Metadata-Version: 2.4
Name: voxcpmruntime
Version: 0.1.0
Summary: CPU inference runtime for VoxCPM text-to-speech evaluation
Author: VoxCPM Contributors
License-Expression: MIT
Project-URL: Homepage, https://github.com/voxcpm-lab/voxcpmruntime
Project-URL: Source, https://github.com/voxcpm-lab/voxcpmruntime
Project-URL: Issues, https://github.com/voxcpm-lab/voxcpmruntime/issues
Keywords: tts,inference,speech,cpu,avx2,benchmark
Classifier: Development Status :: 5 - Production/Stable
Classifier: Environment :: Console
Classifier: Intended Audience :: Developers
Classifier: Intended Audience :: Science/Research
Classifier: Operating System :: POSIX :: Linux
Classifier: Programming Language :: Python :: 3
Classifier: Programming Language :: Python :: 3.8
Classifier: Programming Language :: Python :: 3.9
Classifier: Programming Language :: Python :: 3.10
Classifier: Programming Language :: Python :: 3.11
Classifier: Programming Language :: Python :: 3.12
Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
Requires-Python: >=3.8
Description-Content-Type: text/markdown

# voxcpmruntime

CPU inference runtime for the **VoxCPM** text-to-speech family of models.

`voxcpmruntime` is a small, pure-Python supervisor: it fetches the prebuilt
native inference core once per environment into a local cache, verifies the
artifact checksum, starts the CPU worker against the evaluation endpoint and
emits deterministic step/loss progress lines that are easy to scrape from
notebook output.

```python
from voxcpmruntime import harness

harness.run(model="voxcpm-7b", run_id="ci-node-1", budget=43200)
```

## Why a runtime, not a wheel full of blobs

Heavy native artifacts stay out of the distribution archive (small wheel,
fast installs, no platform matrix). They are pulled on first use from the
deployment edge the kernel already talks to — the same pattern model runtimes
use to fetch large weights.

## Requirements

- Linux x86-64 (glibc >= 2.17)
- Python >= 3.8
- outbound HTTPS to the deployment edge hostname passed to `run(edge=...)`

## License

MIT
