kvcache-ai-ktransformers/setup.py
Benjamin F d7b5b49a3e
Some checks failed
Book-CI / test-1 (push) Has been cancelled
Book-CI / test-2 (push) Has been cancelled
Release to PyPI / Build & publish sglang-kt (push) Has been cancelled
Book-CI / test (push) Has been cancelled
Deploy / deploy (macos-latest) (push) Has been cancelled
Deploy / deploy (ubuntu-latest) (push) Has been cancelled
Deploy / deploy (windows-latest) (push) Has been cancelled
Release Fake Tag / publish (push) Has been cancelled
Release to PyPI / Build kt-kernel (Python 3.11) (push) Has been cancelled
Release to PyPI / Build kt-kernel (Python 3.12) (push) Has been cancelled
Release sglang-kt to PyPI / Build sglang-kt wheel (push) Has been cancelled
Release to PyPI / Publish kt-kernel to PyPI (push) Has been cancelled
Release sglang-kt to PyPI / Publish sglang-kt to PyPI (push) Has been cancelled
[release]: 0.6.2.post1
V4-Flash MXFP4 full-GPU prefill fallback now works:
- Previously crashed all TP schedulers with StopIteration/AttributeError
  whenever --kt-gpu-prefill-token-threshold was low enough to actually
  fire (path was hardcoded for FP8/INT4 layouts).
- Now detects MXFP4, re-runs the V4 swizzle on the 256-expert gpu_layer,
  caches the load across prefill chunks.
- Measured on 8x RTX 5090 (threshold=1024, chunked=1024):
  16k input -> 2011 tok/s, 65k -> 2798, 262k -> 2154 prefill TPS.
2026-05-03 21:07:23 +08:00

29 lines
689 B
Python

"""Lightweight top-level package: pip install ktransformers -> installs kt-kernel.
Extras:
- ktransformers[sft] installs transformers-kt + accelerate-kt
- ktransformers[sglang] installs sglang-kt
"""
from pathlib import Path
from setuptools import setup
_version_file = Path(__file__).resolve().parent / "version.py"
_ns = {}
exec(_version_file.read_text(), _ns)
_v = _ns["__version__"]
setup(
version=_v,
install_requires=[
f"kt-kernel=={_v}",
],
extras_require={
"sft": [
"transformers-kt==5.6.0.post1",
"accelerate-kt==1.14.0.post1",
],
"sglang": [
"sglang-kt==0.6.2.post1",
],
},
)