PyTorch
How vibe coders and indie builders use PyTorch: 17 projects shared since May 2026 were built with it, and 3 were built for it.
How to read this
We count the projects builders share on Reddit each week. A project is counted once, in the week it first appears in our sources — that’s not necessarily when it was built or launched. Because some weeks are busier than others, we compare each tool's share of the projects that say what they're built with, not raw counts, so a busy week doesn't make everything look like it's growing. “Up 70%” means a tool turns up in 70% more of those projects than before. In the Oct 5 – 11 edition, the last 4 weeks are Sep 14 – Oct 11, compared with the previous 8 weeks, Jul 20 – Sep 13; the last 12 weeks are Jul 20 – Oct 11.
Projects built with it
17
Share of stated stacks, last 4 weeks
0.1%
Adjusted trend, last 4 weeks vs previous 8 weeks
too few to tell
Add-ons built for it
3
Share of stated stacks, each week
Share rather than counts, so changes in how many posts a subreddit lets through don't show up as trends.
Often used with
What's built with it
Latest projects built with PyTorch
- IronBee Gamer — Plays browser games by reading the rendered page, using LLM-written rules and a small local model to choose each move.“We built a bot that learns browser games from the page alone: an LLM writes the rules, a small local model makes each move in ~25 ms”
- Lyrebird — Clones a voice from a short audio sample and renders typed scripts, including multi-voice dialogue, into speech audio offline.“Clone Your Voice in 10 Seconds, Free & Offline (Lyrebird)”
- Accord Guard — Detects sensitive identifiers in text locally in the browser and replaces them before they are submitted to generative AI tools.“Built a free, locally running privacy extension for ChatGPT. Looking for feedback from other extension developers.”
- JAZZ — Provides a customized Arch Linux desktop environment with a Quickshell UI and local AI engineering tooling.“Jazz: Developed Arch Linux Desktop.”
- I made a tiny Diffusion model,… — Trains a small diffusion model that generates images by denoising random noise, built from a custom noise scheduler, forward diffusion, and UNet.“I made a tiny Diffusion model, here's what I learned”
- Mission Control — Combines hardware telemetry, game configuration, local AI assistance and an in-game HUD into one Windows gaming dashboard.“We built a gaming-focused AI dashboard/overlay/assistant — looking for brutally honest engineering feedback”
- DubForge & DubStage — Cuts a video scene into timed audio lines and lets users record their own voice over each line, comparing their take against the original waveform.“DubForge & DubStage - dub a video scene with your own voice, GPL-3.0, fully offline”
- ml-quant-trading — Runs an end-to-end quantitative trading research pipeline with PyTorch factor models, portfolio optimization, and cost-aware backtesting.“I open-sourced a PyTorch multi-factor trading research stack with 213 factors”
- GENESIS-165M — Trains and generates text with a from-scratch GPT-style Transformer language model on WikiText-103.“Building a 165M parameter GPT-style language model from scratch as a student”
- NeuroBranch — Lets users build neural networks by wiring typed visual cards on a canvas and compiling the graph into runnable PyTorch code.“I got tired of neural net code being unreadable, so I built a visual editor that compiles drag-and-drop graphs into real, runnable PyTorch”
- PDF Redactor — Permanently removes text from PDFs, including hidden layers, metadata and embedded data, instead of drawing black boxes over it.“Day 6 of vibecoding a PDF Redactor -- It's become a hole puncher thru reality”
- First time building a vision… — Trains a vision model to track a moving ball with physics in a 2D simulated world.“First time building a vision based AI model (Claude Code assisted).”
- librarian-press — Runs an end-to-end pipeline that trains a tokenizer, pretrains and fine-tunes small GPT-style language models, and evaluates and exports them.“Built an open source LLM training framework from scratch — 1B param model trains for under 100 bucks”
- AutoVidify — Generates subtitles for videos by transcribing speech, cleaning up and translating the text with an LLM, separating vocals, and rendering the subtitles onto video.“I spent 9 months building a full AI SaaS, got zero users, and open-sourced it. What would you have done in my place?”
- Inferix — Runs user-supplied model Docker images as serverless inference endpoints on AMD MI300X GPUs, scaling to zero when idle and billing per second.“I built a POC for serverless inference platform on AMD GPUs — 5-min demo, would love feedback before opening up”
Latest add-ons built for PyTorch
- tensorViz — Renders PyTorch nn.Module code as an expandable graph of layers with captured tensor shapes, linked back to source.“I made a visual map for PyTorch models that opens beside the code”
- GPUOpt — Benchmarks PyTorch/CUDA inference workloads on NVIDIA GPUs and tests optimization candidates, keeping only changes that measurably improve performance.“I’m building GPUOpt — a measurement-first GPU optimization tool for PyTorch/CUDA inference”
- Aether-Vault — Versions machine learning code, model weights and datasets together in a Git-like self-hosted system with deduplicated storage.“Aether-Vault: Versioning for ml and agentic model training”
Pitched as alternatives to PyTorch
- vpipe — Runs local AI models such as LLMs, image and video generation, and speech recognition on Apple Silicon using a C++/Metal runtime instead of PyTorch or MLX.“Lightweight C++/Metal runtime for local AI on Apple Silicon”
- Godwit — Runs large mixture-of-experts language models on small Apple Silicon Macs by streaming expert weights from SSD as needed.“I built an engine in Swift that runs a 120-billion-parameter AI model on a 16GB MacBook Air”
- Shorokoo — Defines, trains, and runs neural networks written in C# instead of Python.“I made a C# alternative for PyTorch/TensorFlow”