Metadata-Version: 2.4
Name: basert
Version: 0.4.6
Summary: Fast local LLM inference on Apple Silicon and NVIDIA GB10: the basert CLI, engine, and Python bindings
Author: BaseCompute
License-Expression: Apache-2.0 AND LicenseRef-BaseRT-Engine
Project-URL: Homepage, https://github.com/basecompute/baseRT
Project-URL: Documentation, https://docs.basecompute.co
Project-URL: Changelog, https://github.com/basecompute/baseRT/blob/main/CHANGELOG.md
Keywords: llm,inference,apple-silicon,metal,cuda,openai
Classifier: Development Status :: 3 - Alpha
Classifier: Environment :: GPU
Classifier: Intended Audience :: Developers
Classifier: Operating System :: MacOS
Classifier: Operating System :: POSIX :: Linux
Classifier: Programming Language :: Python :: 3
Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
Requires-Python: >=3.9
Description-Content-Type: text/markdown
License-File: LICENSE
License-File: NOTICE
License-File: baseRT/_engine/LICENSE
License-File: baseRT/_engine/NOTICE
Dynamic: license-file

# BaseRT

Fast local LLM inference on Apple Silicon (Metal) and NVIDIA GB10 / DGX Spark
(CUDA). This package installs the `basert` CLI, the prebuilt engine, and the
Python bindings in one step.

```sh
pip install basert
```

Requires Apple Silicon with macOS 15+, or Linux/arm64 with an NVIDIA GPU and
the CUDA 13 runtime (cuBLAS).

## CLI

```sh
basert pull Qwen/Qwen3-4B      # download from Hugging Face and convert to .base
basert chat Qwen/Qwen3-4B      # chat in the terminal
basert serve Qwen/Qwen3-4B --api-key "$(uuidgen)" --port 8080   # OpenAI-compatible API
```

## Python

```python
import baseRT

with baseRT.Model("model.base") as model:
    print(model.generate_text("The capital of France is", max_tokens=64))
```

Docs: <https://docs.basecompute.co> · Source: <https://github.com/basecompute/baseRT>

## License

The Python bindings and the `basert` CLI launcher are Apache-2.0. The bundled
engine binaries (`baseRT/_engine/`) are under the BaseRT Engine license
included with the package.
