PyTorch is reaching more hardware, with caveats.
TorchTitan reported shorter checkpoint times in a 512-GPU DeepSeek V3 test. FlyDSL speeds select operations on AMD MI350 GPUs; TorchTPU runs PyTorch on Google TPUs without JAX. Kernel coverage and distributed performance…
Read more on Kimbodo:
TorchTitan reported shorter checkpoint times in a 512-GPU DeepSeek V3 test. FlyDSL speeds select operations on AMD MI350 GPUs; TorchTPU runs PyTorch on Google TPUs without JAX. Kernel coverage and distributed performance…
Read more on Kimbodo:
PyTorch Is Expanding Across GPUs and TPUs: What It Means for AI Infrastructure Choices
What Happened Recent PyTorch ecosystem updates focus on training at scale and running workloads across more hardware. TorchTitan added broader model support, 5D parallelism, performance optimizations, and TitanRL for asynchronous…
kimbodo.com
November 5, 2026 at 3:50 PM
PyTorch is reaching more hardware, with caveats.
TorchTitan reported shorter checkpoint times in a 512-GPU DeepSeek V3 test. FlyDSL speeds select operations on AMD MI350 GPUs; TorchTPU runs PyTorch on Google TPUs without JAX. Kernel coverage and distributed performance…
Read more on Kimbodo:
TorchTitan reported shorter checkpoint times in a 512-GPU DeepSeek V3 test. FlyDSL speeds select operations on AMD MI350 GPUs; TorchTPU runs PyTorch on Google TPUs without JAX. Kernel coverage and distributed performance…
Read more on Kimbodo:
#SemiAnalysis on Amazon #Trainium3
Are Amazon successfully delivering a true all-rounder?
open.substack.com/pub/semianal...
Are Amazon successfully delivering a true all-rounder?
open.substack.com/pub/semianal...
Amazon’s Trainium Takes a Strike at the Training King
Trainium Best for Inference? But now good for Training too, Native Trainium PyTorch, High Quality CI, TorchTitan, GPU Impossible Profiler
newsletter.semianalysis.com
November 5, 2026 at 7:53 PM
#SemiAnalysis on Amazon #Trainium3
Are Amazon successfully delivering a true all-rounder?
open.substack.com/pub/semianal...
Are Amazon successfully delivering a true all-rounder?
open.substack.com/pub/semianal...
Zephon keeps TorchTitan batches stable across GPU counts.
In a reported test, a 16-GPU checkpoint resumed on 8, 32 or 64 GPUs with identical global-batch hashes. In-run data prep avoids separate corpora per experiment and helps isolate data-order effects, but cannot ensure…
Read more on Kimbodo:
In a reported test, a 16-GPU checkpoint resumed on 8, 32 or 64 GPUs with identical global-batch hashes. In-run data prep avoids separate corpora per experiment and helps isolate data-order effects, but cannot ensure…
Read more on Kimbodo:
How Deterministic Data Loading Makes PyTorch Training Easier to Scale and Reproduce
What Happened Zephon brings data preparation into TorchTitan training runs: it tokenizes, packs, shuffles, and mixes inputs without requiring a separate preprocessed corpus for every experiment. Its pipeline separates data…
kimbodo.com
October 30, 2026 at 6:50 PM
Zephon keeps TorchTitan batches stable across GPU counts.
In a reported test, a 16-GPU checkpoint resumed on 8, 32 or 64 GPUs with identical global-batch hashes. In-run data prep avoids separate corpora per experiment and helps isolate data-order effects, but cannot ensure…
Read more on Kimbodo:
In a reported test, a 16-GPU checkpoint resumed on 8, 32 or 64 GPUs with identical global-batch hashes. In-run data prep avoids separate corpora per experiment and helps isolate data-order effects, but cannot ensure…
Read more on Kimbodo:
Prime Intellect just dropped a trainable Mega MoE kernel that's 3.11× faster on the forward pass—speed boost for Qwen3, TorchTitan and any MoE training. Curious how it reshapes backward pass performance? Dive in! #PrimeIntellect #MegaMoE #MoETraining
🔗 aidailypost.com/news/prime-i...
🔗 aidailypost.com/news/prime-i...
October 29, 2026 at 8:56 PM
Prime Intellect just dropped a trainable Mega MoE kernel that's 3.11× faster on the forward pass—speed boost for Qwen3, TorchTitan and any MoE training. Curious how it reshapes backward pass performance? Dive in! #PrimeIntellect #MegaMoE #MoETraining
🔗 aidailypost.com/news/prime-i...
🔗 aidailypost.com/news/prime-i...
Prime Intellect just dropped a trainable Mega MoE kernel that's 3.11× faster on the forward pass—speed boost for Qwen3, TorchTitan and any MoE training. Curious how it reshapes backward pass performance? Dive in! #PrimeIntellect #MegaMoE #MoETraining
🔗 aidailypost.com/news/prime-i...
🔗 aidailypost.com/news/prime-i...
October 29, 2026 at 8:56 PM
Prime Intellect just dropped a trainable Mega MoE kernel that's 3.11× faster on the forward pass—speed boost for Qwen3, TorchTitan and any MoE training. Curious how it reshapes backward pass performance? Dive in! #PrimeIntellect #MegaMoE #MoETraining
🔗 aidailypost.com/news/prime-i...
🔗 aidailypost.com/news/prime-i...
🤖 **Train Qwen3 on AWS Trainium with torchtitan**
Train Qwen3 on AWS Trainium with torchtitan three ways on a single trn2.48xlarge: the Hugging Face Transformers v5 modeling backend, torchtitan's native Qwen3, and NKI mixture-of-experts kernels. T...
📰 Source: Artificial Intelligence
🔗 Link […]
Train Qwen3 on AWS Trainium with torchtitan three ways on a single trn2.48xlarge: the Hugging Face Transformers v5 modeling backend, torchtitan's native Qwen3, and NKI mixture-of-experts kernels. T...
📰 Source: Artificial Intelligence
🔗 Link […]
Original post on igeek.gamer-geek-news.com
igeek.gamer-geek-news.com
October 23, 2026 at 8:17 PM
🤖 **Train Qwen3 on AWS Trainium with torchtitan**
Train Qwen3 on AWS Trainium with torchtitan three ways on a single trn2.48xlarge: the Hugging Face Transformers v5 modeling backend, torchtitan's native Qwen3, and NKI mixture-of-experts kernels. T...
📰 Source: Artificial Intelligence
🔗 Link […]
Train Qwen3 on AWS Trainium with torchtitan three ways on a single trn2.48xlarge: the Hugging Face Transformers v5 modeling backend, torchtitan's native Qwen3, and NKI mixture-of-experts kernels. T...
📰 Source: Artificial Intelligence
🔗 Link […]
AWS Trainiumとtorchtitanを用いた大規模言語モデル分散学習の全貌
AWS Trainiumとtorchtitanを活用した大規模モデルの効率的な分散学習手法について解説します。
#AWSTrainium #torchtitan #LLM #DistributedTraining #PyTorch
AWS Trainiumとtorchtitanを活用した大規模モデルの効率的な分散学習手法について解説します。
#AWSTrainium #torchtitan #LLM #DistributedTraining #PyTorch
AWS Trainiumとtorchtitanを用いた大規模言語モデル分散学習の全貌
AWS Trainiumとtorchtitanを活用した大規模モデルの効率的な分散学習手法について解説します。
ai.warp-studio.com
October 21, 2026 at 9:04 PM
AWS Trainiumとtorchtitanを用いた大規模言語モデル分散学習の全貌
AWS Trainiumとtorchtitanを活用した大規模モデルの効率的な分散学習手法について解説します。
#AWSTrainium #torchtitan #LLM #DistributedTraining #PyTorch
AWS Trainiumとtorchtitanを活用した大規模モデルの効率的な分散学習手法について解説します。
#AWSTrainium #torchtitan #LLM #DistributedTraining #PyTorch
AWS swaps PyTorch/XLA for a native PyTorch backend, TorchNeuron, so Trainium chips can more easily run code written for Nvidia's CUDA
Amazon’s Trainium Takes a Strike at the Training King
Trainium Best for Inference? But now good for Training too, Native Trainium PyTorch, High Quality CI, TorchTitan, GPU Impossible Profiler
newsletter.semianalysis.com
October 23, 2026 at 11:00 PM
AWS swaps PyTorch/XLA for a native PyTorch backend, TorchNeuron, so Trainium chips can more easily run code written for Nvidia's CUDA
AWS Machine Learning: Train Qwen3 on AWS Trainium with torchtitan
https://ai-news.ypetit.net/?a=0extd3o9cf7
https://ai-news.ypetit.net/?a=0extd3o9cf7
October 21, 2026 at 9:14 PM
AWS Machine Learning: Train Qwen3 on AWS Trainium with torchtitan
https://ai-news.ypetit.net/?a=0extd3o9cf7
https://ai-news.ypetit.net/?a=0extd3o9cf7
Train Qwen3 on AWS Trainium with torchtitan #machinelearning #ai
Train Qwen3 on AWS Trainium with torchtitan
Train Qwen3 on AWS Trainium with torchtitan
0 views
Skip to first unread message
Eyal Estrin
unread,
4:36 AM (30 minutes ago)
to
https://aws.amazon.com/blogs/machine-learning/train-qwen3-on-aws-trainium-with-torchtitan/
Eyal Estrin
Author | Cloud Architect | AWS • Azure • GCP Insights
Social: @eyalestrin
Connect: https://linktr.ee/eyalestrin Blog: https://security-24-7.com
Reply all
Reply to author
Forward
dlvr.it
October 22, 2026 at 9:05 AM
Train Qwen3 on AWS Trainium with torchtitan #machinelearning #ai
This article demonstrates three methods to train Qwen3 on AWS Trainium using torchtitan, including Hugging Face Transformers v5, native torchtitan models, and custom NKI kernels.
Train Qwen3 on AWS Trainium with torchtitan
This article demonstrates three methods to train Qwen3 on AWS Trainium using torchtitan, including Hugging Face Transformers v5, native torchtitan models, and custom NKI kernels.
aws-news.com
October 21, 2026 at 8:01 PM
This article demonstrates three methods to train Qwen3 on AWS Trainium using torchtitan, including Hugging Face Transformers v5, native torchtitan models, and custom NKI kernels.
3,500+ attendees came together to accelerate open intelligence at #PyTorchCon 2026.
Catch up on new members, TorchTPU, TorchNeuron, TorchTitan, and community highlights:
https://pytorch.org/blog/pytorch-conference-north-america-2026-accelerating-open-intelligence-together/
Catch up on new members, TorchTPU, TorchNeuron, TorchTitan, and community highlights:
https://pytorch.org/blog/pytorch-conference-north-america-2026-accelerating-open-intelligence-together/
October 22, 2026 at 11:58 PM
3,500+ attendees came together to accelerate open intelligence at #PyTorchCon 2026.
Catch up on new members, TorchTPU, TorchNeuron, TorchTitan, and community highlights:
https://pytorch.org/blog/pytorch-conference-north-america-2026-accelerating-open-intelligence-together/
Catch up on new members, TorchTPU, TorchNeuron, TorchTitan, and community highlights:
https://pytorch.org/blog/pytorch-conference-north-america-2026-accelerating-open-intelligence-together/
Integrating TorchTPU into TorchTitan and Optimizing Models for TPU #googlecloud #ai https://developers.googleblog.com/integrating-torchtpu-into-torchtitan-and-optimizing-models-for-tpu/
Integrating TorchTPU into TorchTitan and Optimizing Models for TPU- Google Developers Blog
developers.googleblog.com
October 21, 2026 at 10:02 AM
Integrating TorchTPU into TorchTitan and Optimizing Models for TPU #googlecloud #ai https://developers.googleblog.com/integrating-torchtpu-into-torchtitan-and-optimizing-models-for-tpu/
📰 New article by Josh Longenecker, Alex Hoffman
Train Qwen3 on AWS Trainium with torchtitan
#AWS #AI #MachineLearning
Train Qwen3 on AWS Trainium with torchtitan
#AWS #AI #MachineLearning
Train Qwen3 on AWS Trainium with torchtitan
Train Qwen3 on AWS Trainium with torchtitan three ways on a single trn2.48xlarge: the Hugging Face Transformers v5 modeling backend, torchtitan’s native Qwen3, and NKI mixture-of-experts kernels. The same torchrun launcher and config structure work unchanged across all three paths, including loading and saving published Hugging Face safetensors.
aws.amazon.com
October 21, 2026 at 8:06 PM
📰 New article by Josh Longenecker, Alex Hoffman
Train Qwen3 on AWS Trainium with torchtitan
#AWS #AI #MachineLearning
Train Qwen3 on AWS Trainium with torchtitan
#AWS #AI #MachineLearning
🤖 **Train Qwen3 on AWS Trainium with torchtitan**
Train Qwen3 on AWS Trainium with torchtitan three ways on a single trn2.48xlarge: the Hugging Face Transformers v5 modeling backend, torchtitan's native Qwen3, and NKI mixture-of-experts kernels. T...
📰 Source: Artificial Intelligence
🔗 Link […]
Train Qwen3 on AWS Trainium with torchtitan three ways on a single trn2.48xlarge: the Hugging Face Transformers v5 modeling backend, torchtitan's native Qwen3, and NKI mixture-of-experts kernels. T...
📰 Source: Artificial Intelligence
🔗 Link […]
Original post on igeek.gamer-geek-news.com
igeek.gamer-geek-news.com
October 21, 2026 at 8:10 PM
🤖 **Train Qwen3 on AWS Trainium with torchtitan**
Train Qwen3 on AWS Trainium with torchtitan three ways on a single trn2.48xlarge: the Hugging Face Transformers v5 modeling backend, torchtitan's native Qwen3, and NKI mixture-of-experts kernels. T...
📰 Source: Artificial Intelligence
🔗 Link […]
Train Qwen3 on AWS Trainium with torchtitan three ways on a single trn2.48xlarge: the Hugging Face Transformers v5 modeling backend, torchtitan's native Qwen3, and NKI mixture-of-experts kernels. T...
📰 Source: Artificial Intelligence
🔗 Link […]
TorchTPU is now open source: Run native PyTorch on TPUs #machinelearning #ai
TorchTPU is now open source: Run native PyTorch on TPUs
Groups
Sign in
TorchTPU is now open source: Run native PyTorch on TPUs
0 views
Eyal Estrin
unread,
3:36 AM (29 minutes ago)
to
https://cloud.google.com/blog/topics/ai-infrastructure/torchtpu-is-now-open-source-run-native-pytorch-on-tpus
https://developers.googleblog.com/integrating-torchtpu-into-torchtitan-and-optimizing-models-for-tpu/
https://developers.googleblog.com/announcing-new-vllm-and-sglang-inference-engines-on-tpu/
https://www.helpnetsecurity.com/2026/10/21/torchtpu-pytorch-on-tpus/
Eyal Estrin
Author | Cloud Architect | AWS • Azure • GCP Insights
Social: @eyalestrin
Connect: https://linktr.ee/eyalestrin Blog: https://security-24-7.com
Reply all
Reply to author
Forward
dlvr.it
October 22, 2026 at 9:05 AM
TorchTPU is now open source: Run native PyTorch on TPUs #machinelearning #ai
The TorchTitan team landed massive improvements in the past year without losing its minimal, extensible ethos. Enhancements include support for SOTA models, dramatic improvements of out-the-box performance at scale, RL support, and new communication features. Learn more: pytorch.org/blog/torchti...
October 21, 2026 at 11:53 PM
The TorchTitan team landed massive improvements in the past year without losing its minimal, extensible ethos. Enhancements include support for SOTA models, dramatic improvements of out-the-box performance at scale, RL support, and new communication features. Learn more: pytorch.org/blog/torchti...
Google says TorchTitan recipes have been validated on up to 8,000 Ironwood TPUs. This quarter, testing distributed workloads alongside single-device ports will reveal whether easier migration also preserves useful performance at scale.
October 21, 2026 at 11:38 AM
Google says TorchTitan recipes have been validated on up to 8,000 Ironwood TPUs. This quarter, testing distributed workloads alongside single-device ports will reveal whether easier migration also preserves useful performance at scale.
Amazon says TorchNeuron integrates with TorchTitan for distributed training. This week, testing multi-instance scaling alongside a single-chip port can show whether the lower migration effort survives production workload sizes.
October 20, 2026 at 10:31 AM
Amazon says TorchNeuron integrates with TorchTitan for distributed training. This week, testing multi-instance scaling alongside a single-chip port can show whether the lower migration effort survives production workload sizes.
Google open-sourced TorchTPU, so PyTorch on TPUs just got a lot easier
Google released TorchTPU as open source, giving PyTorch developers a native way to run models on the company’s Tensor Processing Units. Google also announced TPU support in the vLLM and SGLang serv…
#hackernews #huggingface #llm
Google released TorchTPU as open source, giving PyTorch developers a native way to run models on the company’s Tensor Processing Units. Google also announced TPU support in the vLLM and SGLang serv…
#hackernews #huggingface #llm
Google open-sourced TorchTPU, so PyTorch on TPUs just got a lot easier
Google released TorchTPU as open source, giving PyTorch developers a native way to run models on the company’s Tensor Processing Units. Google also announced TPU support in the vLLM and SGLang serving engines, the TorchTitan training framework, Ray, and Hugging Face’s libraries. Teams that write their models in PyTorch and run them on GPUs by default have the most to gain. Google says those workloads now move to TPUs, its custom AI chips, with minimal …
www.helpnetsecurity.com
October 21, 2026 at 10:18 PM
Google open-sourced TorchTPU, so PyTorch on TPUs just got a lot easier
Google released TorchTPU as open source, giving PyTorch developers a native way to run models on the company’s Tensor Processing Units. Google also announced TPU support in the vLLM and SGLang serv…
#hackernews #huggingface #llm
Google released TorchTPU as open source, giving PyTorch developers a native way to run models on the company’s Tensor Processing Units. Google also announced TPU support in the vLLM and SGLang serv…
#hackernews #huggingface #llm
Google & Meta launch TorchTPU for native PyTorch support on Google TPUs!
🔥True eager execution & torch.compile backend
⚡Fast inference via vLLM & SGLang
🏋️Near-linear scaling for Llama 3 405B & Qwen 3 MoE via TorchTitan
Read more here: https://bit.ly/4hV19Ho
🔥True eager execution & torch.compile backend
⚡Fast inference via vLLM & SGLang
🏋️Near-linear scaling for Llama 3 405B & Qwen 3 MoE via TorchTitan
Read more here: https://bit.ly/4hV19Ho
October 20, 2026 at 5:45 PM
Google & Meta launch TorchTPU for native PyTorch support on Google TPUs!
🔥True eager execution & torch.compile backend
⚡Fast inference via vLLM & SGLang
🏋️Near-linear scaling for Llama 3 405B & Qwen 3 MoE via TorchTitan
Read more here: https://bit.ly/4hV19Ho
🔥True eager execution & torch.compile backend
⚡Fast inference via vLLM & SGLang
🏋️Near-linear scaling for Llama 3 405B & Qwen 3 MoE via TorchTitan
Read more here: https://bit.ly/4hV19Ho
Zephon’s TorchTitan integration preserved identical global batches when a 1B-parameter model resumed from a 16-GPU checkpoint on 8, 32, and 64 GPUs.
Read more: https://pytorch.org/blog/bringing-elastically-deterministic-data-loading-to-torchtitan-with-zephon/
Read more: https://pytorch.org/blog/bringing-elastically-deterministic-data-loading-to-torchtitan-with-zephon/
October 16, 2026 at 4:39 PM
Zephon’s TorchTitan integration preserved identical global batches when a 1B-parameter model resumed from a 16-GPU checkpoint on 8, 32, and 64 GPUs.
Read more: https://pytorch.org/blog/bringing-elastically-deterministic-data-loading-to-torchtitan-with-zephon/
Read more: https://pytorch.org/blog/bringing-elastically-deterministic-data-loading-to-torchtitan-with-zephon/
Today, Meta is open sourcing the weights for Muse Spark Open, the most capable US open-weights model to date. It’s integrated with the PyTorch ecosystem: it ships with Helion kernels, and TorchTitan can be used to further tune the model.
More in the blog: https://pytorch.org/blog/muse-spark-open/
More in the blog: https://pytorch.org/blog/muse-spark-open/
October 15, 2026 at 5:57 PM
Today, Meta is open sourcing the weights for Muse Spark Open, the most capable US open-weights model to date. It’s integrated with the PyTorch ecosystem: it ships with Helion kernels, and TorchTitan can be used to further tune the model.
More in the blog: https://pytorch.org/blog/muse-spark-open/
More in the blog: https://pytorch.org/blog/muse-spark-open/
At #PyTorchCon North America 2026, Liz Li (AMD) and Shekhar Pandey (AMD) will discuss scalable MXFP8 pretraining on MI355X GPUs, including TorchAO kernel optimization and TorchTitan integration for an end-to-end upstream PyTorch training stack.
https://hubs.la/Q04v4SL60
https://hubs.la/Q04v4SL60
September 24, 2026 at 6:58 PM
At #PyTorchCon North America 2026, Liz Li (AMD) and Shekhar Pandey (AMD) will discuss scalable MXFP8 pretraining on MI355X GPUs, including TorchAO kernel optimization and TorchTitan integration for an end-to-end upstream PyTorch training stack.
https://hubs.la/Q04v4SL60
https://hubs.la/Q04v4SL60