mxfp8

cuDNN Frontend is NVIDIA's modern, open-source entry point to the cuDNN library and a growing collection of high-performance open-source kernels.

LLM fine-tuning with LoRA + NVFP4/MXFP8 on NVIDIA DGX Spark (Blackwell GB10)

Easy recipes to speed up latency of Flux, QwenImage, and LTX-2 with NVFP4 and MXFP8 on Blackwell.

Patches + recipe to deploy festr2/MiMo-V2.5-Pro-NVFP4-MXFP8-attn-TP8 on 8-node DGX Spark sm_121 (Ray + vLLM, TP=8). Fixes the fused-qkv loader bug that mis-slotted Q values as K/V on 7 of 8 ranks.

Three ComfyUI nodes that show the real runtime precision of MODEL, CLIP, and VAE objects.

Block-scaled FP8 / FP4 / INT4 tensor primitive with Triton scaled-matmul at FP32 parity on H100. NumPy / PyTorch / MLX / JAX backends.

Blackwell-optimized llama.cpp bundle with DFlash2, MXFP6/MXFP8/NVFP4, TurboQuant KV, and GPU-resident speculative handoff.

Dual DGX Spark TP=2 serving for GLM-5.3 Flash with an NVFP4 target and MXFP8 DFlash2 draft

MXFP/NVFP/FP16 to GGUF conversion with automatic safe routing