.5
     dokka_version: 2.1.0
     asciidoctorj_version: 3.0.0
diff --git a/docs/modules/ROOT/pages/reference/kernel-support-matrix.adoc b/docs/modules/ROOT/pages/reference/kernel-support-matrix.adoc
@@ -1,7 +1,7 @@
 = Kernel × platform support matrix
 :description: Which compute-kernel provider serves each weight format on each KMP target.
 
-Generated from `kernel-support.json` (version `0.56.0`) by `KernelSupportMatrixTest` — registry introspection of the registered `KernelProvider` implementations. Do not edit by hand; run `./gradlew generateKernelMatrix` to refresh.
+Generated from `kernel-support.json` (version `0.57.0`) by `KernelSupportMatrixTest` — registry introspection of the registered `KernelProvider` implementations. Do not edit by hand; run `./gradlew generateKernelMatrix` to refresh.
 
 Each cell is the best (highest-priority) provider that serves `Float32 × format` `matmul` on that platform: *native-ffm* (100) → *panama-vector* (50) → *scalar* (0). An empty cell (`—`) means no provider carries a kernel there (the format is dequant-to-FP32 only).
 

diff --git a/docs/modules/ROOT/pages/reference/operators/generated/index.adoc b/docs/modules/ROOT/pages/reference/operators/generated/index.adoc
@@ -1,6 +1,6 @@
 = AI-NET Operators Reference
 
-Generated from version `0.56.0` on 2026-09-20
+Generated from version `0.57.0` on 2026-09-24
 
 == Operators by Modality
 

diff --git a/docs/modules/ROOT/pages/reference/ops-status-matrix.adoc b/docs/modules/ROOT/pages/reference/ops-status-matrix.adoc
@@ -1,7 +1,7 @@
 = Operator Coverage Matrix
 :description: Cross-backend status for every operator function in SKaiNET.
 
-Generated from `operators.json` version `0.56.0` on 2026-09-20.
+Generated from `operators.json` version `0.57.0` on 2026-09-24.
 
 Rows are `Operator.function` pairs. The `Validated` column shows whether the function's documentation has been DARC-validated by a reviewer (see xref:contributing/darc-workflow.adoc[DARC workflow]). Remaining columns are backends that appear in any function's `statusByBackend` map — a missing entry means the backend makes no claim about the function (treat it as "unknown", not "not supported").
 

diff --git a/docs/modules/ROOT/pages/tutorials/android-classifier-getting-started.adoc b/docs/modules/ROOT/pages/tutorials/android-classifier-getting-started.adoc
@@ -19,12 +19,12 @@ An Android project with Kotlin. Add the SKaiNET modules:
 [source,kotlin]
 ----
 dependencies {
-    implementation("sk.ainet.core:skainet-lang-core:0.56.0")     // tensors, DSL, training
-    implementation("sk.ainet.core:skainet-backend-cpu:0.56.0")   // CPU ops
-    implementation("sk.ainet.core:skainet-compile-dag:0.56.0")   // autograd (training context)
-    implementation("sk.ainet.core:skainet-data-api:0.56.0")      // Dataset / DataBatch
-    implementation("sk.ainet.core:skainet-data-simple:0.56.0")   // embedded Iris
-    runtimeOnly("sk.ainet.core:skainet-backend-jni-cpu:0.56.0")  // NEON kernels (see below)
+    implementation("sk.ainet.core:skainet-lang-core:0.57.0")     // tensors, DSL, training
+    implementation("sk.ainet.core:skainet-backend-cpu:0.57.0")   // CPU ops
+    implementation("sk.ainet.core:skainet-compile-dag:0.57.0")   // autograd (training context)
+    implementation("sk.ainet.core:skainet-data-api:0.57.0")      // Dataset / DataBatch
+    implementation("sk.ainet.core:skainet-data-simple:0.57.0")   // embedded Iris
+    runtimeOnly("sk.ainet.core:skainet-backend-jni-cpu:0.57.0")  // NEON kernels (see below)
 }
 ----
 

diff --git a/gradle.properties b/gradle.properties
@@ -1,5 +1,5 @@
 GROUP=sk.ainet.core
-VERSION_NAME=0.56.0
+VERSION_NAME=0.57.0
 POM_DESCRIPTION=SKaiNET
 
 POM_URL=https://github.com/SKaiNET-developers/skainet/