I hope everyone appreciates that I didn't refer to Huang's fake girlfriend as "hardware," which perfects the parallelism. The thought is there but describing a person as hardware is gross. I am very restrained.
November 10, 2026 at 11:21 AM
I hope everyone appreciates that I didn't refer to Huang's fake girlfriend as "hardware," which perfects the parallelism. The thought is there but describing a person as hardware is gross. I am very restrained.
genomic parallelism vindicated
November 9, 2026 at 3:44 AM
genomic parallelism vindicated
. o O ( personality parallelism )
November 7, 2026 at 7:19 AM
. o O ( personality parallelism )
I see it as parallelism between "image of God" and "male and female", saying that they are one and the same.
November 8, 2026 at 8:11 PM
I see it as parallelism between "image of God" and "male and female", saying that they are one and the same.
Parallelism in landscape framing creates order in wild scenes. Stacking the horizontal rock layers builds a calm, grounded rhythm that balances the sweeping curves below. #photographytips
November 7, 2026 at 12:14 AM
Parallelism in landscape framing creates order in wild scenes. Stacking the horizontal rock layers builds a calm, grounded rhythm that balances the sweeping curves below. #photographytips
We’re happy to announce the release of HPX 2.0.0, the C++ Standard Library for Parallelism and Concurrency! Read our latest blog for the highlights: hpsf.io/blog/2026/hp...
November 10, 2026 at 7:09 PM
We’re happy to announce the release of HPX 2.0.0, the C++ Standard Library for Parallelism and Concurrency! Read our latest blog for the highlights: hpsf.io/blog/2026/hp...
Susav Shrestha, Brad Settlemyer, Narasimha Reddy: Decode Token Parallelism: Memory-Efficient Distributed Attention Parallelism for Scalable LLM Decoding https://arxiv.org/abs/2611.13293 https://arxiv.org/pdf/2611.13293 https://arxiv.org/html/2611.13293
November 11, 2026 at 6:40 AM
Susav Shrestha, Brad Settlemyer, Narasimha Reddy: Decode Token Parallelism: Memory-Efficient Distributed Attention Parallelism for Scalable LLM Decoding https://arxiv.org/abs/2611.13293 https://arxiv.org/pdf/2611.13293 https://arxiv.org/html/2611.13293
DMuon transforms matrix optimization in AI training by assigning ownership and reducing communication overhead. This innovation reaches 2.1x throughput of current methods, enhancing efficiency for data-parallel runtimes like PyTorch FSDP and speeding training. https://arxiv.org/abs/2611.07084
DMuon: Scaling Matrix Optimizers in Data Parallelism with Topology-Aware Matrix Ownership and Communication
ArXiv link for DMuon: Scaling Matrix Optimizers in Data Parallelism with Topology-Aware Matrix Ownership and Communication
arxiv.org
November 7, 2026 at 11:00 AM
DMuon transforms matrix optimization in AI training by assigning ownership and reducing communication overhead. This innovation reaches 2.1x throughput of current methods, enhancing efficiency for data-parallel runtimes like PyTorch FSDP and speeding training. https://arxiv.org/abs/2611.07084
They should try an Itanium but you have to manually manage your processor cache instead of parallelism
November 10, 2026 at 2:41 PM
They should try an Itanium but you have to manually manage your processor cache instead of parallelism
Secondary #Genesis 4: Epilogue Orantuozegulanaganobulan_4Quashu, The Long Twilight
hf894.blogspot.com/2026/11/seco...
#Myth #Tech Parallelism☠️🤬🐺🤬
The Dismemberment of #Osiris and Postwar Disarmament myth encodes a timeless political-technical logic🦉☠️🐺🦉
theorionblogown.blogspot.com/2025/10/myth...
hf894.blogspot.com/2026/11/seco...
#Myth #Tech Parallelism☠️🤬🐺🤬
The Dismemberment of #Osiris and Postwar Disarmament myth encodes a timeless political-technical logic🦉☠️🐺🦉
theorionblogown.blogspot.com/2025/10/myth...
November 10, 2026 at 9:41 PM
Secondary #Genesis 4: Epilogue Orantuozegulanaganobulan_4Quashu, The Long Twilight
hf894.blogspot.com/2026/11/seco...
#Myth #Tech Parallelism☠️🤬🐺🤬
The Dismemberment of #Osiris and Postwar Disarmament myth encodes a timeless political-technical logic🦉☠️🐺🦉
theorionblogown.blogspot.com/2025/10/myth...
hf894.blogspot.com/2026/11/seco...
#Myth #Tech Parallelism☠️🤬🐺🤬
The Dismemberment of #Osiris and Postwar Disarmament myth encodes a timeless political-technical logic🦉☠️🐺🦉
theorionblogown.blogspot.com/2025/10/myth...
Terraform Weekly #295 - Secretless Azure pipelines, RDS passwords in Secrets Manager, Terraform Kubernetes provider, Kumoss AI, and why parallelism 100 backfires
www.weekly.tf/p/issue-295-...
www.weekly.tf/p/issue-295-...
Issue #295 - Secretless Azure pipelines, RDS passwords in Secrets Manager, Terraform Kubernetes provider, Kumoss AI, and why parallelism 100 backfires
A weekly newsletter about Terraform ecosystem (posts, tools, tips&tricks, open-source) with humble opinions by Anton Babenko.
www.weekly.tf
November 6, 2026 at 1:21 PM
Terraform Weekly #295 - Secretless Azure pipelines, RDS passwords in Secrets Manager, Terraform Kubernetes provider, Kumoss AI, and why parallelism 100 backfires
www.weekly.tf/p/issue-295-...
www.weekly.tf/p/issue-295-...
A little political parallelism.
November 6, 2026 at 8:37 PM
A little political parallelism.
For parallelism, it should be “all important walks” I think
November 3, 2026 at 10:04 PM
For parallelism, it should be “all important walks” I think
threadpoolctl 3.8.0 is out!
Most changes are results of a collaboration between @quansight.com and @probabl.ai as part of the NASA ROSES grant.
This release includes many building blocks to improve thread-based parallelism in the scientific Python stack, in particular w free-threading Python.
Most changes are results of a collaboration between @quansight.com and @probabl.ai as part of the NASA ROSES grant.
This release includes many building blocks to improve thread-based parallelism in the scientific Python stack, in particular w free-threading Python.
November 5, 2026 at 11:21 AM
threadpoolctl 3.8.0 is out!
Most changes are results of a collaboration between @quansight.com and @probabl.ai as part of the NASA ROSES grant.
This release includes many building blocks to improve thread-based parallelism in the scientific Python stack, in particular w free-threading Python.
Most changes are results of a collaboration between @quansight.com and @probabl.ai as part of the NASA ROSES grant.
This release includes many building blocks to improve thread-based parallelism in the scientific Python stack, in particular w free-threading Python.
DeepSeek open-sourced DeepInfer (Apache-2.0), the inference framework behind its own services: expert parallelism, speculative decoding, deterministic inference.
The thread's question: if inference is this efficient, how low can prices go?
https://github.com/deepseek-ai/DeepInfer
The thread's question: if inference is this efficient, how low can prices go?
https://github.com/deepseek-ai/DeepInfer
November 10, 2026 at 5:23 PM
DeepSeek open-sourced DeepInfer (Apache-2.0), the inference framework behind its own services: expert parallelism, speculative decoding, deterministic inference.
The thread's question: if inference is this efficient, how low can prices go?
https://github.com/deepseek-ai/DeepInfer
The thread's question: if inference is this efficient, how low can prices go?
https://github.com/deepseek-ai/DeepInfer
Google Cloud's MaxText framework now supports long-context training up to 14 million tokens on Cloud TPUs.
Source: Google Developers
Source: Google Developers
Announcing Long-Context Training on TPUs with MaxText Context Parallelism- Google Developers Blog
Learn how MaxText enables efficient long-context LLM training up to 14M tokens on Cloud TPUs using advanced context parallelism strategies and Tokamax Splash Attention.
developers.googleblog.com
November 9, 2026 at 9:52 PM
Google Cloud's MaxText framework now supports long-context training up to 14 million tokens on Cloud TPUs.
Source: Google Developers
Source: Google Developers
Minimize $ spent on Apple compute, it is a terrible value when you consider that MacOS cuts parallelism potential by like 4x+
November 9, 2026 at 1:30 AM
Minimize $ spent on Apple compute, it is a terrible value when you consider that MacOS cuts parallelism potential by like 4x+
Resource-Efficient Reinforcement Learning Post-Training through Elastic Expert Parallelism
Jia Wei et al.
#arXiv #cs.AI #cs.DC
Jia Wei et al.
#arXiv #cs.AI #cs.DC
Resource-Efficient Reinforcement Learning Post-Training through Elastic Expert Parallelism
Reinforcement learning post-training improves the capabilities of large language models, but generating training responses is a major computational bottleneck. In mixture-of-experts models, workers that finish their assigned responses may remain occupied because their experts are still needed by wo…
arxiv.org
November 6, 2026 at 11:39 PM
Those changes will be leveraged by the following PR in joblib:
github.com/joblib/jobli...
Early benchmark results show that free-threading-based parallelism can finally beat process-based parallelism for real ML workloads involving nested parallelism between Python and BLAS/OpenMP.
github.com/joblib/jobli...
Early benchmark results show that free-threading-based parallelism can finally beat process-based parallelism for real ML workloads involving nested parallelism between Python and BLAS/OpenMP.
PERF Limit OpenMP and BLAS in threaded backend as well by itamarst · Pull Request #1871 · joblib/joblib
Fixes #1846
Assumes joblib/threadpoolctl#257 will get merged; if not, or if API changes, this will need to change too. It is compatible with older threadpoolctl though.
PR in changelog
Figure ou...
github.com
November 5, 2026 at 11:21 AM
Those changes will be leveraged by the following PR in joblib:
github.com/joblib/jobli...
Early benchmark results show that free-threading-based parallelism can finally beat process-based parallelism for real ML workloads involving nested parallelism between Python and BLAS/OpenMP.
github.com/joblib/jobli...
Early benchmark results show that free-threading-based parallelism can finally beat process-based parallelism for real ML workloads involving nested parallelism between Python and BLAS/OpenMP.
The scaling laws are great, but optimizing inference latency for complex agentic loops often forces brutal architectural tradeoffs. Is sparsity the future, or will better parallelism win?
November 11, 2026 at 7:20 AM
The scaling laws are great, but optimizing inference latency for complex agentic loops often forces brutal architectural tradeoffs. Is sparsity the future, or will better parallelism win?
Resource-Efficient Reinforcement Learning Post-Training through Elastic Expert Parallelism
Jia Wei et al.
#arXiv #cs.AI #cs.DC
Jia Wei et al.
#arXiv #cs.AI #cs.DC
Resource-Efficient Reinforcement Learning Post-Training through Elastic Expert Parallelism
Reinforcement learning post-training improves the capabilities of large language models, but generating training responses is a major computational bottleneck. In mixture-of-experts models, workers that finish their assigned responses may remain occupied because their experts are still needed by wo…
arxiv.org
November 7, 2026 at 12:38 AM
The em dash thing has been known forever, but I didn't realize Oxford commas were also a red flag for AI-written content.
It was bad enough being concious every time I used an em dash, I don't want to start doubting my Oxford commas now too.
It was bad enough being concious every time I used an em dash, I don't want to start doubting my Oxford commas now too.
November 5, 2026 at 2:34 PM
The em dash thing has been known forever, but I didn't realize Oxford commas were also a red flag for AI-written content.
It was bad enough being concious every time I used an em dash, I don't want to start doubting my Oxford commas now too.
It was bad enough being concious every time I used an em dash, I don't want to start doubting my Oxford commas now too.
Speculative decoding in LLMs: Use a cheap assistant to draft and a smart reviewer to verify. Parallelism makes batch validation faster than sequential guessing.
#AI #SystemDesign #DistributedSystems
#AI #SystemDesign #DistributedSystems
November 9, 2026 at 11:02 PM
Speculative decoding in LLMs: Use a cheap assistant to draft and a smart reviewer to verify. Parallelism makes batch validation faster than sequential guessing.
#AI #SystemDesign #DistributedSystems
#AI #SystemDesign #DistributedSystems
NASA Curiosity Blog: Sols 5056-5062 Charging Back Up
Curiosity completed three power-positive sols before resuming drilling at the Nazca target, including remote science and a new parallelism engineering test on Sol 5061.
Curiosity completed three power-positive sols before resuming drilling at the Nazca target, including remote science and a new parallelism engineering test on Sol 5061.
NASA Curiosity Blog: Sols 5056-5062 Charging Back Up
Curiosity completed three power-positive sols before resuming drilling at the Nazca target, including remote science and a new parallelism engineering test on Sol 5061.
hermes-vector.buzz
November 11, 2026 at 4:49 AM
NASA Curiosity Blog: Sols 5056-5062 Charging Back Up
Curiosity completed three power-positive sols before resuming drilling at the Nazca target, including remote science and a new parallelism engineering test on Sol 5061.
Curiosity completed three power-positive sols before resuming drilling at the Nazca target, including remote science and a new parallelism engineering test on Sol 5061.
AutoParallel automatically optimizes distributed training parallelism in PyTorch, matching or beating hand-written expert recipes and discovering novel strategies that scale efficiently to 128 H100s.
https://arxiv.org/abs/2611.08660
#cloud #AWS
https://arxiv.org/abs/2611.08660
#cloud #AWS
November 11, 2026 at 7:30 AM
AutoParallel automatically optimizes distributed training parallelism in PyTorch, matching or beating hand-written expert recipes and discovering novel strategies that scale efficiently to 128 H100s.
https://arxiv.org/abs/2611.08660
#cloud #AWS
https://arxiv.org/abs/2611.08660
#cloud #AWS