ameva-runtime 2.2.2__tar.gz → 2.2.4__tar.gz

{ameva_runtime-2.2.2/python/ameva_runtime.egg-info → ameva_runtime-2.2.4}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: ameva-runtime
-Version: 2.2.2
+Version: 2.2.4
 Summary: Unified Next-Gen Hardware Orchestration & AI Acceleration Runtime for Mobile & Edge
 Home-page: https://github.com/uno-km/ameva-runtime
 Author: Eunho Kim
@@ -84,21 +84,44 @@ To install with full multi-modal engine integrations (STT, TTS, LLM, Diffusion,
 pip install "ameva-runtime[all]"
 ```
 
-### 1.3 1-Click Native Hardware Asset Provisioning (One-Touch Auto-Install)
-Provision precompiled ARM64 Bionic binaries, native compute shaders, and hardware drivers in a single command without building from source:
+### 1.3 1-Click Native Hardware Asset Provisioning & Dual-Track Compatibility
+Rather than requiring users to manually compile C/C++ source trees, setup CMake/Clang toolchains, or configure OpenMP and Vulkan loaders, AMEVA-Runtime includes an automated, atomic hardware provisioner (`NativeAssetManager`):
+
 ```bash
-# Provision all modalities (Diffusion, STT, TTS, OpenMP, EGL Shim, SPIR-V Shaders)
+# 1-Click Auto-Provision all 6 modalities and native hardware drivers
 ameva install --all
 
-# Or provision a specific modality with force overwrite
+# Or alias
+ameva setup
+
+# Force overwrite existing binaries with fresh GitHub Release assets
+ameva install --all --force
+
+# Provision a single targeted modality
 ameva install --modality diffusion --force
 ```
-This automatically provisions and links:
-* **Stable Diffusion CLI (`sd-cli`)**: `~/.local/bin/sd-cli` (and legacy path `~/.cache/termux-diffusion/bin/sd-cli`)
-* **Whisper STT (`whisper-cli`)**: `~/.local/bin/whisper-cli` (and `$PREFIX/bin/whisper-cli`)
-* **Sherpa-NCNN TTS (`sherpa-ncnn-offline-tts`)**: `~/.local/bin/sherpa-ncnn-offline-tts`
-* **Vulkan HAL Shim & OpenMP (`libegl_shim.so`, `libomp.so`)**: `$PREFIX/lib/`
-* **SPIR-V Zero-Stride Bypass Shader (`matmul.spv`)**: `~/.local/share/ameva/shaders/matmul.spv`
+
+#### Dual-Track Compatibility Architecture (Track A + Track B)
+AMEVA-Runtime enforces a **Dual-Track Deployment Architecture** to ensure zero-regression interoperability between modern unified environments and existing ecosystem toolchains:
+
+| Asset / Engine | Clean Unified Path (Track B) | Legacy Bridge / Symlink (Track A) | Engine Type & Target Modality |
+| :--- | :--- | :--- | :--- |
+| **`sd-cli`** (36.2 MB) | `~/.local/bin/sd-cli` | `~/.cache/termux-diffusion/bin/sd-cli` | Stable Diffusion On-Device Vulkan Engine |
+| **`whisper-cli`** (3.3 MB) | `~/.local/bin/whisper-cli` | `$PREFIX/bin/whisper-cli`, `~/.local/bin/whisper-cpp` | Whisper Speech-to-Text ARM64 Engine |
+| **`sherpa-ncnn-offline-tts`** (3.9 MB) | `~/.local/bin/sherpa-ncnn-offline-tts` | `~/sherpa-ncnn/build-vulkan/bin/`, `$PREFIX/bin/` | Sherpa-NCNN Vulkan Neural Speech Synthesis |
+| **`libomp.so`** (1.1 MB) | `$PREFIX/lib/libomp.so` | `~/.local/lib/libomp.so` | OpenMP High-Throughput Threading Runtime |
+| **`libegl_shim.so`** (8.4 KB) | `$PREFIX/lib/libegl_shim.so` | `~/.local/lib/libegl_shim.so` | Android Termux Headless EGL/GBM Driver Shim |
+| **`matmul.spv`** (4.7 KB) | `~/.local/share/ameva/shaders/matmul.spv` | `$PREFIX/share/ameva/shaders/matmul.spv` | Mali/Adreno Zero-Stride Workaround SPIR-V Shader |
+
+#### Programmatic Python Provisioning API
+You can also trigger atomic asset provisioning directly within Python workflows:
+```python
+from ameva_runtime import provision_native_assets
+
+# Provision all native binaries and link compatibility bridges
+results = provision_native_assets(force=False)
+print("Provisioning Status:", results)
+```
 
 ### 1.4 Node.js / TypeScript SDK & CLI Installation
 Install globally or as a project dependency via `npm`:

{ameva_runtime-2.2.2 → ameva_runtime-2.2.4}/README.md

@@ -33,21 +33,44 @@ To install with full multi-modal engine integrations (STT, TTS, LLM, Diffusion,
 pip install "ameva-runtime[all]"
 ```
 
-### 1.3 1-Click Native Hardware Asset Provisioning (One-Touch Auto-Install)
-Provision precompiled ARM64 Bionic binaries, native compute shaders, and hardware drivers in a single command without building from source:
+### 1.3 1-Click Native Hardware Asset Provisioning & Dual-Track Compatibility
+Rather than requiring users to manually compile C/C++ source trees, setup CMake/Clang toolchains, or configure OpenMP and Vulkan loaders, AMEVA-Runtime includes an automated, atomic hardware provisioner (`NativeAssetManager`):
+
 ```bash
-# Provision all modalities (Diffusion, STT, TTS, OpenMP, EGL Shim, SPIR-V Shaders)
+# 1-Click Auto-Provision all 6 modalities and native hardware drivers
 ameva install --all
 
-# Or provision a specific modality with force overwrite
+# Or alias
+ameva setup
+
+# Force overwrite existing binaries with fresh GitHub Release assets
+ameva install --all --force
+
+# Provision a single targeted modality
 ameva install --modality diffusion --force
 ```
-This automatically provisions and links:
-* **Stable Diffusion CLI (`sd-cli`)**: `~/.local/bin/sd-cli` (and legacy path `~/.cache/termux-diffusion/bin/sd-cli`)
-* **Whisper STT (`whisper-cli`)**: `~/.local/bin/whisper-cli` (and `$PREFIX/bin/whisper-cli`)
-* **Sherpa-NCNN TTS (`sherpa-ncnn-offline-tts`)**: `~/.local/bin/sherpa-ncnn-offline-tts`
-* **Vulkan HAL Shim & OpenMP (`libegl_shim.so`, `libomp.so`)**: `$PREFIX/lib/`
-* **SPIR-V Zero-Stride Bypass Shader (`matmul.spv`)**: `~/.local/share/ameva/shaders/matmul.spv`
+
+#### Dual-Track Compatibility Architecture (Track A + Track B)
+AMEVA-Runtime enforces a **Dual-Track Deployment Architecture** to ensure zero-regression interoperability between modern unified environments and existing ecosystem toolchains:
+
+| Asset / Engine | Clean Unified Path (Track B) | Legacy Bridge / Symlink (Track A) | Engine Type & Target Modality |
+| :--- | :--- | :--- | :--- |
+| **`sd-cli`** (36.2 MB) | `~/.local/bin/sd-cli` | `~/.cache/termux-diffusion/bin/sd-cli` | Stable Diffusion On-Device Vulkan Engine |
+| **`whisper-cli`** (3.3 MB) | `~/.local/bin/whisper-cli` | `$PREFIX/bin/whisper-cli`, `~/.local/bin/whisper-cpp` | Whisper Speech-to-Text ARM64 Engine |
+| **`sherpa-ncnn-offline-tts`** (3.9 MB) | `~/.local/bin/sherpa-ncnn-offline-tts` | `~/sherpa-ncnn/build-vulkan/bin/`, `$PREFIX/bin/` | Sherpa-NCNN Vulkan Neural Speech Synthesis |
+| **`libomp.so`** (1.1 MB) | `$PREFIX/lib/libomp.so` | `~/.local/lib/libomp.so` | OpenMP High-Throughput Threading Runtime |
+| **`libegl_shim.so`** (8.4 KB) | `$PREFIX/lib/libegl_shim.so` | `~/.local/lib/libegl_shim.so` | Android Termux Headless EGL/GBM Driver Shim |
+| **`matmul.spv`** (4.7 KB) | `~/.local/share/ameva/shaders/matmul.spv` | `$PREFIX/share/ameva/shaders/matmul.spv` | Mali/Adreno Zero-Stride Workaround SPIR-V Shader |
+
+#### Programmatic Python Provisioning API
+You can also trigger atomic asset provisioning directly within Python workflows:
+```python
+from ameva_runtime import provision_native_assets
+
+# Provision all native binaries and link compatibility bridges
+results = provision_native_assets(force=False)
+print("Provisioning Status:", results)
+```
 
 ### 1.4 Node.js / TypeScript SDK & CLI Installation
 Install globally or as a project dependency via `npm`:

{ameva_runtime-2.2.2 → ameva_runtime-2.2.4}/README.pypi.md

@@ -33,21 +33,44 @@ To install with full multi-modal engine integrations (STT, TTS, LLM, Diffusion,
 pip install "ameva-runtime[all]"
 ```
 
-### 1.3 1-Click Native Hardware Asset Provisioning (One-Touch Auto-Install)
-Provision precompiled ARM64 Bionic binaries, native compute shaders, and hardware drivers in a single command without building from source:
+### 1.3 1-Click Native Hardware Asset Provisioning & Dual-Track Compatibility
+Rather than requiring users to manually compile C/C++ source trees, setup CMake/Clang toolchains, or configure OpenMP and Vulkan loaders, AMEVA-Runtime includes an automated, atomic hardware provisioner (`NativeAssetManager`):
+
 ```bash
-# Provision all modalities (Diffusion, STT, TTS, OpenMP, EGL Shim, SPIR-V Shaders)
+# 1-Click Auto-Provision all 6 modalities and native hardware drivers
 ameva install --all
 
-# Or provision a specific modality with force overwrite
+# Or alias
+ameva setup
+
+# Force overwrite existing binaries with fresh GitHub Release assets
+ameva install --all --force
+
+# Provision a single targeted modality
 ameva install --modality diffusion --force
 ```
-This automatically provisions and links:
-* **Stable Diffusion CLI (`sd-cli`)**: `~/.local/bin/sd-cli` (and legacy path `~/.cache/termux-diffusion/bin/sd-cli`)
-* **Whisper STT (`whisper-cli`)**: `~/.local/bin/whisper-cli` (and `$PREFIX/bin/whisper-cli`)
-* **Sherpa-NCNN TTS (`sherpa-ncnn-offline-tts`)**: `~/.local/bin/sherpa-ncnn-offline-tts`
-* **Vulkan HAL Shim & OpenMP (`libegl_shim.so`, `libomp.so`)**: `$PREFIX/lib/`
-* **SPIR-V Zero-Stride Bypass Shader (`matmul.spv`)**: `~/.local/share/ameva/shaders/matmul.spv`
+
+#### Dual-Track Compatibility Architecture (Track A + Track B)
+AMEVA-Runtime enforces a **Dual-Track Deployment Architecture** to ensure zero-regression interoperability between modern unified environments and existing ecosystem toolchains:
+
+| Asset / Engine | Clean Unified Path (Track B) | Legacy Bridge / Symlink (Track A) | Engine Type & Target Modality |
+| :--- | :--- | :--- | :--- |
+| **`sd-cli`** (36.2 MB) | `~/.local/bin/sd-cli` | `~/.cache/termux-diffusion/bin/sd-cli` | Stable Diffusion On-Device Vulkan Engine |
+| **`whisper-cli`** (3.3 MB) | `~/.local/bin/whisper-cli` | `$PREFIX/bin/whisper-cli`, `~/.local/bin/whisper-cpp` | Whisper Speech-to-Text ARM64 Engine |
+| **`sherpa-ncnn-offline-tts`** (3.9 MB) | `~/.local/bin/sherpa-ncnn-offline-tts` | `~/sherpa-ncnn/build-vulkan/bin/`, `$PREFIX/bin/` | Sherpa-NCNN Vulkan Neural Speech Synthesis |
+| **`libomp.so`** (1.1 MB) | `$PREFIX/lib/libomp.so` | `~/.local/lib/libomp.so` | OpenMP High-Throughput Threading Runtime |
+| **`libegl_shim.so`** (8.4 KB) | `$PREFIX/lib/libegl_shim.so` | `~/.local/lib/libegl_shim.so` | Android Termux Headless EGL/GBM Driver Shim |
+| **`matmul.spv`** (4.7 KB) | `~/.local/share/ameva/shaders/matmul.spv` | `$PREFIX/share/ameva/shaders/matmul.spv` | Mali/Adreno Zero-Stride Workaround SPIR-V Shader |
+
+#### Programmatic Python Provisioning API
+You can also trigger atomic asset provisioning directly within Python workflows:
+```python
+from ameva_runtime import provision_native_assets
+
+# Provision all native binaries and link compatibility bridges
+results = provision_native_assets(force=False)
+print("Provisioning Status:", results)
+```
 
 ### 1.4 Node.js / TypeScript SDK & CLI Installation
 Install globally or as a project dependency via `npm`:

{ameva_runtime-2.2.2 → ameva_runtime-2.2.4}/python/ameva_runtime/_version.py

@@ -5,4 +5,4 @@ Single Source of Truth (SSOT) for all runtime packages.
 Modifying this single string updates ameva_runtime, ameva_runtime.vulkan,
 and the package metadata simultaneously.
 """
-__version__ = "2.2.2"
+__version__ = "2.2.4"

{ameva_runtime-2.2.2 → ameva_runtime-2.2.4}/python/ameva_runtime/adapters/diffusion.py

@@ -151,6 +151,8 @@ class DiffusionAdapter(BaseAdapter):
     def get_execution_env(cls, extra_env: Optional[dict[str, str]] = None) -> dict[str, str]:
         """Assemble environment variables including LD_PRELOAD shim for mobile Vulkan HAL."""
         env = dict(os.environ)
+        # Essential bypass for Qualcomm Adreno & mobile Vulkan debug CPU check overhead
+        env.setdefault("GGML_VULKAN_SKIP_CHECKS", "999999999")
         if extra_env:
             env.update(extra_env)
 
@@ -171,14 +173,17 @@ class DiffusionAdapter(BaseAdapter):
         output_path: str,
         width: int = 512,
         height: int = 512,
-        steps: int = 4,
+        steps: int = 2,
         cfg_scale: float = 1.0,
+        guidance: Optional[float] = None,
         threads: Any = "auto",
         target_backend: str = "auto",
         sampling_method: Optional[str] = None,
         seed: Optional[int] = None,
         vae_path: Optional[str] = None,
         vae_tiling: bool = False,
+        backend: Optional[str] = None,
+        is_adreno: bool = False,
     ) -> list[str]:
         """최신 sd-cli 규격에 부합하는 안전하고 검증된 Diffusion CLI 인자 목록을 조립합니다."""
         if threads == "auto" or threads is None:
@@ -199,9 +204,15 @@ class DiffusionAdapter(BaseAdapter):
             "--cfg-scale", str(cfg_scale),
         ]
 
+        if guidance is not None:
+            cmd.extend(["--guidance", str(guidance)])
+
         if target_backend == "cpu":
             cmd.append("--offload-to-cpu")
-        # In Vulkan GPU or auto mode on supported hardware, sd-cli runs native Vulkan without broken flags
+        elif backend:
+            cmd.extend(["--backend", str(backend)])
+        elif is_adreno or target_backend in ("vulkan", "gpu"):
+            cmd.extend(["--backend", "clip=vulkan0,diffusion=vulkan0,vae=vulkan0"])
 
         if vae_tiling:
             cmd.append("--vae-tiling")

{ameva_runtime-2.2.2 → ameva_runtime-2.2.4/python/ameva_runtime.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: ameva-runtime
-Version: 2.2.2
+Version: 2.2.4
 Summary: Unified Next-Gen Hardware Orchestration & AI Acceleration Runtime for Mobile & Edge
 Home-page: https://github.com/uno-km/ameva-runtime
 Author: Eunho Kim
@@ -84,21 +84,44 @@ To install with full multi-modal engine integrations (STT, TTS, LLM, Diffusion,
 pip install "ameva-runtime[all]"
 ```
 
-### 1.3 1-Click Native Hardware Asset Provisioning (One-Touch Auto-Install)
-Provision precompiled ARM64 Bionic binaries, native compute shaders, and hardware drivers in a single command without building from source:
+### 1.3 1-Click Native Hardware Asset Provisioning & Dual-Track Compatibility
+Rather than requiring users to manually compile C/C++ source trees, setup CMake/Clang toolchains, or configure OpenMP and Vulkan loaders, AMEVA-Runtime includes an automated, atomic hardware provisioner (`NativeAssetManager`):
+
 ```bash
-# Provision all modalities (Diffusion, STT, TTS, OpenMP, EGL Shim, SPIR-V Shaders)
+# 1-Click Auto-Provision all 6 modalities and native hardware drivers
 ameva install --all
 
-# Or provision a specific modality with force overwrite
+# Or alias
+ameva setup
+
+# Force overwrite existing binaries with fresh GitHub Release assets
+ameva install --all --force
+
+# Provision a single targeted modality
 ameva install --modality diffusion --force
 ```
-This automatically provisions and links:
-* **Stable Diffusion CLI (`sd-cli`)**: `~/.local/bin/sd-cli` (and legacy path `~/.cache/termux-diffusion/bin/sd-cli`)
-* **Whisper STT (`whisper-cli`)**: `~/.local/bin/whisper-cli` (and `$PREFIX/bin/whisper-cli`)
-* **Sherpa-NCNN TTS (`sherpa-ncnn-offline-tts`)**: `~/.local/bin/sherpa-ncnn-offline-tts`
-* **Vulkan HAL Shim & OpenMP (`libegl_shim.so`, `libomp.so`)**: `$PREFIX/lib/`
-* **SPIR-V Zero-Stride Bypass Shader (`matmul.spv`)**: `~/.local/share/ameva/shaders/matmul.spv`
+
+#### Dual-Track Compatibility Architecture (Track A + Track B)
+AMEVA-Runtime enforces a **Dual-Track Deployment Architecture** to ensure zero-regression interoperability between modern unified environments and existing ecosystem toolchains:
+
+| Asset / Engine | Clean Unified Path (Track B) | Legacy Bridge / Symlink (Track A) | Engine Type & Target Modality |
+| :--- | :--- | :--- | :--- |
+| **`sd-cli`** (36.2 MB) | `~/.local/bin/sd-cli` | `~/.cache/termux-diffusion/bin/sd-cli` | Stable Diffusion On-Device Vulkan Engine |
+| **`whisper-cli`** (3.3 MB) | `~/.local/bin/whisper-cli` | `$PREFIX/bin/whisper-cli`, `~/.local/bin/whisper-cpp` | Whisper Speech-to-Text ARM64 Engine |
+| **`sherpa-ncnn-offline-tts`** (3.9 MB) | `~/.local/bin/sherpa-ncnn-offline-tts` | `~/sherpa-ncnn/build-vulkan/bin/`, `$PREFIX/bin/` | Sherpa-NCNN Vulkan Neural Speech Synthesis |
+| **`libomp.so`** (1.1 MB) | `$PREFIX/lib/libomp.so` | `~/.local/lib/libomp.so` | OpenMP High-Throughput Threading Runtime |
+| **`libegl_shim.so`** (8.4 KB) | `$PREFIX/lib/libegl_shim.so` | `~/.local/lib/libegl_shim.so` | Android Termux Headless EGL/GBM Driver Shim |
+| **`matmul.spv`** (4.7 KB) | `~/.local/share/ameva/shaders/matmul.spv` | `$PREFIX/share/ameva/shaders/matmul.spv` | Mali/Adreno Zero-Stride Workaround SPIR-V Shader |
+
+#### Programmatic Python Provisioning API
+You can also trigger atomic asset provisioning directly within Python workflows:
+```python
+from ameva_runtime import provision_native_assets
+
+# Provision all native binaries and link compatibility bridges
+results = provision_native_assets(force=False)
+print("Provisioning Status:", results)
+```
 
 ### 1.4 Node.js / TypeScript SDK & CLI Installation
 Install globally or as a project dependency via `npm`: