xl03-bit/sm120-nvfp4-training
Library
Can I trust it?
No red flags in our checks.
Stars from real accounts No data
We haven't checked its stars for fakes yet.
Active OK
Last change 3 days ago.
Licence OK
MIT: you can use it for free, also in a commercial product. Keep the licence notice.
Quick checks from public data. They help, but they are no guarantee.
What it is
A PyTorch package for NVFP4 training on GeForce RTX 50-series and RTX PRO 6000 Blackwell (SM120) GPUs. It implements stochastic rounding and random Hadamard transforms in software, uses FP4 Tensor Cores for matrix multiplies, and offers six profiles that trade memory for speed…
Who it's for
ML engineers training or fine-tuning models in PyTorch on Blackwell consumer or workstation GPUs
How to use it
Install
python -m pip install --no-deps .From its README. Check it there before running it.
Ask your AI to set it up
Paste this into Claude, ChatGPT or your coding assistant.
Help me install and set up xl03-bit/sm120-nvfp4-training in my project. Follow its README: https://github.com/xl03-bit/sm120-nvfp4-training. Explain each step and check it is safe before running anything.
Star curve
Alternatives
Similar projects named in its profile. Each link searches the directory.
Get alerts for repos like this
We'll email you when something in Models & inference starts rising. Free, once a week, only what matches.
What arrives: one email on Mondays with the repos that match, never more. See last week's. Privacy
For maintainers
Add this badge to your README. It shows the repo's place on the radar, updates every hour and links back to this page.
[](https://repoalerts.com/repo/xl03-bit/sm120-nvfp4-training?ref=badge)Similar repos
- astding/mmaemu
A Python library that emulates bit-exact FP8 tensor-core GEMMs, matching the output of torch._scaled_mm and Transformer Engine on NVIDIA H100 PCIe or RTX 4060 GPUs, so results can be reproduced on a C
- jesdga95/basalt
A C++ local inference engine for the Qwen3.8-Flash-Next mixture-of-experts model on NVIDIA Blackwell GPUs (sm_120) under Linux and Windows. Experts that do not fit in VRAM are kept in system RAM, opti
- HalfVulpes/Deepinfer-SM80
Ports DeepSeek's DeepInfer kernels and engine optimizations from Blackwell to NVIDIA Ampere SM80/SM86 GPUs for vLLM V1. Ships two vLLM patches plus drop-in Python, Triton and TileLang modules targetin
- KohakuBlueleaf/KohakUnlock
Research repository on unlocking tensor-compute throughput on NVIDIA GPUs via firmware-level changes. It documents a Linux initramfs port for one TU102 card and reproduction notes for Ampere, with C s
- OpenBMB/ForgeKernel
A self-evolving GPU kernel agent that starts from a reference kernel, profiles it, and runs a milestone-based coding-agent loop to produce faster kernels, promoting each accepted kernel as the next re
- NVIDIA/hyperflare-inference
A local inference server for NVIDIA SM120/SM121 GPUs aimed at long conversations, coding agents and repeated tool use. Models are prepared at build time with PyTorch, Transformers and FlashInfer, then