HyFF Andante reports 29x UMA-S-1.2 MD speed: what would you test before switching?
I came across Tencent's HyFF release. The claim that got my attention was the 29x MD speedup, but I'd want to count the evaluation work before calling it a saving. I haven't run it.
The maintainers' README reports Andante at 29x UMA-S-1.2's MD speed, with force MAE of 7.65 versus 7.55 meV/Å. The speed test is a 2,586-atom solvated chorismate system on one NVIDIA H20, using 1 fs NVE with bond constraints and including neighbour-list rebuilds. Accuracy is measured separately on the full OMol25 validation set.
That separation caught my attention. I wouldn't treat a validation-set error and one system's throughput as enough to estimate the cost of moving an existing project. The README also says other GPUs are untested, which matters before assuming the speedup will carry over to the hardware someone already owns.
My instinct would be to start with a small system whose behaviour is already understood, keep the comparison settings explicit, and count the time spent checking the result as part of the migration cost.
For people who actually use ML potentials: what would be your first acceptance test here? Long-trajectory stability, behaviour outside the training distribution, performance on your own GPU, or something else? I'm especially interested in what would make you stop evaluating a model early, before spending days integrating it.
5 points · 2 comments
Tencent Hunyuan released Hyra 2.0, a research agent built to make reusable scientific tools 🔬 It has already trained and open-sourced two protein models, HyInv and HyPLM-Flash, with weights and code.
#AI #AIAgents #Tencent
Tencent-Hunyuan/hyra-results
1 like
Regarde ça 🔥
Hy-Nomas de Tencent est dispo en open-source !
World model interactif : tu génères un monde vidéo segment par segment, tu le diriges en live avec du texte + caméra… et ça tourne sur une seule RTX 4090.
➡️ Tencent-Hunyuan/Hy-Nomas
Apache 2.0, weights dispo.
5 likes · 51 views
Check out Tencent-Hunyuan/Hy-Nomas instead, it's newer
That's from 5-6 months ago. The new kid on the block is Tencent-Hunyuan/Hy-Nomas
Tencent Hy-Nomas to revolutionize Interactive Fiction and worldbuilding?
This one is pretty new - An AI world builder that allows interactive prompt updates between settings - think the AI based stories that you see advertised here - but as interactive 3D worlds - where you can move around and generate new content on the fly.
hy-nomas.hy.tencent.com
Tencent-Hunyuan/Hy-Nomas
This seems to bring us one step closer to the Matrix - I thought I'd post quickly before heading off to bed - and I think I need to update my Comfy desktop to load it - that's next. Seems it can run on a 4090 - so we'll see how my 4070TiS goes.
You can see a demo here:
hy-nomas.hy.tencent.com
Has anyone heard more about this since it's release?
3 points · 3 comments
Surprising how Tencent's HY-World 2.0 open weights world model didn't get as much attention:
Tencent-Hunyuan/HY-World-2.0
Which of these six projects deserves a full deep dive? 👇
Official project links:
1. Hy-MT2 — Tencent-Hunyuan/Hy-MT2
2. Qwen-Image 2.1 — QwenLM/Qwen-Image-2.1
3. REA — morluto/rea
4. AI-agent collaboration — project link pending verification
5. PhotoCraft — photocraft-app/photocraft
6. Xiaomi MiMo-V2.6-Pro — mimo.mi.com
Benchmark reference: artificialanalysis.ai
**Vote in the comments:** `TRANSLATION`, `IMAGE AI`, `REVERSE ENGINEERING`, `AGENTS`, `PHOTOCRAFT`, or `MIMO`.
I'll prioritize the project you want me to test and break down next.
腾讯混元开源了个同传，10 月 22 日挂的仓库，延迟 3 秒，人做同传得 4 到 5 秒，分数还压过五个商业系统!!
结果五天过去，就几颗星。
这俩对不上啊朋友们!!!
Tencent-Hunyuan/Hy-SimulMT
5 likes · 400 views
Hy-Nomas now runs on two more backends.
SGLang (main branch), by @sgl_project : offline video jobs and streaming sessions.
sglang#46124
ComfyUI, by @LatentSpacer : chained segments and WASD camera control.
PedroMarinhoDev/ComfyUI-Hy-Nomas
Thank you both for building on it.
25 likes · 10 reposts · 911 views
Mentions
HyFF Andante reports 29x UMA-S-1.2 MD speed: what would you test before switching? I came across Tencent's HyFF release. The claim that got my attention was the 29x MD speedup, but I'd want to count the evaluation work before calling it a saving. I haven't run it. The maintainers' README reports Andante at 29x UMA-S-1.2's MD speed, with force MAE of 7.65 versus 7.55 meV/Å. The speed test is a 2,586-atom solvated chorismate system on one NVIDIA H20, using 1 fs NVE with bond constraints and including neighbour-list rebuilds. Accuracy is measured separately on the full OMol25 validation set. That separation caught my attention. I wouldn't treat a validation-set error and one system's throughput as enough to estimate the cost of moving an existing project. The README also says other GPUs are untested, which matters before assuming the speedup will carry over to the hardware someone already owns. My instinct would be to start with a small system whose behaviour is already understood, keep the comparison settings explicit, and count the time spent checking the result as part of the migration cost. For people who actually use ML potentials: what would be your first acceptance test here? Long-trajectory stability, behaviour outside the training distribution, performance on your own GPU, or something else? I'm especially interested in what would make you stop evaluating a model early, before spending days integrating it.
Tencent Hunyuan released Hyra 2.0, a research agent built to make reusable scientific tools 🔬 It has already trained and open-sourced two protein models, HyInv and HyPLM-Flash, with weights and code. #AI #AIAgents #Tencent Tencent-Hunyuan/hyra-results
Regarde ça 🔥 Hy-Nomas de Tencent est dispo en open-source ! World model interactif : tu génères un monde vidéo segment par segment, tu le diriges en live avec du texte + caméra… et ça tourne sur une seule RTX 4090. ➡️ Tencent-Hunyuan/Hy-Nomas Apache 2.0, weights dispo.
Check out Tencent-Hunyuan/Hy-Nomas instead, it's newer
That's from 5-6 months ago. The new kid on the block is Tencent-Hunyuan/Hy-Nomas
Tencent Hy-Nomas to revolutionize Interactive Fiction and worldbuilding? This one is pretty new - An AI world builder that allows interactive prompt updates between settings - think the AI based stories that you see advertised here - but as interactive 3D worlds - where you can move around and generate new content on the fly. hy-nomas.hy.tencent.com Tencent-Hunyuan/Hy-Nomas This seems to bring us one step closer to the Matrix - I thought I'd post quickly before heading off to bed - and I think I need to update my Comfy desktop to load it - that's next. Seems it can run on a 4090 - so we'll see how my 4070TiS goes. You can see a demo here: hy-nomas.hy.tencent.com Has anyone heard more about this since it's release?
Surprising how Tencent's HY-World 2.0 open weights world model didn't get as much attention: Tencent-Hunyuan/HY-World-2.0
Which of these six projects deserves a full deep dive? 👇 Official project links: 1. Hy-MT2 — Tencent-Hunyuan/Hy-MT2 2. Qwen-Image 2.1 — QwenLM/Qwen-Image-2.1 3. REA — morluto/rea 4. AI-agent collaboration — project link pending verification 5. PhotoCraft — photocraft-app/photocraft 6. Xiaomi MiMo-V2.6-Pro — mimo.mi.com Benchmark reference: artificialanalysis.ai **Vote in the comments:** `TRANSLATION`, `IMAGE AI`, `REVERSE ENGINEERING`, `AGENTS`, `PHOTOCRAFT`, or `MIMO`. I'll prioritize the project you want me to test and break down next.
腾讯混元开源了个同传，10 月 22 日挂的仓库，延迟 3 秒，人做同传得 4 到 5 秒，分数还压过五个商业系统!! 结果五天过去，就几颗星。 这俩对不上啊朋友们!!! Tencent-Hunyuan/Hy-SimulMT
Hy-Nomas now runs on two more backends. SGLang (main branch), by @sgl_project : offline video jobs and streaming sessions. sglang#46124 ComfyUI, by @LatentSpacer : chained segments and WASD camera control. PedroMarinhoDev/ComfyUI-Hy-Nomas Thank you both for building on it.