# projects **URL:** https://discuss.pytorch.org/c/projects/33.md [Latest](https://discuss.pytorch.org/latest.md) · [Categories](https://discuss.pytorch.org/categories.md) --- ## [About the projects category](https://discuss.pytorch.org/t/about-the-projects-category/117648)
**Author:** [@suraj.pt](https://discuss.pytorch.org/u/suraj.pt)\ **Replies:** 0
Tell the community how you’re using PyTorch! --- ## [Released gabp-sparse-inv 0.4.1 - differentiable selected inversion in PyTorch](https://discuss.pytorch.org/t/released-gabp-sparse-inv-0-4-1-differentiable-selected-inversion-in-pytorch/225573)
**Author:** [@Vanshdeep\_Sehrawat](https://discuss.pytorch.org/u/Vanshdeep_Sehrawat)\ **Replies:** 0\ **Last updated:** [October 19, 2026, 7:40am UTC](https://discuss.pytorch.org/t/released-gabp-sparse-inv-0-4-1-differentiable-selected-inversion-in-pytorch/225573 "2026-10-19T07:40:04Z")
I’ve released an open-source Python package for computing selected inverse blocks of sparse block-structured matrices without forming the dense inverse. It supports PyTorch autograd, sparse solves, log-determinants, Gau… --- ## [Kpnn2: sparsely connected PyTorch layers from a named graph, with attributions by node name](https://discuss.pytorch.org/t/kpnn2-sparsely-connected-pytorch-layers-from-a-named-graph-with-attributions-by-node-name/225558)
**Author:** [@Thomas\_Rauter](https://discuss.pytorch.org/u/Thomas_Rauter)\ **Replies:** 0\ **Last updated:** [October 16, 2026, 11:28am UTC](https://discuss.pytorch.org/t/kpnn2-sparsely-connected-pytorch-layers-from-a-named-graph-with-attributions-by-node-name/225558 "2026-10-16T11:28:51Z")
kpnn2 turns a graph of named entities into sparsely connected PyTorch layers. Each node becomes a unit, each edge becomes a weight, pairs without an edge stay unconnected, and attribution scores come back labeled by node… --- ## [TorchYAND: A Matrix-Free PyTorch Implementation for Higher-Moment Portfolio Optimization](https://discuss.pytorch.org/t/torchyand-a-matrix-free-pytorch-implementation-for-higher-moment-portfolio-optimization/225549)
**Author:** [@JohnnyZPiggy](https://discuss.pytorch.org/u/JohnnyZPiggy)\ **Replies:** 0\ **Last updated:** [October 15, 2026, 2:34pm UTC](https://discuss.pytorch.org/t/torchyand-a-matrix-free-pytorch-implementation-for-higher-moment-portfolio-optimization/225549 "2026-10-15T14:34:09Z")
Hi everyone, I’ve been experimenting with implementing optimization algorithms in PyTorch as a side project. TorchYAND is a PyTorch implementation of the YAND-MVSK algorithm for higher-moment portfolio optimization. S… --- ## [CrashLens: local GPU log triage for OOM, NCCL and Xid errors](https://discuss.pytorch.org/t/crashlens-local-gpu-log-triage-for-oom-nccl-and-xid-errors/225493)
**Author:** [@Vivaan](https://discuss.pytorch.org/u/Vivaan)\ **Replies:** 2\ **Last updated:** [October 10, 2026, 12:41pm UTC](https://discuss.pytorch.org/t/crashlens-local-gpu-log-triage-for-oom-nccl-and-xid-errors/225493 "2026-10-10T12:41:01Z")
Hi everyone — I’m Vivaan, building ThermaCompute AI to help teams understand GPU failures and wasted compute. Our current work is passive diagnostics: turning logs and telemetry into evidence engineers can investigate, w… --- ## [Rgpu: a PyTorch device whose tensors live on a remote GPU](https://discuss.pytorch.org/t/rgpu-a-pytorch-device-whose-tensors-live-on-a-remote-gpu/225481)
**Author:** [@yanmi](https://discuss.pytorch.org/u/yanmi)\ **Replies:** 1\ **Last updated:** [October 3, 2026, 6:11am UTC](https://discuss.pytorch.org/t/rgpu-a-pytorch-device-whose-tensors-live-on-a-remote-gpu/225481 "2026-10-03T06:11:26Z")
Hi all. I’ve been building rgpu, which lets you write device="rgpu" on a laptop while the tensors and kernels live on a GPU host you reach over SSH. pip install rgpu import torch, rgpu x = torch.ones(4, device="rgpu") … --- ## [TraceML: always-on runtime visibility for PyTorch training](https://discuss.pytorch.org/t/traceml-always-on-runtime-visibility-for-pytorch-training/224679)
**Author:** [@abhinavsriva](https://discuss.pytorch.org/u/abhinavsriva)\ **Replies:** 1\ **Last updated:** [October 1, 2026, 9:37am UTC](https://discuss.pytorch.org/t/traceml-always-on-runtime-visibility-for-pytorch-training/224679 "2026-10-01T09:37:00Z")
Hi everyone, I am building TraceML, an open-source tool for always-on step-level training visibility in PyTorch. It shows where time and memory go inside each training step while training is still running, making it ea… --- ## [ShrinkAI: knowledge distillation and compression package](https://discuss.pytorch.org/t/shrinkai-knowledge-distillation-and-compression-package/225411)
**Author:** [@elouanzer](https://discuss.pytorch.org/u/elouanzer)\ **Replies:** 0\ **Last updated:** [September 14, 2026, 9:54am UTC](https://discuss.pytorch.org/t/shrinkai-knowledge-distillation-and-compression-package/225411 "2026-09-14T09:54:28Z")
I just released shrinkai, a python package for reducing neural network size and inference latency (distillation + pruning + quantization), that could be useful for edge/mobile deployment. Built to feel native if you alre… --- ## [Clifra: a differentiable Clifford algebra computation layer for PyTorch](https://discuss.pytorch.org/t/clifra-a-differentiable-clifford-algebra-computation-layer-for-pytorch/225410)
**Author:** [@Concode](https://discuss.pytorch.org/u/Concode)\ **Replies:** 0\ **Last updated:** [September 14, 2026, 1:58am UTC](https://discuss.pytorch.org/t/clifra-a-differentiable-clifford-algebra-computation-layer-for-pytorch/225410 "2026-09-14T01:58:49Z")
Hi, I’ve been working on clifra, a differentiable Clifford algebra computation layer for PyTorch. The main design goal is to keep Clifford algebra inside the normal PyTorch programming model. Values are ordinary torch.… --- ## [What changes when a PyTorch prototype has to run reliably in production?](https://discuss.pytorch.org/t/what-changes-when-a-pytorch-prototype-has-to-run-reliably-in-production/225332)
**Author:** [@wyattlewiiss](https://discuss.pytorch.org/u/wyattlewiiss)\ **Replies:** 0\ **Last updated:** [August 25, 2026, 7:57pm UTC](https://discuss.pytorch.org/t/what-changes-when-a-pytorch-prototype-has-to-run-reliably-in-production/225332 "2026-08-25T19:57:31Z")
I’ve been thinking about the gap between getting a PyTorch model working in a notebook and actually operating it as part of a production application. In projects I’ve worked around, the model itself is often only one pa… --- ## [Built a 380M param LLM from scratch (no PyTorch/TF/JAX) and found a nasty residual connection bug via a custom debugger (that works across all major ml frameworks)](https://discuss.pytorch.org/t/built-a-380m-param-llm-from-scratch-no-pytorch-tf-jax-and-found-a-nasty-residual-connection-bug-via-a-custom-debugger-that-works-across-all-major-ml-frameworks/225284)
**Author:** [@codeyash09](https://discuss.pytorch.org/u/codeyash09)\ **Replies:** 0\ **Last updated:** [August 14, 2026, 5:29am UTC](https://discuss.pytorch.org/t/built-a-380m-param-llm-from-scratch-no-pytorch-tf-jax-and-found-a-nasty-residual-connection-bug-via-a-custom-debugger-that-works-across-all-major-ml-frameworks/225284 "2026-08-14T05:29:46Z")
Since December, I’ve been building a 380M parameter LLM entirely from scratch, using only NumPy and CuPy (no PyTorch, TensorFlow, or JAX). What started as an exercise to deeply understand transformers by implementing “At… --- ## [Orchestration layer](https://discuss.pytorch.org/t/orchestration-layer/225275)
**Author:** [@Jam2](https://discuss.pytorch.org/u/Jam2)\ **Replies:** 0\ **Last updated:** [August 11, 2026, 3:46pm UTC](https://discuss.pytorch.org/t/orchestration-layer/225275 "2026-08-11T15:46:06Z")
Hi friends, is there an orchestration layer that can sit around PyTorch and handle process like experiment tracking, evaluation, and optimization without requiring the existing training code to be rewritten? --- ## [sal-torch : training-time head masking for compression-resilient transformers](https://discuss.pytorch.org/t/sal-torch-training-time-head-masking-for-compression-resilient-transformers/225129)
**Author:** [@zeekmartin](https://discuss.pytorch.org/u/zeekmartin)\ **Replies:** 0\ **Last updated:** [June 26, 2026, 12:20pm UTC](https://discuss.pytorch.org/t/sal-torch-training-time-head-masking-for-compression-resilient-transformers/225129 "2026-06-26T12:20:30Z")
Hi everyone, I’ve been working on a different approach to model compression and wanted to share the resulting package. The idea: instead of pruning a trained model and hoping it holds up, what if we trained the model t… --- ## [TorchJD: Computing and aggregating Jacobians with PyTorch](https://discuss.pytorch.org/t/torchjd-computing-and-aggregating-jacobians-with-pytorch/224655)
**Author:** [@valerian.rey](https://discuss.pytorch.org/u/valerian.rey)\ **Replies:** 1\ **Last updated:** [June 7, 2026, 11:35am UTC](https://discuss.pytorch.org/t/torchjd-computing-and-aggregating-jacobians-with-pytorch/224655 "2026-06-07T11:35:17Z")
TorchJD Hi everyone! I’d like to quickly present my biggest project so far, in collaboration with @Pierre\_Quinton: TorchJD. When you train a model with multiple losses, TorchJD allows you to compute the Jacobian of the v… --- ## [Encoding arbitrarily nested, structured data with trees of MHSA + pooling blocks](https://discuss.pytorch.org/t/encoding-arbitrarily-nested-structured-data-with-trees-of-mhsa-pooling-blocks/225023)
**Author:** [@granthamtaylor](https://discuss.pytorch.org/u/granthamtaylor)\ **Replies:** 0\ **Last updated:** [June 1, 2026, 5:54pm UTC](https://discuss.pytorch.org/t/encoding-arbitrarily-nested-structured-data-with-trees-of-mhsa-pooling-blocks/225023 "2026-06-01T17:54:32Z")
I have spent the last several years to wrangle TensorDict and @tensorclass into an extensible framework able to encode arbitrarily nested (structured) json-like objects for predictive modeling tasks. Basically, a tree o… --- ## [How to cast/convert a JAX function into a Pytorch autograd differentiable function](https://discuss.pytorch.org/t/how-to-cast-convert-a-jax-function-into-a-pytorch-autograd-differentiable-function/225004)
**Author:** [@JosueCom](https://discuss.pytorch.org/u/JosueCom)\ **Replies:** 0\ **Last updated:** [May 26, 2026, 7:27am UTC](https://discuss.pytorch.org/t/how-to-cast-convert-a-jax-function-into-a-pytorch-autograd-differentiable-function/225004 "2026-05-26T07:27:42Z")
A way to convert a pure JAX function into an autograd differentiable PyTorch function that allows the function to be used in backward() and autograd.grad() calls while supporting higher orders of differentiability. impo… --- ## [Torchdiag — lightweight model health diagnostics for PyTorch](https://discuss.pytorch.org/t/torchdiag-lightweight-model-health-diagnostics-for-pytorch/224999)
**Author:** [@Aditya\_Mehra](https://discuss.pytorch.org/u/Aditya_Mehra)\ **Replies:** 0\ **Last updated:** [May 24, 2026, 2:43am UTC](https://discuss.pytorch.org/t/torchdiag-lightweight-model-health-diagnostics-for-pytorch/224999 "2026-05-24T02:43:56Z")
I just open-sourced torchdiag, a small diagnostic toolkit for PyTorch models. It brings SRE-style observability to model training with five commands: torchdiag.summary(model) — parameter counts, memory footprint, devic… --- ## [TorchRL DQN with Flame](https://discuss.pytorch.org/t/torchrl-dqn-with-flame/224941)
**Author:** [@k82cn](https://discuss.pytorch.org/u/k82cn)\ **Replies:** 0\ **Last updated:** [May 14, 2026, 7:03pm UTC](https://discuss.pytorch.org/t/torchrl-dqn-with-flame/224941 "2026-05-14T19:03:37Z")
Hi team, I wrote an example of TrochRL DQN with Flame which is an distributed engine for AI application. And that’ll be great if any suggestions on more complex examples/applications for the distribution, feel free to l… --- ## [I created a minimal one-file implementations (160loc) of JEPA family (ijepa, vjepa, vjepa2, cjepa) for educational purposes \[P\]](https://discuss.pytorch.org/t/i-created-a-minimal-one-file-implementations-160loc-of-jepa-family-ijepa-vjepa-vjepa2-cjepa-for-educational-purposes-p/224932)
**Author:** [@keonkim](https://discuss.pytorch.org/u/keonkim)\ **Replies:** 0\ **Last updated:** [May 13, 2026, 10:22am UTC](https://discuss.pytorch.org/t/i-created-a-minimal-one-file-implementations-160loc-of-jepa-family-ijepa-vjepa-vjepa2-cjepa-for-educational-purposes-p/224932 "2026-05-13T10:22:25Z")
Hi all, I made my own minimal implementation of JEPA algorithms. Making things minimal and removing all the things needed for scaling the algorithm always helped me understanding. So I stripped everything but the algor… --- ## [GYRO: A PyTorch optimizer with geometric gradient projection to dampen oscillations](https://discuss.pytorch.org/t/gyro-a-pytorch-optimizer-with-geometric-gradient-projection-to-dampen-oscillations/224889)
**Author:** [@Sunder\_Flow\_Research](https://discuss.pytorch.org/u/Sunder_Flow_Research)\ **Replies:** 2\ **Last updated:** [May 4, 2026, 9:27am UTC](https://discuss.pytorch.org/t/gyro-a-pytorch-optimizer-with-geometric-gradient-projection-to-dampen-oscillations/224889 "2026-05-04T09:27:24Z")
Hey, I’ve been working on a small side project and wanted to share it and get some thoughts from people who know this space better than I do :hugs: GYRO (Geometric Yield Rotation Optimizer) is a PyTorch optimizer that w… --- ## [Implementation of recent meta ai paper](https://discuss.pytorch.org/t/implementation-of-recent-meta-ai-paper/224893)
**Author:** [@reinforced](https://discuss.pytorch.org/u/reinforced)\ **Replies:** 1\ **Last updated:** [May 3, 2026, 4:58pm UTC](https://discuss.pytorch.org/t/implementation-of-recent-meta-ai-paper/224893 "2026-05-03T16:58:48Z")
paper arxiv link : \[2604.16529v1\] Scaling Test-Time Compute for Agentic Coding github link : genji970/facebook-paper\_harness-inference-scale-agent: unofficial implementation of Meta(facebook) Ai paper “Scaling test time… --- ## [MODTORCH: a meta-language to build PyTorch dynamically](https://discuss.pytorch.org/t/modtorch-a-meta-language-to-build-pytorch-dynamically/224848)
**Author:** [@riccstef](https://discuss.pytorch.org/u/riccstef)\ **Replies:** 0\ **Last updated:** [April 23, 2026, 4:19pm UTC](https://discuss.pytorch.org/t/modtorch-a-meta-language-to-build-pytorch-dynamically/224848 "2026-04-23T16:19:50Z")
Hello, I developed MODTORCH a meta-language for building PyTorch networks on the fly, without having to write/change PyTorch classes manually every time. It makes me really easier to test different architectures. Maybe … --- ## [NOTORCH: because we're all tired of overrated, normalized bloat](https://discuss.pytorch.org/t/notorch-because-were-all-tired-of-overrated-normalized-bloat/224774)
**Author:** [@ariannamethod](https://discuss.pytorch.org/u/ariannamethod)\ **Replies:** 0\ **Last updated:** [April 6, 2026, 4:10pm UTC](https://discuss.pytorch.org/t/notorch-because-were-all-tired-of-overrated-normalized-bloat/224774 "2026-04-06T16:10:42Z")
NOTORCH — the entire stack rewritten in C. no Python. no pip. no mercy. NO PyTorch dependency hell. NO Python. NO pip. NO conda. No RAM — RAM, not vRAM — abuse disguised as “innovation”. No “eGosystem”. ENOUGH. Ge… --- ## [Supporting Data Lineage](https://discuss.pytorch.org/t/supporting-data-lineage/224759)
**Author:** [@clodaghwalsh17](https://discuss.pytorch.org/u/clodaghwalsh17)\ **Replies:** 0\ **Last updated:** [April 3, 2026, 8:19am UTC](https://discuss.pytorch.org/t/supporting-data-lineage/224759 "2026-04-03T08:19:36Z")
Myself and my collegue are presenting our thoughts on supporting data lineage at Pytorch EU next week and wanted to start a broader discussion on the topic in the community. We will present our ideas on how PyTorch could… --- ## [MCCL: First Native MPS distributed backend for Pytorch](https://discuss.pytorch.org/t/mccl-first-native-mps-distributed-backend-for-pytorch/224729)
**Author:** [@copythatpasta](https://discuss.pytorch.org/u/copythatpasta)\ **Replies:** 0\ **Last updated:** [March 21, 2026, 6:27pm UTC](https://discuss.pytorch.org/t/mccl-first-native-mps-distributed-backend-for-pytorch/224729 "2026-03-21T18:27:02Z")
Created a DDP backend for apple silicon devices, tested it working across a m1max + m4max MBP. There’s a demo video in the repo showing it working: https://github.com/mps-ddp/mccl I’m definitely looking for additional … --- ## [GSoC 2026 Interest: Building on my previous contributions (PR #175990)](https://discuss.pytorch.org/t/gsoc-2026-interest-building-on-my-previous-contributions-pr-175990/224673)
**Author:** [@mhusnain-tech](https://discuss.pytorch.org/u/mhusnain-tech)\ **Replies:** 0\ **Last updated:** [March 8, 2026, 7:09pm UTC](https://discuss.pytorch.org/t/gsoc-2026-interest-building-on-my-previous-contributions-pr-175990/224673 "2026-03-08T19:09:21Z")
Hi PyTorch Team, I am M Husnain, a BSCS student at MNS UET Multan with a 3.8 CGPA. I have a deep-rooted interest in Neural Networks and have completed several specializations including Deep Learning, TensorFlow Develope… --- ## [Dev Discuss is Read Only](https://discuss.pytorch.org/t/dev-discuss-is-read-only/224656)
**Author:** [@Morrison\_Turnansky](https://discuss.pytorch.org/u/Morrison_Turnansky)\ **Replies:** 1\ **Last updated:** [March 6, 2026, 6:23pm UTC](https://discuss.pytorch.org/t/dev-discuss-is-read-only/224656 "2026-03-06T18:23:45Z")
Notifying that dev-discuss is currently read only. Message is “This site is in read-only mode because there are overdue hosting payments.” --- ## [MacBook Air M4 Chip PyTorch Compatability](https://discuss.pytorch.org/t/macbook-air-m4-chip-pytorch-compatability/223131)
**Author:** [@PyTorch24107](https://discuss.pytorch.org/u/PyTorch24107)\ **Replies:** 1\ **Last updated:** [January 21, 2026, 2:43pm UTC](https://discuss.pytorch.org/t/macbook-air-m4-chip-pytorch-compatability/223131 "2026-01-21T14:43:33Z")
Hello, I wanted to check whether PyTorch runs on Apple MacBook Air M4 Chip 2025, with 10 core CPU, 10 core GPU, 16 core neural engine. Thank you. --- ## [ResDAG: A modern, GPU-accelerated reservoir computing library for PyTorch](https://discuss.pytorch.org/t/resdag-a-modern-gpu-accelerated-reservoir-computing-library-for-pytorch/224374)
**Author:** [@El3ssar](https://discuss.pytorch.org/u/El3ssar)\ **Replies:** 0\ **Last updated:** [January 18, 2026, 4:36pm UTC](https://discuss.pytorch.org/t/resdag-a-modern-gpu-accelerated-reservoir-computing-library-for-pytorch/224374 "2026-01-18T16:36:04Z")
I’ve been working on reservoir computing for my PhD research and found myself repeatedly building ESN infrastructure from scratch. So I packaged it into a library: ResDAG. It’s a pure PyTorch implementation of Echo Stat… --- ## [Pytorch Symbolic: an equivalent of Keras Functional API](https://discuss.pytorch.org/t/pytorch-symbolic-an-equivalent-of-keras-functional-api/165395)
**Author:** [@gahaalt](https://discuss.pytorch.org/u/gahaalt)\ **Replies:** 3\ **Last updated:** [January 18, 2026, 4:24pm UTC](https://discuss.pytorch.org/t/pytorch-symbolic-an-equivalent-of-keras-functional-api/165395 "2026-01-18T16:24:37Z")
Hello! I just hit 1.0.0 version of a library I’ve been developing for the past months as a side project. Pytorch Symbolic A library that aims to provide a concise API for neural network creation in PyTorch. The API and… [Next page](https://discuss.pytorch.org/c/projects/33.md?page=1)