Prepare 0.8.3 release

CHANGELOG.md

@@ -1,9 +1,10 @@
-## Unreleased
+## 0.8.3
 
-* Preload Windows backend modules with `LOAD_WITH_ALTERED_SEARCH_PATH` from
-  the resolved native backend bundle directory before handing them to
-  llama.cpp, so CUDA backend discovery can resolve colocated CUDA
-  redistributables without app `PATH` changes.
+* Fixed Windows CUDA backend discovery when the native asset bundle directory is
+  not on the app `PATH`. llama.cpp backend modules are now loaded from their
+  resolved bundle path in a way that lets colocated CUDA redistributables such
+  as `cudart64_12.dll`, `cublas64_12.dll`, and `cublasLt64_12.dll` resolve
+  correctly.
 
 ## 0.8.2
 

README.md

@@ -52,7 +52,7 @@ JavaScript runtime.
 
 ```yaml
 dependencies:
-  llamadart: ^0.8.2
+  llamadart: ^0.8.3
 ```
 
 Flutter iOS/macOS apps that want Swift Package Manager-linked Apple
@@ -61,7 +61,7 @@ they ship:
 
 ```yaml
 dependencies:
-  llamadart: ^0.8.2
+  llamadart: ^0.8.3
   llamadart_llama_cpp_flutter: ^0.0.3 # GGUF / llama.cpp
   llamadart_litert_lm_flutter: ^0.0.2 # .litertlm / LiteRT-LM
 ```

packages/llamadart_litert_lm_flutter/README.md

@@ -10,7 +10,7 @@ core package's native-assets fallback.
 
 ```yaml
 dependencies:
-  llamadart: ^0.8.2
+  llamadart: ^0.8.3
   llamadart_litert_lm_flutter: ^0.0.2
 ```
 

packages/llamadart_llama_cpp_flutter/README.md

@@ -9,7 +9,7 @@ core package's native-assets fallback.
 
 ```yaml
 dependencies:
-  llamadart: ^0.8.2
+  llamadart: ^0.8.3
   llamadart_llama_cpp_flutter: ^0.0.3
 ```
 

pubspec.yaml

@@ -1,6 +1,6 @@
 name: llamadart
 description: Dart and Flutter local LLM inference with llama.cpp GGUF and LiteRT-LM across native platforms and web.
-version: 0.8.2
+version: 0.8.3
 homepage: https://github.com/leehack/llamadart
 repository: https://github.com/leehack/llamadart
 issue_tracker: https://github.com/leehack/llamadart/issues

website/docs/changelog/recent-releases.md

@@ -7,6 +7,13 @@ For canonical full release notes, use:
 
 - [`CHANGELOG.md`](https://github.com/leehack/llamadart/blob/main/CHANGELOG.md)
 
+## 0.8.3
+
+- Fixed Windows CUDA backend discovery when the native asset bundle directory is
+  not on the app `PATH`. Apps using the CUDA llama.cpp backend can now resolve
+  bundled CUDA redistributables beside `ggml-cuda.dll` without manually adding
+  `.dart_tool/lib` or the native bundle path to `PATH`.
+
 ## 0.8.2
 
 - Updated the default llama.cpp native runtime pin to

website/docs/getting-started/installation.md

@@ -27,15 +27,15 @@ In Xcode, set `IPHONEOS_DEPLOYMENT_TARGET = 16.4` or
 
 ```yaml
 dependencies:
-  llamadart: ^0.8.2
+  llamadart: ^0.8.3
 ```
 
 For Flutter iOS/macOS apps that should link Apple XCFrameworks through Swift
 Package Manager, also add the runtime companion packages you need:
 
 ```yaml
 dependencies:
-  llamadart: ^0.8.2
+  llamadart: ^0.8.3
   llamadart_llama_cpp_flutter: ^0.0.3 # GGUF / llama.cpp
   llamadart_litert_lm_flutter: ^0.0.2 # .litertlm / LiteRT-LM
 ```