Prepare 0.19.8 release

CHANGELOG.md

@@ -1,11 +1,12 @@
 Releases before 0.10.0 are in the
 [changelog archive](https://github.com/leehack/llamadart/blob/main/doc/changelog_archive.md).
 
-## Unreleased
+## 0.19.8
 
-- Fixed constrained JSON (`responseFormat`) ignoring the grammar after a `ThinkingBudget` closes a template-opened thought on native llama.cpp (Qwen3.5 and similar), and the same for required tool calls on templates whose grammar takes that thought.
-- Fixed a required tool call on Seed-OSS being unconstrained after a `ThinkingBudget` closes a thought the model opened, on native llama.cpp ([#1958](https://github.com/leehack/llamadart/issues/1958)).
-- Aligned the default WebGPU bridge assets to `v0.1.84`: constrained JSON (`responseFormat`) and required tool calls that take the thought stay constrained after a `ThinkingBudget` closes it on WebGPU, and older assets throw `LlamaUnsupportedException` for that combination ([#1954](https://github.com/leehack/llamadart/issues/1954)). Web keeps native llama.cpp `v0.9.0` parity (upstream `v0.9.0@cd959bae99ff4bed22bef2604f45241c09bbb8c8`) and `@litert-lm/[email protected]`. Manifest SHA-256: `55abeb377597148156059a8ee5400f2314c6254ad91b57db65b709e304c0dea0`.
+- Fixed constrained JSON (`responseFormat`) and required tool calls losing their grammar after a `ThinkingBudget` closes a thought on native llama.cpp: an eager grammar that takes the forced-open thought (Qwen3.5 and similar, and Seed-OSS required calls) now stays applied, and every other grammar pauses until reasoning ends ([#1953](https://github.com/leehack/llamadart/issues/1953), [#1958](https://github.com/leehack/llamadart/issues/1958)).
+- Aligned the default WebGPU bridge assets to `v0.1.84`, so WebGPU keeps those grammars applied too; older assets throw `LlamaUnsupportedException` for that combination ([#1954](https://github.com/leehack/llamadart/issues/1954)).
+
+Runtime baseline: native llama.cpp `v0.9.0` and Web use upstream `v0.9.0@cd959bae99ff4bed22bef2604f45241c09bbb8c8`; native LiteRT-LM uses `v0.19.1-4` and stable-diffusion uses `v0.3.0-3`. Web LiteRT-LM uses `@litert-lm/[email protected]`. Manifest SHA-256: `55abeb377597148156059a8ee5400f2314c6254ad91b57db65b709e304c0dea0`.
 
 ## 0.19.7
 

README.md

@@ -35,7 +35,7 @@ LFM2 replies cut off at a reported length limit expose raw `partialToolCalls` di
 
 ```yaml
 dependencies:
-  llamadart: ^0.19.7
+  llamadart: ^0.19.8
 ```
 
 ```bash
@@ -56,10 +56,10 @@ companion shares its core's `MAJOR.MINOR` and constrains `llamadart` to it.
 
 ```yaml
 dependencies:
-  llamadart: ^0.19.7
-  llamadart_llama_cpp_flutter: ^0.19.7 # GGUF / llama.cpp
-  llamadart_litert_lm_flutter: ^0.19.7 # Apple .litertlm / LiteRT-LM targets
-  llamadart_stable_diffusion_flutter: ^0.19.7 # Apple image generation, opt-in
+  llamadart: ^0.19.8
+  llamadart_llama_cpp_flutter: ^0.19.8 # GGUF / llama.cpp
+  llamadart_litert_lm_flutter: ^0.19.8 # Apple .litertlm / LiteRT-LM targets
+  llamadart_stable_diffusion_flutter: ^0.19.8 # Apple image generation, opt-in
 ```
 
 Pairing rules, the SwiftPM bootstrap and the privacy manifest are in

example/chat_app/pubspec.lock

@@ -421,28 +421,28 @@ packages:
       path: "../.."
       relative: true
     source: path
-    version: "0.19.7"
+    version: "0.19.8"
   llamadart_litert_lm_flutter:
     dependency: "direct main"
     description:
       path: "../../packages/llamadart_litert_lm_flutter"
       relative: true
     source: path
-    version: "0.19.7"
+    version: "0.19.8"
   llamadart_llama_cpp_flutter:
     dependency: "direct main"
     description:
       path: "../../packages/llamadart_llama_cpp_flutter"
       relative: true
     source: path
-    version: "0.19.7"
+    version: "0.19.8"
   llamadart_stable_diffusion_flutter:
     dependency: "direct main"
     description:
       path: "../../packages/llamadart_stable_diffusion_flutter"
       relative: true
     source: path
-    version: "0.19.7"
+    version: "0.19.8"
   llamadart_validation:
     dependency: "direct main"
     description:

packages/llamadart_litert_lm_flutter/CHANGELOG.md

@@ -1,3 +1,8 @@
+## 0.19.8
+
+- Refresh README installation guidance for core `0.19.8`.
+- Pair with core `>=0.19.8 <0.20.0`; Apple SwiftPM stays pinned to `leehack/[email protected]`.
+
 ## 0.19.7
 
 - Refresh README installation guidance for core `0.19.7`.

packages/llamadart_litert_lm_flutter/README.md

@@ -27,8 +27,8 @@ also request an x86_64 Simulator slice must exclude x86_64 for LiteRT-LM builds.
 
 ```yaml
 dependencies:
-  llamadart: ^0.19.7
-  llamadart_litert_lm_flutter: ^0.19.7
+  llamadart: ^0.19.8
+  llamadart_litert_lm_flutter: ^0.19.8
 ```
 
 This package has no runtime Dart API of its own. Import `package:llamadart`

packages/llamadart_litert_lm_flutter/pubspec.yaml

@@ -1,6 +1,6 @@
 name: llamadart_litert_lm_flutter
 description: Flutter Apple SwiftPM runtime companion package for llamadart LiteRT-LM support.
-version: 0.19.7
+version: 0.19.8
 repository: https://github.com/leehack/llamadart/tree/main/packages/llamadart_litert_lm_flutter
 homepage: https://github.com/leehack/llamadart
 issue_tracker: https://github.com/leehack/llamadart/issues
@@ -17,7 +17,7 @@ environment:
 dependencies:
   flutter:
     sdk: flutter
-  llamadart: '>=0.19.7 <0.20.0'
+  llamadart: '>=0.19.8 <0.20.0'
 
 dev_dependencies:
   flutter_test:

packages/llamadart_llama_cpp_flutter/CHANGELOG.md

@@ -1,3 +1,7 @@
+## 0.19.8
+
+- Refresh README installation guidance for core `0.19.8`; Apple SwiftPM stays pinned to `leehack/[email protected]`.
+
 ## 0.19.7
 
 - Refresh README installation guidance for core `0.19.7`; Apple SwiftPM stays pinned to `leehack/[email protected]`.

packages/llamadart_llama_cpp_flutter/README.md

@@ -13,8 +13,8 @@ with core `0.19.x`. Upgrade both together. Earlier pairings are in the
 
 ```yaml
 dependencies:
-  llamadart: ^0.19.7
-  llamadart_llama_cpp_flutter: ^0.19.7
+  llamadart: ^0.19.8
+  llamadart_llama_cpp_flutter: ^0.19.8
 ```
 
 This package has no runtime Dart API of its own. Import `package:llamadart`

packages/llamadart_llama_cpp_flutter/pubspec.yaml

@@ -1,6 +1,6 @@
 name: llamadart_llama_cpp_flutter
 description: Flutter Apple SwiftPM runtime companion package for llamadart llama.cpp and GGUF support.
-version: 0.19.7
+version: 0.19.8
 repository: https://github.com/leehack/llamadart/tree/main/packages/llamadart_llama_cpp_flutter
 homepage: https://github.com/leehack/llamadart
 issue_tracker: https://github.com/leehack/llamadart/issues

packages/llamadart_stable_diffusion_flutter/CHANGELOG.md

@@ -1,3 +1,7 @@
+## 0.19.8
+
+- Refresh README installation guidance for core `0.19.8`; Apple SwiftPM stays pinned to `leehack/[email protected]`.
+
 ## 0.19.7
 
 - Refresh README installation guidance for core `0.19.7`; Apple SwiftPM stays pinned to `leehack/[email protected]`.

packages/llamadart_stable_diffusion_flutter/README.md

@@ -19,8 +19,8 @@ with core `0.19.x`. Upgrade both together. Earlier pairings are in the
 
 ```yaml
 dependencies:
-  llamadart: ^0.19.7
-  llamadart_stable_diffusion_flutter: ^0.19.7
+  llamadart: ^0.19.8
+  llamadart_stable_diffusion_flutter: ^0.19.8
 ```
 
 Adding the package selects the `stable_diffusion` runtime for Flutter iOS and

packages/llamadart_stable_diffusion_flutter/pubspec.yaml

@@ -1,6 +1,6 @@
 name: llamadart_stable_diffusion_flutter
 description: Flutter Apple SwiftPM runtime companion package for llamadart stable-diffusion.cpp image generation.
-version: 0.19.7
+version: 0.19.8
 repository: https://github.com/leehack/llamadart/tree/main/packages/llamadart_stable_diffusion_flutter
 homepage: https://github.com/leehack/llamadart
 issue_tracker: https://github.com/leehack/llamadart/issues

packages/llamadart_validation/pubspec.lock

@@ -215,7 +215,7 @@ packages:
       path: "../.."
       relative: true
     source: path
-    version: "0.19.7"
+    version: "0.19.8"
   logging:
     dependency: transitive
     description:

pubspec.yaml

@@ -1,6 +1,6 @@
 name: llamadart
 description: Cross-platform, on-device LLM inference for Dart and Flutter. Run GGUF (llama.cpp) and LiteRT-LM models locally on Android, iOS, macOS, Windows, Linux and web.
-version: 0.19.7
+version: 0.19.8
 homepage: https://llamadart.leehack.com
 repository: https://github.com/leehack/llamadart
 issue_tracker: https://github.com/leehack/llamadart/issues

website/docs/changelog/recent-releases.md

@@ -11,11 +11,12 @@ For upgrade planning, combine this page with the
 [Upgrade Checklist](../migration/upgrade-checklist). For breaking changes,
 validate against the exact release tag notes.
 
-## Unreleased
+## 0.19.8
 
-- Fixed constrained JSON (`responseFormat`) ignoring the grammar after a `ThinkingBudget` closes a template-opened thought on native llama.cpp (Qwen3.5 and similar), and the same for required tool calls on templates whose grammar takes that thought.
-- Fixed a required tool call on Seed-OSS being unconstrained after a `ThinkingBudget` closes a thought the model opened, on native llama.cpp ([#1958](https://github.com/leehack/llamadart/issues/1958)).
-- Aligned default WebGPU bridge assets to `v0.1.84`: constrained JSON (`responseFormat`) and required tool calls that take the thought stay constrained after a `ThinkingBudget` closes it on WebGPU, and older assets throw `LlamaUnsupportedException` for that combination ([#1954](https://github.com/leehack/llamadart/issues/1954)). Web keeps native llama.cpp `v0.9.0` parity (upstream `v0.9.0@cd959bae99ff4bed22bef2604f45241c09bbb8c8`) and `@litert-lm/[email protected]`. Manifest SHA-256: `55abeb377597148156059a8ee5400f2314c6254ad91b57db65b709e304c0dea0`.
+- Fixed constrained JSON (`responseFormat`) and required tool calls losing their grammar after a `ThinkingBudget` closes a thought on native llama.cpp: an eager grammar that takes the forced-open thought (Qwen3.5 and similar, and Seed-OSS required calls) now stays applied, and every other grammar pauses until reasoning ends ([#1953](https://github.com/leehack/llamadart/issues/1953), [#1958](https://github.com/leehack/llamadart/issues/1958)).
+- Aligned default WebGPU bridge assets to `v0.1.84`, so WebGPU keeps those grammars applied too; older assets throw `LlamaUnsupportedException` for that combination ([#1954](https://github.com/leehack/llamadart/issues/1954)).
+
+Runtime baseline: native llama.cpp `v0.9.0` and Web use upstream `v0.9.0@cd959bae99ff4bed22bef2604f45241c09bbb8c8`; native LiteRT-LM uses `v0.19.1-4` and stable-diffusion uses `v0.3.0-3`. Web LiteRT-LM uses `@litert-lm/[email protected]`. Manifest SHA-256: `55abeb377597148156059a8ee5400f2314c6254ad91b57db65b709e304c0dea0`.
 
 ## 0.19.7
 

website/docs/getting-started/installation.md

@@ -23,7 +23,7 @@ For Genkit or Flutter Edge AI integration setup, see
 
 ```yaml
 dependencies:
-  llamadart: ^0.19.7
+  llamadart: ^0.19.8
 ```
 
 Then resolve packages:
@@ -42,10 +42,10 @@ Apple XCFrameworks:
 
 ```yaml
 dependencies:
-  llamadart: ^0.19.7
-  llamadart_llama_cpp_flutter: ^0.19.7 # GGUF / llama.cpp
-  llamadart_litert_lm_flutter: ^0.19.7 # Apple .litertlm / LiteRT-LM targets
-  llamadart_stable_diffusion_flutter: ^0.19.7 # Apple image generation, opt-in
+  llamadart: ^0.19.8
+  llamadart_llama_cpp_flutter: ^0.19.8 # GGUF / llama.cpp
+  llamadart_litert_lm_flutter: ^0.19.8 # Apple .litertlm / LiteRT-LM targets
+  llamadart_stable_diffusion_flutter: ^0.19.8 # Apple image generation, opt-in
 ```
 
 Companions use the core's `MAJOR.MINOR`; pub refuses a mismatched pair.

website/examples/observability/pubspec.lock

@@ -271,7 +271,7 @@ packages:
       path: "../../.."
       relative: true
     source: path
-    version: "0.19.7"
+    version: "0.19.8"
   logging:
     dependency: transitive
     description: