Skip to main content

Unapologetically SM120-only CuTe DSL kernels for NVFP4 GEMM and MoE.

Project description

b12x is an SM120/SM121 CuTe DSL kernel library for (primarily) NVFP4 LLM inference.

It is intentionally narrow. This is not a generic CUDA kernel collection or a full model-serving stack. It does not intend to target any other GPU architectures, including SM100. It is a focused package for a small number of high-performance kernels plus the runtime glue needed to launch them cleanly from sglang/vllm.

Currently supported kernels:

  • NVFP4 fused MoE GEMM
  • NVFP4 dense GEMM
  • BF16/FP8 paged attention
  • Sparse MLA attention (for DSA/NSA only).
pip install b12x

Ask your friendly neighborhood AI agent for further information on how to use this library.

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

b12x-0.13.7.tar.gz (491.8 kB view details)

Uploaded Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

b12x-0.13.7-py3-none-any.whl (546.8 kB view details)

Uploaded Python 3

File details

Details for the file b12x-0.13.7.tar.gz.

File metadata

  • Download URL: b12x-0.13.7.tar.gz
  • Upload date:
  • Size: 491.8 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/6.2.0 CPython/3.14.4

File hashes

Hashes for b12x-0.13.7.tar.gz
Algorithm Hash digest
SHA256 464de6708c2bc61943ab4811febd204190ad6269bcfea525e0260d8b9f7d785e
MD5 6826e49bfaa1334d9a2cf3272630e98c
BLAKE2b-256 40ad5faea7f74ffb0b9d9772275107a6010d3ab70dda9cd2cd3d6549d53ac28e

See more details on using hashes here.

File details

Details for the file b12x-0.13.7-py3-none-any.whl.

File metadata

  • Download URL: b12x-0.13.7-py3-none-any.whl
  • Upload date:
  • Size: 546.8 kB
  • Tags: Python 3
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/6.2.0 CPython/3.14.4

File hashes

Hashes for b12x-0.13.7-py3-none-any.whl
Algorithm Hash digest
SHA256 30fabbbb2b8d108c06f7c94acf69ae7504193124d38bb51136540ec1e11ed74b
MD5 489364cbaf45c43811794dda0a67b390
BLAKE2b-256 34e3780ca3636c7d724d4c05f234d2e46f61ef98c2df38cc0181eb0a324fdc7c

See more details on using hashes here.

Supported by

AWS Cloud computing and Security Sponsor Datadog Monitoring Depot Continuous Integration Fastly CDN Google Download Analytics Pingdom Monitoring Sentry Error logging StatusPage Status page