mirror of
https://github.com/TheTom/turboquant_plus.git
synced 2026-10-22 02:53:13 -05:00
TheTomandClaude Opus 4.7
eb0ebe3573
- pyproject: add refract-sglang extra (requests + transformers) - README: install snippet, platform support table (vLLM + SGLang verified on Ubuntu MI300X), backends summary, repo tree - QUICKSTART: alpha-warning bullet, prereqs (vLLM + SGLang sections), --backend selector help, backends status table - backends/__init__.py: docstring status block (mlx + vllm + sglang marked production, was SKELETON) - CHANGELOG: new v0.3.2.1 entry documenting the cross-engine bench validation Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> Co-Authored-By: tturney1@gmail.com <tturney1@gmail.com>
85 lines
2.6 KiB
TOML
85 lines
2.6 KiB
TOML
[project]
|
|
name = "turboquant"
|
|
version = "0.1.0"
|
|
description = "TurboQuant + REFRACT: KV-cache quantization (TurboQuant) and quality oracle (REFRACT)"
|
|
readme = "README.md"
|
|
license = { text = "MIT" }
|
|
requires-python = ">=3.9"
|
|
authors = [
|
|
{ name = "Tom (TheTom)" },
|
|
]
|
|
keywords = ["llm", "kv-cache", "quantization", "evaluation", "llama-cpp", "mlx"]
|
|
classifiers = [
|
|
"Development Status :: 3 - Alpha",
|
|
"Intended Audience :: Developers",
|
|
"Intended Audience :: Science/Research",
|
|
"License :: OSI Approved :: MIT License",
|
|
"Operating System :: MacOS",
|
|
"Operating System :: POSIX :: Linux",
|
|
"Operating System :: Microsoft :: Windows",
|
|
"Programming Language :: Python :: 3",
|
|
"Programming Language :: Python :: 3.9",
|
|
"Programming Language :: Python :: 3.10",
|
|
"Programming Language :: Python :: 3.11",
|
|
"Programming Language :: Python :: 3.12",
|
|
"Topic :: Scientific/Engineering :: Artificial Intelligence",
|
|
]
|
|
|
|
# REFRACT itself has zero non-stdlib deps — kept that way deliberately so
|
|
# `pip install .` for the alpha is one command and pulls nothing extra.
|
|
# numpy / scipy / etc. live under the `turboquant` extra below; they're
|
|
# only needed for the TurboQuant Python implementation, NOT for REFRACT.
|
|
dependencies = []
|
|
|
|
[project.optional-dependencies]
|
|
# REFRACT alpha testers: `pip install -e .` is enough. These extras are
|
|
# only needed if you run the matching backend.
|
|
refract-mlx = [
|
|
"mlx>=0.20",
|
|
"mlx-lm>=0.31", # qwen3_5 + gemma-4 support requires 0.31+
|
|
]
|
|
refract-vllm = [
|
|
"vllm>=0.5",
|
|
]
|
|
refract-sglang = [
|
|
# SGLang backend is HTTP-based; the user runs the SGLang server
|
|
# separately (typically via Docker). REFRACT only needs a HTTP client
|
|
# and a tokenizer to construct prompt token IDs.
|
|
"requests>=2.28",
|
|
"transformers>=4.40",
|
|
]
|
|
# TurboQuant Python implementation (separate from REFRACT). Install with
|
|
# `pip install -e .[turboquant]` if you're running benchmarks or unit tests
|
|
# under turboquant/.
|
|
turboquant = [
|
|
"numpy>=1.24",
|
|
"scipy>=1.10",
|
|
]
|
|
dev = [
|
|
"pytest>=7.0",
|
|
"pytest-cov>=4.0",
|
|
"coverage>=7.0",
|
|
]
|
|
bench = [
|
|
"matplotlib",
|
|
]
|
|
|
|
[project.urls]
|
|
Homepage = "https://github.com/TheTom/turboquant_plus"
|
|
Repository = "https://github.com/TheTom/turboquant_plus"
|
|
Issues = "https://github.com/TheTom/turboquant_plus/issues"
|
|
|
|
[project.scripts]
|
|
refract = "refract.cli:main"
|
|
|
|
[tool.setuptools.packages.find]
|
|
include = ["turboquant*", "refract*"]
|
|
|
|
[build-system]
|
|
requires = ["setuptools>=68"]
|
|
build-backend = "setuptools.build_meta"
|
|
|
|
[tool.pytest.ini_options]
|
|
# Both turboquant/ and refract/ ship their own tests/ subtree.
|
|
testpaths = ["tests", "refract/tests"]
|