Refactor OpenVINO backend: isolate op support, manage buffers, and document API

.github/workflows/build-openvino.yml

@@ -78,18 +78,16 @@ jobs:
 
       - name: Test (CPU)
         id: cmake_test_cpu
-        # TODO: fix and re-enable the `test-llama-archs` test below
         run: |
           cd ${{ github.workspace }}
-          ctest --test-dir build/ReleaseOV -L main -E "test-llama-archs" --verbose --timeout 2000
+          ctest --test-dir build/ReleaseOV -L main --verbose --timeout 2000
 
       - name: Test (GPU)
         id: cmake_test_gpu
-        # TODO: fix and re-enable the `test-llama-archs` test below
         run: |
           cd ${{ github.workspace }}
           export GGML_OPENVINO_DEVICE=GPU
-          ctest --test-dir build/ReleaseOV -L main -E "test-llama-archs" --verbose --timeout 3000
+          ctest --test-dir build/ReleaseOV -L main --verbose --timeout 3000
 
   openvino-windows-2022:
     runs-on: windows-2022
@@ -159,11 +157,10 @@ jobs:
       - name: Test (CPU)
         id: cmake_test_cpu
         shell: cmd
-        # TODO: fix and re-enable the `test-llama-archs` test below
         run: |
           REM Find extracted OpenVINO folder dynamically
           for /d %%i in (openvino_toolkit\*) do set OPENVINO_ROOT=%%i
           call "%OPENVINO_ROOT%\setupvars.bat"
 
           cd build
-          ctest --test-dir ReleaseOV -L main -E "test-llama-archs" -C Release --verbose --timeout 3000
+          ctest --test-dir ReleaseOV -L main -C Release --verbose --timeout 3000

docs/backend/OPENVINO.md

@@ -206,7 +206,7 @@ cmake -B build/ReleaseOV -G Ninja -DCMAKE_BUILD_TYPE=Release -DGGML_OPENVINO=ON
 cmake --build build/ReleaseOV --parallel
 ```
 
-- **Windows:** Open a **Developer Command Prompt for VS 2022** (so the MSVC toolchain is on `PATH`), then run:
+- **Windows:** Open **x64 Native Tools Command Prompt for VS** (so the MSVC toolchain is on `PATH`), then run:
 
 ```cmd
 C:\Intel\openvino\setupvars.bat
@@ -710,11 +710,15 @@ Boolean flags follow a uniform convention: set to a **positive integer** (e.g. `
 |-----------------------------------|-----------|------------|-------------------------------------------------------------------------------------------------------------|
 | `GGML_OPENVINO_DEVICE`            | String    | `CPU`      | Specify the target device (CPU, GPU, NPU). On systems with multiple GPUs, use `GPU.0` or `GPU.1` to explicitly target specific GPU. See [OpenVINO GPU Device](https://docs.openvino.ai/2026/openvino-workflow/running-inference/inference-devices-and-modes/gpu-device.html). When set to **NPU**, static compilation mode is enabled for optimal performance. |
 | `GGML_OPENVINO_CACHE_DIR`         | String    | `not set`  | Directory for OpenVINO model caching (recommended: `/tmp/ov_cache`). Enables model caching when set. **Not supported on NPU devices.** |
+| `GGML_OPENVINO_COMPILED_MODEL_CACHE_DIR` | String | `not set` | Directory for the frontend compiled-model cache. When set, OpenVINO compiled models are exported as blobs and imported on later runs to skip weight requantization, graph conversion, and compilation for matching single-graph models. |
 | `GGML_OPENVINO_PREFILL_CHUNK_SIZE`| Integer   | `256`      | Token chunk size for **NPU** prefill (NPU-only; ignored on CPU/GPU). Must be a positive integer; otherwise the default is used. |
 | `GGML_OPENVINO_STATEFUL_EXECUTION`| Boolean   | `0`        | Enable stateful KV cache for better performance. Recommended on CPU, GPU.                                   |
 | `GGML_OPENVINO_DISABLE_CACHE`     | Boolean   | `0`        | Disable the in-process compiled-model / decoder cache (cache is on by default). Set to `1` to disable.      |
 | `GGML_OPENVINO_DISABLE_KV_SLICE`  | Boolean   | `0`        | Disable the KV-cache input-tensor slicing optimization (slicing is on by default on CPU/GPU). Set to `1` to disable. |
 | `GGML_OPENVINO_MANUAL_GQA_ATTN`   | Boolean   | device-based | Tri-state. When **unset**, manual GQA attention is enabled by default on `GPU` and disabled on other devices. Set to a positive integer to force-enable, or `0` to force-disable. |
+| `GGML_OPENVINO_MEMORY_OPTIMIZE`   | Boolean   | `0`        | Umbrella switch for compile-time memory reductions. Enables `GGML_OPENVINO_REDUCE_COMPILE_MEM` and, on GPU, `GGML_OPENVINO_RELEASE_WEIGHTS` unless those fine-grained variables are explicitly set. |
+| `GGML_OPENVINO_REDUCE_COMPILE_MEM`| Boolean   | inherits from `GGML_OPENVINO_MEMORY_OPTIMIZE` | Reduce compile-time host memory use by streaming weight requantization and avoiding extra weight-node materialization where possible. Set explicitly to override the umbrella switch. |
+| `GGML_OPENVINO_RELEASE_WEIGHTS`   | Boolean   | inherits from `GGML_OPENVINO_MEMORY_OPTIMIZE` on GPU | GPU-only. Release host weight buffers after the compiled model cache can reuse the device/plugin copy. Requires stable graph shapes; dynamic workloads that need recompilation should leave this disabled. |
 | `GGML_OPENVINO_PROFILING`         | Boolean   | `0`        | Enable execution-time profiling.                                                                            |
 | `GGML_OPENVINO_DUMP_CGRAPH`       | Boolean   | `0`        | Dump the GGML compute graph to `cgraph_ov.txt`.                                                             |
 | `GGML_OPENVINO_DUMP_IR`           | Boolean   | `0`        | Serialize OpenVINO IR files with timestamps.                                                                |

ggml/include/ggml-openvino.h

@@ -11,25 +11,53 @@ extern "C" {
 #define GGML_OPENVINO_NAME "OPENVINO"
 
 // backend API
+/// \brief Initialize an OpenVINO backend instance for the given device.
+/// \param device OpenVINO device index to initialize.
+/// \return OpenVINO backend instance, or nullptr if initialization fails.
 GGML_BACKEND_API ggml_backend_t ggml_backend_openvino_init(int device);
 
+/// \brief Return whether the backend instance is an OpenVINO backend.
+/// \param backend Backend instance to query.
+/// \return True if the backend is an OpenVINO backend.
 GGML_BACKEND_API bool ggml_backend_is_openvino(ggml_backend_t backend);
 
+/// \brief Return whether the buffer instance is an OpenVINO backend buffer.
+/// \param buffer Backend buffer instance to query.
+/// \return True if the buffer is an OpenVINO backend buffer.
 GGML_BACKEND_API bool ggml_backend_buffer_is_openvino(ggml_backend_buffer_t buffer);
 
+/// \brief Return whether the buffer type is an OpenVINO device buffer type.
+/// \param buft Backend buffer type to query.
+/// \return True if the buffer type is an OpenVINO device buffer type.
 GGML_BACKEND_API bool ggml_backend_buft_is_openvino(ggml_backend_buffer_type_t buft);
 
+/// \brief Return whether the buffer type is an OpenVINO host buffer type.
+/// \param buft Backend buffer type to query.
+/// \return True if the buffer type is an OpenVINO host buffer type.
 GGML_BACKEND_API bool ggml_backend_buft_is_openvino_host(ggml_backend_buffer_type_t buft);
 
+/// \brief Return the context identifier for an OpenVINO backend buffer.
+/// \param buffer Backend buffer instance to query.
+/// \return OpenVINO buffer context identifier, or 0 if the buffer is not an OpenVINO buffer.
 GGML_BACKEND_API size_t ggml_backend_openvino_buffer_get_ctx_id(ggml_backend_buffer_t buffer);
 
 // device buffer
+/// \brief Return the OpenVINO device buffer type for the given device.
+/// \param device OpenVINO device index.
+/// \return OpenVINO device buffer type.
 GGML_BACKEND_API ggml_backend_buffer_type_t ggml_backend_openvino_buffer_type(int device);
 
+/// \brief Return the OpenVINO host buffer type for the given device.
+/// \param device OpenVINO device index.
+/// \return OpenVINO host buffer type.
 GGML_BACKEND_API ggml_backend_buffer_type_t ggml_backend_openvino_host_buffer_type(int device);
 
+/// \brief Return the number of OpenVINO devices available to the backend.
+/// \return Number of OpenVINO devices.
 GGML_BACKEND_API int ggml_backend_openvino_get_device_count(void);
 
+/// \brief Return the OpenVINO backend registry instance.
+/// \return OpenVINO backend registry instance.
 GGML_BACKEND_API ggml_backend_reg_t ggml_backend_openvino_reg(void);
 
 #ifdef __cplusplus