#fp4

The latest AI and developer news about #fp4, with the original source and a short summary.

The official DeepSeek V4 paper has been released, adding FP4 QAT and stabilization techniques.

·2026.05.09 17:00

The Dell Pro Max GB10 ran a 70B-class LLM locally in real time using 128GB of unified memory.

·2026.05.08 13:00

Different FP4 inference support has been added to llama.cpp and ik_llama.cpp.

·2·2026.04.26 00:00

On 4x Tesla M40 12GB with an fp4 setup, Qwen3.6 27B speed was reported at 26 tok/s PP and 8 tok/s TG.

·1·2026.04.25 00:00

Explains the bit composition, value range, and E2M1 format of FP4.

·2·2026.04.19 02:00

DeepSeek has updated DeepGEMM with Blackwell optimizations and Mega MoE support.

·2026.04.16 20:00

NVIDIA Sol-RL reduces cost and accelerates convergence by separating FP4 rollouts from BF16 updates.

·2026.04.10 09:00