AMEVA-Runtime (Python)

Next-Gen Unified On-Device Hardware Orchestration & 6-Modality AI Acceleration Runtime (with BitNet 1.58-bit Vulkan Compute) for Mobile & Edge

Installation

pip install ameva-runtime

Quickstart

import ameva_runtime as ameva
from ameva_runtime import vulkan

# Inspect hardware
profile = ameva.detect_hardware()
print(f"Target: {profile.soc_name} | {profile.gpu_vendor}")

# Run diagnostics
doc = vulkan.Doctor()
report = doc.run_self_test()
print(f"Passed: {report.passed_stages}/{report.total_stages}")

# Run Bionic Native Vulkan TTS (Zero-Silent-Fallback)
from ameva_runtime.adapters.tts import TTSAdapter
adapter = TTSAdapter(backend="vulkan", model_tier="medium")
output_wav = adapter.synthesize(
    text="온디바이스 네이티브 하드웨어 가속 음성 합성 테스트입니다.",
    output_path="output.wav"
)
print(f"TTS Synthesized: {output_wav}, RTF: {adapter.last_rtf:.3f}x")

Empirical Benchmarks

Galaxy S22 (Adreno 730):
- MeloTTS Universal Bilingual (Vulkan GPU): 2,750 ms, RTF 0.88x (Real-time).
Galaxy S21 (Mali-G78):
- Piper VITS On-chip Tiled (Vulkan GPU): 560 ms, RTF 0.18x (5.5x faster than RT).
Galaxy S20 (Mali-G77):
- Piper VITS On-chip Tiled (Vulkan GPU): 750 ms, RTF 0.24x (4.1x faster than RT).
Galaxy S25 (Adreno 830):
- BitNet 1.58-bit LLM: 17.558 t/s (12.58x speedup), Prompt Eval 205.9 ms.
- Qwen2.5-0.5B LLM: 35.80 t/s (25/25 VRAM layers).
- Whisper STT: 4,401 ms, Supertonic 3 TTS: 380 ms (RTF 0.12x).
Galaxy A35 (Mali-G68 MP5):
- BitNet 1.58-bit LLM: 3.471 t/s (5.94x speedup), Prompt Eval 1,552.8 ms.
- Qwen2.5-0.5B LLM: 4.44 t/s (+26.9% vs NEON).
- Whisper STT: 360.60s (2.26x speedup), Piper TTS: RTF 1.146x.

Documentation

License

vulkan-compute mobile-gpu hardware-acceleration hardware-abstraction-layer adreno-gpu arm-mali snapdragon-8-elite exynos termux on-device-ai edge-ai tensor-acceleration spir-v compute-shaders zero-silent-fallback llamacpp whisper-cpp sherpa-onnx stable-diffusion vision-language-models bitnet gguf ncnn bionic-loader arm64 aarch64 cgroup-management cpu-neon thermal-throttling power-efficiency smart-router hardware-orchestration multi-modal-ai subgroup-operations gemm-acceleration mobile-vlm speech-to-text text-to-speech image-generation edge-inference unprivileged-userspace termux-wake-lock phantom-process-killer android-ai-runtime valhall-gpu adreno-830 mali-g68 ameva-foundation uno-km open-source-ai

Release files for ameva-runtime 2.6.1

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Source distribution (sdist)

Source distribution for ameva-runtime 2.6.1
File	Size	Uploaded
ameva_runtime-2.6.1.tar.gz	85.3 kB	Sep 14, 2026	Details

Built distribution (wheel)

Table of built distributions (wheels) for ameva-runtime 2.6.1
File	Interpreter	ABI	Platform	Reset
ameva_runtime-2.6.1-py3-none-any.whl 93.6 kB Sep 14, 2026	Python 3	none	any	Details

Total release size: 178.9 kB

Release files / ameva_runtime-2.6.1.tar.gz

Download URL	ameva_runtime-2.6.1.tar.gz
Size	85.3 kB
Tags	Source
SHA-256 checksum How to use checksums	`4e7300d8381b0defca4de10082d532899b70877778ff109d43ab04fc4b749592`
BLAKE2b-256 checksum How to use checksums	`0d775612a01ddd200620390b5977ac46f396d15cd003deb3e56b6fb03cedae51`
Upload date	Sep 14, 2026
Uploaded using Trusted Publishing? What is trusted publishing?	No
Uploaded via	`twine/7.0.0 CPython/3.12.0`

Release files / ameva_runtime-2.6.1-py3-none-any.whl

Download URL	ameva_runtime-2.6.1-py3-none-any.whl
Size	93.6 kB
Tags	Python 3
SHA-256 checksum How to use checksums	`9277f2152452256a7481a9973e3a77430495fde0a00b7b66ab4df600005f8c3e`
BLAKE2b-256 checksum How to use checksums	`403667d0964c7bbffae63c6ec47bd088700f02baaa7419bf470308eef08ef85b`
Upload date	Sep 14, 2026
Uploaded using Trusted Publishing? What is trusted publishing?	No
Uploaded via	`twine/7.0.0 CPython/3.12.0`