zml/llmd 20261106.0
More models, less waiting.
We are pleased to announce the release of zml/llmd 20261106.0. This week, we are bringing two new models, asynchronous scheduling in beta, and more performance work across our backends.
zml/llmd 20261029.0
Le Chonk.
Oct 29, 2026
We are pleased to announce the release of zml/llmd 20261029.0. This week, we are bringing more models:
zml/llmd 20261023.0
More speed, more models, more fun.
Oct 22, 2026
We are pleased to announce the release of zml/llmd 20261023.0.
zml/llmd 20261009.0
Performance is all around me, and so the feeling grows.
Oct 9, 2026
Another week, another release. We’re happy to announce the release of zml/llmd 20261009.0, which brings considerable performance improvements across the board.
zml/llmd 20260929.0
DeepSeek-V4.1-Flash lands on NVIDIA and AMD.
Sep 29, 2026
We’re happy to announce the release of zml/llmd 20260929.0, which notably brings deepseek-ai/DeepSeek-V4.1-Flash support on NVIDIA and AMD GPUs.
Laguna S 2.1
Poolside's open-weight model for agentic coding and long-horizon work.
Jul 21, 2026
In close collaboration with Poolside, we’re happy to announce the support of Laguna S 2.1, their new open-weight model for agentic coding and long-horizon work.
ZML/LLMD alpha
The universal LLM server.
Jul 8, 2026
Today we’re releasing ZML/LLMD. It’s a self-contained inference server that runs LLaMa, Gemma, Qwen and Mistral LLMs transparently on 5 architectures: NVIDIA CUDA, AMD ROCm, Google TPU, Intel oneAPI and Apple Metal.
10x faster tokenization
Integrating with the IREE tokenizer for a 10x uplift in tokenization performance.
Apr 7, 2026
Today we merged a pretty nice performance improvement to ZML.
Introducing zml-smi
A universal diagnostic and monitoring tool for GPUs, TPUs and NPUs.
Mar 30, 2026
zml-smi is a universal diagnostic and monitoring tool for GPUs, TPUs and NPUs. It provides real-time insights into the performance and health of your hardware.
Introducing ZML/v2
Frontier performance through composability.
Mar 24, 2026
ZML is an inference stack built close to the hardware. It lowers models directly onto NVIDIA, AMD, TPU, and Trainium targets from a single codebase, without depending on and suffering from the Python-heavy runtime layers that most of the ecosystem is built around.