MODEL INFRA / MIXED-FORMAT QUANTIZATIONOpen research prototype

TyloQuant MFQ

Put every bit where precision matters most

TyloQuant MFQ co-designs quantization formats, precision allocation, and inference kernels so packed weights can run directly through CUDA kernels and a C++ runtime for high-fidelity local LLM inference.

View on GitHubThis page summarizes the public repository. Refer to the repository for implementation details, test conditions, and reproduction notes.
010.84 – 8.30

BPW encoding range

Custom weight encodings span quality tiers from below 1 bit to above 8 bits.

0216

Production encodings

NINT, NVQ, NPQ, NEPQ, and related formats serve different precision budgets.

03CUDA + C++

Direct inference runtime

Packed weights execute without materializing complete FP16 weights.

METHOD / QUANTIZATION–RUNTIME CO-DESIGN

The precision budget follows the full path from weight structure to execution

01

Mixed-format encoding

NINT, NVQ, NPQ, and NEPQ create different quality and size tiers instead of forcing one format on every weight.

02

Precision allocation

Gradient calibration selects precision per compute group; MoE models can also allocate bitrate per expert.

03

Heterogeneous expert container

NINTM v2 stores heterogeneous formats in one MoE tensor so individual experts can receive different precision.

04

Direct packed execution

CUDA kernels and the C++ runtime execute compressed weights directly, with an OpenAI-compatible API and local WebUI.

CURRENT SCOPE

The public implementation covers several recent model architectures

Qwen3.5Qwen3.6 routed MoEGemma4DeepSeek-V4-FlashOpenAI-compatible APIApple Silicon / Metal in development
BOUNDARY

This is an inspectable research prototype, not an unconditional performance promise

Full-model serving currently remains primarily a single-GPU CUDA research prototype. Quality, speed, and memory requirements vary with architecture, precision tier, hardware, and runtime configuration. The Metal HTTP server remains in development.

[ INTERACT TO CREATE LIFE ]