Tech Meridian ← ENTITY INDEX
PROMY MERIDIAN RU

COMPANY · ENTITY #11944

MXFP8

Related event timeline, sources and context from the news index.

EVENT TIMELINE

3

MODELS · 2 SOURCES · Hugging Face · The Decoder

Thinking Machines Lab releases Inkling 2 Mini — looped MoE model for local use

Inkling 2 Mini is a new multimodal looped transformer Mixture-of-Experts from Thinking Machines Lab (~26B physical params, ~3B active per token) that reuses 24 physical layers with a triple-pass loop to behave like a deeper model, supports 1M-token context, image+text inputs, mixed-precision NVFP4/MXFP8 expert weights, and a DSpark drafter; a variable-loop research checkpoint (TML-d24-varloop-research) is also provided. The model is packaged for local and datacenter inference with transformers, llama.cpp, MLX and SGLang support and is reported to lead many benchmarks in its class.

7.0

RESEARCH · 1 SOURCE · arXiv cs.AI

Paper evaluates MXFP6/MXFP8 quantization for R&R on AMD MI350X, reports measured GEMM speedups

arXiv:2611.17501v1 presents an empirical evaluation of MXFP8 and a native 6-bit MXFP6 format for industrial-scale recommendation & ranking (R&R) inference on the AMD MI350X. The authors report MXFP8 GEMM kernel speedups over FP16 of 1.33–1.55x and MXFP6 speedups of 1.73–1.99x (up to 2.90x on some layers), with end-to-end throughput gains from selective MXFP6 of 1.5–5.3% for one production ads ranking model (Model A) and 7.6–14.3% for a second (Model B), while respecting a stringent 0.01% Normalized Entropy (NE) parity target.

7.0

MODELS · 1 SOURCE · Hugging Face

Olmo launches Olmo-core 3 — open, scalable MoE training infrastructure

Olmo has released Olmo-core 3, a redesigned open mixture-of-experts (MoE) training framework intended to scale MoE models into the trillion-parameter range while preserving computational efficiency. The stack replaces an FSDP-based implementation with a DDP-centered system, combines expert/pipeline parallelism and other optimizations, reports a 2.7× throughput uplift on a 47B-parameter MoE across eight NVIDIA B300 GPUs, and shows MXFP8 delivering about 21% higher throughput than BF16 in controlled tests.

7.0