#TwoTower
LightNews LightNews
sustainablex.in
💣💣💣 Breaking News: NVIDIA unleashes Nemotron-TwoTower-30B-A3B-Base-BF16, a game-changing diffusion-based language model.

💣💣💣 Say goodbye to slow generation times!

This beast retains 98.7% of baseline quality, yet cranks out results 2
NVIDIA has released Nemotron-TwoTower-30B-A3B-Base-BF16, an unusual diffusion-ba
October 27, 2026 at 4:36 AM
aub.bsky.social
they did bicameralism on a diffusion model
NVIDIA Nemotron-Labs-TwoTower: 2.42× Diffusion LLM Guide (2026) | explainx.ai Blog
NVIDIA Nemotron-Labs-TwoTower — split 30B Nemotron Nano into context + denoiser towers. 98.7% quality, 2.42× speed. Hugging Face, vLLM, mask diffusion explained.
www.explainx.ai
September 17, 2026 at 5:25 AM
mustaphaunubimomoh.bsky.social
#RecommenderSystems #recommendations #MLOps #TwoTower #recsys #machinelearning
May 21, 2026 at 3:34 PM
moisesjafet.com
"Autoregressive (AR) models decode one token at a time. That serial process caps generation throughput. Discrete diffusion language models take another route. They generate tokens in parallel and refine them iteratively."
www.marktechpost.com/2026/07/01/n....
NVIDIA Releases Nemotron-Labs-TwoTower: an Open-Weight Diffusion Language Model Built on a Frozen Autoregressive Nemotron-3-Nano-30B-A3B Backbone
NVIDIA has released Nemotron-Labs-TwoTower, a diffusion language model built on a pretrained autoregressive backbone. It ships as open weights under the NVIDIA Nemotron Open Model License. The release...
www.marktechpost.com
July 4, 2026 at 3:27 PM
rbd.bsky.social
Nvidia apparently working on diffusion too, here's a 30B-A3B version of Nemotron 3 that they claim is ~2.4× faster than the regular autoregressive version: huggingface.co/nvidia/Nemot...
nvidia/Nemotron-TwoTower-30B-A3B-Base-BF16 · Hugging Face
We’re on a journey to advance and democratize artificial intelligence through open source and open science.
huggingface.co
June 25, 2026 at 4:45 PM
aidailypost.com
🚀 NVIDIA just dropped Nemotron‑Labs‑TwoTower, a diffusion model powered by a 128‑expert MoE. Open‑source and ready to remix your AI pipelines. Curious? Dive in for the details! #NVIDIA #TwoTower #MoE

🔗 aidailypost.com/news/nvidia-...
July 1, 2026 at 8:47 AM
qdrddr.bsky.social
#Nvidia TwoTower: split a #model into two:
#Autoregressive and #Diffusion gaining x2.4 throughout, sacrificing 1-3% accuracy.
#LLM #AI
#OpenSource nvidia-open-model lic
huggingface.co/nvidia/Nemot...
nvidia/Nemotron-Labs-TwoTower-30B-A3B-Base-BF16 · Hugging Face
We’re on a journey to advance and democratize artificial intelligence through open source and open science.
huggingface.co
July 2, 2026 at 2:24 PM
arxiv-cs-cl.bsky.social
Fitsum Reda, John Kamalu, Roger Waleffe, Mostofa Patwary, Mohammad Shoeybi, Bryan Catanzaro
Nemotron-TwoTower: Diffusion Language Modeling with Pretrained Autoregressive Context
https://arxiv.org/abs/2606.26493
June 26, 2026 at 12:15 PM
ebibibibibibi.bsky.social
📎 NVIDIAの拡散型LLM「Nemotron-Labs-TwoTower」、既存モデルの上に載せるだけで生成速度2.42倍に
https://www.ebisuda.net/tech/2026/07/25/nvidiallmnemotron-labs-twotower242-nvidia-releases-nemotron-labs-twotower-an-ope/
July 25, 2026 at 11:30 AM
ebibibibibibi.bsky.social
既存モデルに"載せるだけ"で生成速度2.42倍に⚡

NVIDIAが拡散型LLM「Nemotron-Labs-TwoTower」を公開。追加学習だけで既存モデルを高速化できるという発想が面白い🤖

詳しくはこちら👇
https://www.ebisuda.net/tech/2026/07/25/nvidiallmnemotron-labs-twotower242-nvidia-releases-nemotron-labs-twotower-an-ope/

#TechNews #生成AI
July 25, 2026 at 5:45 AM
sungkim.bsky.social
They found it kept 98.7% of the original model’s quality at 2.42× faster generation.

Model: huggingface.co/nvidia/Nemot...
Paper: arxiv.org/abs/2606.26493
nvidia/Nemotron-Labs-TwoTower-30B-A3B-Base-BF16 · Hugging Face
We’re on a journey to advance and democratize artificial intelligence through open source and open science.
huggingface.co
July 2, 2026 at 6:36 PM
sungkim.bsky.social
Nvidia's Nemotron-Labs-TwoTower

A diffusion language model adapted from Nemotron-3-Nano-30B-A3B. Here’s how it works: one half holds the context, the other writes the tokens, with both reusing the pretrained model instead of training a new one from scratch.
July 2, 2026 at 6:36 PM
ryanhe.mistyreverie.org.ap.brid.gy
NVIDIA 發佈 Nemotron-TwoTower-30B-A3B-Base 大語言擴散模型
https://www.reddit.com/r/LocalLLaMA/comments/1uf4azy/nvidia_has_released/
NVIDIA 近期發佈了基於 Nemotron 3 Nano 30B-A3B 構建的 Nemotron-TwoTower-30B-A3B-Base 模型。該模型採用獨特的擴散模型（Diffusion-based）架構，結合了凍結的自回歸上下文塔與擴散去噪塔，使其能夠平行填充 Token 塊，而非傳統的逐字生成。實驗數據顯示，此架構在保留 98 […]
Original post on mistyreverie.org
mistyreverie.org
June 26, 2026 at 9:29 AM
jeremiahgilfoy.bsky.social
Fight for your own, your people. Your community. #TwoTower #LordOfTheRing
a man with long hair and a beard is standing in a dark room with his arms outstretched
ALT: a man with long hair and a beard is standing in a dark room with his arms outstretched
media.tenor.com
March 15, 2026 at 4:35 AM
cscl-bot.bsky.social
Fitsum Reda, John Kamalu, Roger Waleffe, Mostofa Patwary, Mohammad Shoeybi, Bryan Catanzaro: Nemotron-TwoTower: Diffusion Language Modeling with Pretrained Autoregressive Context https://arxiv.org/abs/2606.26493 https://arxiv.org/pdf/2606.26493 https://arxiv.org/html/2606.26493
June 26, 2026 at 6:39 AM
buhane.com.tr
Nemotron-Labs-TwoTower: Diffusion Speed On A Frozen 30B Backbone NVIDIA’s new Nemotron-Labs-TwoTower diffusion language model significantly advances AI text generation. It uses a frozen.... @thecosmicmeta.com #AI

https://u2m.io/x5JFlnow
Nemotron-Labs-TwoTower: Diffusion Speed On A Frozen 30B Backbone
NVIDIA's new Nemotron-Labs-TwoTower diffusion language model significantly advances AI text generation. It uses a frozen Nemotron-3-Nano-30B-A3B backbone and an innovative
thecosmicmeta.com
July 1, 2026 at 10:20 PM
attapi.bsky.social
anyone reading or messing with the twitter simclusters paper?

#simcluster #simclusters #recsys #algorithm #twhin #recommendationsystem #machinelearning #twotower #graph
April 19, 2026 at 3:39 PM