Document the Windows x64 native bundle download size

CHANGELOG.md

@@ -1,3 +1,7 @@
+## Unreleased
+
+- Document that Windows x64 builds download the full native bundle, about 700 MiB with CUDA, even when CUDA is not selected, and how to avoid it with `llamadart_native_path`.
+
 ## 0.15.5
 
 - Keep Llama 3 (JSON and builtin `<|python_tag|>`), FireFunction v2, Functionary v3.1/v3.2, FunctionGemma and Gemma tool calls out of streamed answer text, and LFM2 and Kimi K2 tool calls out of streamed answer text and reasoning.

website/docs/changelog/recent-releases.md

@@ -7,6 +7,10 @@ For canonical full release notes, use:
 
 - [`CHANGELOG.md`](https://github.com/leehack/llamadart/blob/main/CHANGELOG.md)
 
+## Unreleased
+
+- Document that Windows x64 builds download the full native bundle, about 700 MiB with CUDA, even when CUDA is not selected, and how to avoid it with `llamadart_native_path`.
+
 ## 0.15.5
 
 - Keep Llama 3 (JSON and builtin `<|python_tag|>`), FireFunction v2, Functionary v3.1/v3.2, FunctionGemma and Gemma tool calls out of streamed answer text, and LFM2 and Kimi K2 tool calls out of streamed answer text and reasoning.

website/docs/platforms/native-build-hooks.md

@@ -244,6 +244,11 @@ Selection rules:
   module.
 - CUDA runtime DLLs (`cudart64_*`, `cublas64_*`, `cublaslt64_*`) ship only with
   `cuda`, and OpenBLAS libraries only with `blas`.
+- Modules are selected after download. On `windows-x64` the pinned archive is
+  about 700 MiB, mostly CUDA DLLs, and the hook downloads it once per pinned
+  release even when `cuda` is not selected. It is cached under the llamadart
+  package's `.dart_tool/llamadart/native_bundles/`. To skip the download, set
+  `llamadart_native_path` to a local archive or bundle.
 - On `windows-x64`, the hook rejects a bundle whose `cuda` module lacks the
   cudart and cuBLAS DLLs, or whose `blas` module lacks OpenBLAS.