Tencent's research agent trained two open protein models, and you can download both 🧬

I spend a lot of time thinking about AI agents that do real work, not just chat. So this release from Tencent Hunyuan caught my eye. On November 2, the team behind Hyra 2.0, its automated research agent, published two protein models with weights, inference code and technical reports. The interesting part: Tencent says both models were trained by the agent itself.

HyInv: a tiny model that designs sequences

The first model, HyInv, does inverse folding. You give it the backbone of a protein (the 3D shape), and it proposes amino-acid sequences that should fold into that shape. That's a basic building block for drug design, enzyme engineering and building libraries for lab screening.

What I love is how small it is. HyInv is a graph neural network with just 1.96 million parameters, and the weights file is 7.8 MB. It follows the message-passing design of ProteinMPNN, with a 4-layer encoder and a 3-layer decoder, and it only looks at the backbone atoms. It was pretrained with random-order autoregressive modeling and then improved with reinforcement learning, using a reward based on how well the designed sequences refold into the target structure. It runs on CPU or GPU.

Speed is impressive too. According to the technical report, the accelerated BF16 path designs 128 sequences for a 402-residue backbone in about 50.6 ms on a single NVIDIA H20, roughly 52 times faster than the reference sampler. Across five benchmark datasets, the team reports a higher average refolding score than all ten external baselines it tested, at similar or higher sequence diversity.

HyPLM-Flash: a protein language model on a data diet

The second model, HyPLM-Flash, is a 650-million-parameter masked protein language model. Models like this help predict the effect of mutations and support protein design. Tencent says Hyra 2.0 trained it on 0.73 times the data of Meta's ESM-2 650M. In a matched evaluation of 28 public checkpoints on 32 tasks, it reaches a ProteinGym zero-shot Spearman correlation of 0.409, which the team calls statistically tied with ESM-2 650M and other 600M-class models. Its INT8 engine embeds about 130,000 residues per second on one H20.

The same repository also added HyFF, five machine-learning force fields for molecular simulation, on the same day.

The honest caveats

All these numbers come from Tencent itself, and nobody outside has reproduced them yet. The speed results were verified on H20 GPUs only. And a good refolding score on a computer is still a step away from a protein that actually folds and works in a lab; I didn't see wet-lab results in the released material. Open weights plus inference code also isn't a full training reproduction.

Why this matters

Still, this is the kind of agent story I like: not a demo, but downloadable artifacts under the MIT license, with reports you can read and models you can test. If research agents can take a scientific model from idea to release, and then hand it to everyone, that could speed up a lot of small labs and students who could never train these from scratch.

Sources

I'm curious how fast the community will put these small models through independent tests.

Reacties

Een reactie posten

Populaire posts van deze blog

Meta just shared the full recipe for a 30B agent model you can run at home 🧠

When Meta released Muse Glimmer in August, the pitch was simple: an open-weight, Apache 2.0 model built for AI agents that run on your own Mac or PC. What was missing was the "how". On 19 October, Meta Superintelligence Labs published the full technical report, and for anyone who cares about distillation and local agents, it is one of the most useful reads in a while. What the model is Muse Glimmer is a dense model of about 29.6 billion parameters, including a 1.8B perception encoder, so it reads text and images and writes text. It has a 128K+ context window and was trained on data from more than a hundred languages. The design goal was strict: everything has to fit in 32 GB of GPU memory, and a smaller 4-bit variant targets 24 GB cards. The parts I found most interesting Dense beat mixture-of-experts here. That surprised me. The team trained scaling ladders for a dense model and two MoE designs with the same memory footprint. Under a fixed memory ceiling and their compu...
Meer lezen

AI that can actually "see" a crystal 🔬

Most of the solid stuff around us, from battery materials to chips, is made of crystals: atoms stacked in neat, repeating 3D patterns. Large language models are great at reading papers about those materials, but they have never really looked at the structures themselves. A team at MIT just showed a clever way around that, and the trick is one I keep coming back to: good synthetic data. What they built The model is called CLaC, short for Contrastive Language-Crystal model, and it is described in a paper in Nature Communications by Yang Jeong Park, Mayank Kumaran, Chia-Wei Hsu, Elsa Olivetti and Ju Li. MIT News covered it on 18 October 2026. If you know CLIP, the model that taught image generators to connect pictures and words, you already get the idea. CLaC has two branches: a graph neural network that reads the 3D crystal structure, and a text encoder that reads a description. Contrastive training pulls matching structure-text pairs together in one shared space and pushes mismatche...
Meer lezen

A 35B model on a Mac mini, using just 2.9 GiB of memory 🚀

There's something special about running a capable language model on hardware you own: no API bill, no data leaving the house, just a small box on your desk doing the thinking. A paper that's currently #2 on Hugging Face's trending list takes that idea further than I expected: it runs a 35-billion-parameter model on a Mac mini at about 20 tokens per second while using under 3 GiB of memory. The trick: leave the experts on the SSD The paper is called The Other Half of the Memory Wall: Serving 35B MoEs from SSD with Trained Routing Prediction , and the system is named Edge0 . It builds on a Mixture-of-Experts (MoE) model, Qwen3.6-35B-A3B. In an MoE, each token only uses a few "experts" per layer (here 4 out of 256 in each of 40 layers), so most of the weights sit idle at any given moment. Edge0 takes that literally. The expert weights (quantized to int4) stay on the SSD and are streamed in only when needed. The clever part is a small prerouter : a tiny trained he...
Meer lezen