Zero-valued PyTorch loss can run code in the backward pass
IEEE 754 rounding and autodiff disagree, so a loss logged as 0.0 can still set gradients that rewrite weights.
52
Product
PyTorch
- Mentions · 7d8
- Mentions · 30d10
- All mentions10
- First seenSep 26, 2026
- Last mentioned4d ago
Timeline
10 articles · newest firstGoogle open-sources TorchTPU for running PyTorch on TPUs
TPU support also landed in vLLM, SGLang, TorchTitan, Ray, and Hugging Face libraries, with two pieces still in preview.
58
Google open-sources TorchTPU so PyTorch apps can run on TPUs
Amazon released TorchNeuron a day earlier to port PyTorch apps to Trainium without rewriting them.
55
Google open-sources TorchTPU, a native PyTorch back end for TPUs
The feed only shows a Hacker News thread with 42 points and 12 comments.
42
Google open-sources TorchTPU, a PyTorch backend for Cloud TPUs
The supplied page text is a cookie notice and does not describe features, license, or availability.
42
Hugging Face blog covers MPS in PyTorch 2.15
The title names Metal Performance Shaders support in PyTorch 2.15; the post body was not available.
18
AWS releases TorchNeuron so PyTorch apps can run on Trainium
Amazon says one configuration line moves PyTorch apps to Trainium and it will upstream changes to loosen Nvidia ties in PyTorch.
52
doubleAI unveils Maxwell, a runtime model that rewrites running binaries
doubleAI says Maxwell proves each rewrite equivalent to the original and claims large speedups on GPUs and CPUs.
48
AI Models Can Provably Hide Arbitrary Capabilities
A paper shows neural nets can cryptographically hide arbitrary capabilities that elicitation audits cannot reliably surface.
61
End-to-End Multimodal Data Augmentation and Adversarial Robustness Benchmark with AugLy for Images, Text, Audio, and PyTorch
A tutorial shows AugLy augmentations and small robustness tests for image, text, and audio in PyTorch.
32
Appears with
- TorchTPUproduct · 4
- Googlevendor · 4
- Nvidiavendor · 3
- Hugging Faceorganization · 3
- Metaorganization · 2
- Metavendor · 2
- Trainiumproduct · 2
- Linux Foundationorganization · 2
- TorchNeuronproduct · 2
- vLLMproduct · 1
- AugLyproduct · 1
- TorchTitanproduct · 1
- IBMorganization · 1
- Red Hatorganization · 1
- Amazon Web Servicesvendor · 1
- Tensor Processing Unitsproduct · 1
Entities are extracted by the model from each article. First seen , last mentioned . Watching keeps it in this browser only (no account); the watchlist page and dashboard alerts use it.