Alternatives

Products that do what UNet diffusion model in pure CUDA does

Hi HN! I was inspired by Andrej Karpathy's llm.c (https://github.com/karpathy/llm.c), and wrote a full diffusion model training loop in CUDA. I learnt a lot about CUDA from Simon Boehm's Matmul blog (https://siboehm.com/articles/22/CUDA-MMM). Currently there is still a lot of room for optimization: the model is running at 45% speed of PyTorch with torch.compile. I'm curious about any thoughts or CUDA tips for convolutions.

1
Forge CLI▲107
Swarm agents optimize CUDA/Triton for any HF/PyTorch model
Jan 2026
2TV
Tiny-vLLM – high performance LLM inference engine in C++ and CUDA▲205
May 2026 · github.com
3
Seed Diffusion▲158
A faster, more holistic way to generate code
2025
4NG
NanoEuler – GPT-2 scale model in pure C/CUDA from scratch▲55
Hi everyone, I started working on nanoeuler after the ban of anthropic's fable because my ambition and dream is to work in the AI field in anthropic. The two interesting reasons that led me to create nanoeuler were (1) interfacing with llm does not mean understanding how they are composed and (2), working on llm with a very low-level layer to understand the correlation between parameters and data and growth of the model and how the GPU works and how some layers can be optimized. So I started working on it with a research aspect by making nanoeuler grow more and more but doing one step after…
Jun 2026 · github.com
5
Dream 7B▲191
Powerful Open Diffusion LLM, Beyond Autoregressive
2025
6
RightNow CLI▲121
Claude Code for CUDA, an open-source AI CLI for GPU devs
Oct 2025
7I4
Ideogram 4.0 – open-weight 9.3B text-to-image model▲46
It's our new text-to-image model: a 9.3B single-stream diffusion transformer trained entirely from scratch. We focused heavily on controllability through structured JSON prompts, with strong text rendering, spatial awareness through bounding box guidance, and color palette control. It has the best text rendering of any open-weight model we've tested so far, and the NF4 quantized checkpoint runs on a single 24GB GPU. For more technical details and examples see our blog post: https://ideogram.ai/blog/ideogram-4.0/ We will be happy to answer any questions :)
Jun 2026 · github.com
8ML
Minimal LLM Post-Training Experiments on an 8GB GPU (SFT, DPO, GRPO)▲21
Aug 2026 · github.com
9DG
Dockerized GPU Deep Learning Solution (Code and Blog and TensorFlow Demo)▲80
2016 · github.com
10LL
LeetGPU – LeetCode for GPU Programming▲9
After the incredible response to our launch of the first online CUDA playground, we have just shipped something we think all you GPU programming and ML enthusiasts will love. Introducing LeetGPU Challenges--the place to compete on writing the fastest CUDA kernels. We have problems like matrix multiplication, agent simulation, multi-head self-attention, with more dropping every couple of days! We have a lot of really cool things coming up, including support for PyTorch, TensorFlow, JAX, TinyGrad; Multi-GPU programs; H100, V100, A100 GPU options Give it a try and let us know what you think!
2025 · leetgpu.com
11NG
NanoEuler – GPT-2 scale model in pure C/CUDA from scratch▲6
Hi everyone, I started working on nanoeuler after the ban of anthropic's fable because my ambition and dream is to work in the AI field in anthropic. The two interesting reasons that led me to create nanoeuler were the first, interfacing with llm does not mean understanding how they are composed and two, working on llm with a very low-level layer to understand the correlation between parameters and data and growth of the model and how the GPU works and how some layers can be optimized. So I started working on it with a research aspect by making nanoeuler grow more and more but doing one step…
Jun 2026 · github.com
12LR
Linear RNN/Reservoir hybrid generative model, one C file (no deps.)▲7
I just noticed it takes literally ~5 minutes to train millions parameters on slow CPU...but before you call Yudkowsky that "it's over", an important note: the main bottleneck is the corpus size, params are just 'cleverness' but given limited info it's powerless. Anyway, here is the project: https://github.com/bggb7781-collab/lrnnsmdds/tree/main couple of notes: 1. single C file, no dependencies. Below are literally all the "dependencies", not even custom header (copy paste from the top of the single c file): #define _POSIX_C_SOURCE 200809L #include #include…
Apr 2026 · raw.githubusercontent.com
13IE
I embedded 685M public texts in 32 minutes (on 8x A100, Rust, TensorRT)▲7
Quick note on how it works and how I've done my batch embedding engine IgniteMS. The whole thing runs as one process using Rust, reading input, tokenizing, packing batches, keeping the queue full. TensorRT handles inference. Python is only as a wrapper. I built it this way because when you use more than couple of GPUs, the GPUs stop being the problem. CPU cannot feed them fast enough. One A100 can go through batches faster than Python can tokenize and feed, so the GPU just sits there idle waiting for work. Most of my time went into optimizing this. At 8 GPUs that was basically the entire…
Jun 2026 · github.com
14FO
Federation of robots collaboratively train an object manipulation model▲7
This is a quickstart example using LeRobot and Flower that demonstrates how to train a diffusion model collaboratively across 10 individual nodes (each with its own dataset). This example uses the push-t dataset, where the task is to move a letter T object on top of another that is to remain static. The example it's pretty easy to run, and can do so efficiently if you have access to a recent gaming GPU. Although the diffusion model only take 2GB of VRAM (of course you can decide to scale it up), the compute needed to train them isn't negligible. For context, running the example until…
2025 · github.com
15UA
Unstock.ai – Free image generation with SDXL▲9
Hi HN! I've got the barebones of a service running on top of Stable Diffusion XL. I can cheaply run image generations at 1024x1024. And of course there's a limit to how fast I can generate them given the request queue and limited GPUs, but the service is cheap enough that I'm happy to run it out of pocket for now. Let me know your thoughts, I hope you enjoy the service!
2023 · unstock.ai
16RA
RapidFire AI: 16–24x More Experiment Throughput Without Extra GPUs▲7
We built RapidFire AI, an open-source Python tool to speed up LLM fine-tuning and post-training with a powerful level of control not found in most tools: Stop, resume, clone-modify and warm-start configs on the fly—so you can branch experiments while they’re running instead of starting from scratch or running one after another. - Works within your OSS stack: PyTorch, HuggingFace TRL/PEFT), MLflow. - Hyperparallel search: launch as many configs as you want together, even on a single GPU - Dynamic real-time control: stop laggards, resume them later to revisit, branch promising configs in…
Sep 2025 · github.com
17SG
Self-growing neural networks via a custom Rust-to-LLVM compiler▲8
Hi HN, I built NOMA (Neural-Oriented Machine Architecture), a systems language where reverse-mode autodiff is a compiler pass (lowered to LLVM IR). My goal is to treat model parameters as explicit, growable memory buffers. Since NOMA compiles to standalone native binaries (no Python runtime), it allows using realloc on weights mid-training. This makes "self-growing" architectures a system primitive rather than a complex framework hack. I just pushed a reproducible benchmark (Self-Growing XOR) to validate the methodology: it compares NOMA against PyTorch and C++, specifically testing how…
Dec 2025 · github.com
18FM
Flow Matching model inference in C▲5
As the title and description of the GitHub repo suggest, I’m working on a small project for purely educational purposes, with the goal of implementing generative model inference (small models capable of modeling 2D distributions) based on the Flow Matching paradigm in C. I’ve worked on generative AI models based on Flow Matching from a more “abstract” perspective, using frameworks like PyTorch, and I wanted to understand what goes on behind the scenes. The repository is still a work in progress and is also one of my first "serious" projects in C.
Jul 2026 · github.com
19IC
I created a Neural Network from scratch, in scratch▲7
I spent the past week implementing a 1 Layer Neural Net and training it on MNIST within the visual scripting language provided by scratch.mit.edu. It was tedious, but ultimately not too difficult. The code runs incredibly slowly, so much so that 64 samples of MNIST takes 5+ hours to train on my machine. There were a lot of little mini challenges that were fun to overcome (implementing softmax was very tricky). If you're interested, I encourage you to try and improve on it! More details in the linked blog post.
2024 · bell-boy.github.io
20OS
Open-source fine-tuning in a Colab notebook▲5
Posted before, but wanted to share if you want an open source alternative to OpenAI fine-tuning, give Unsloth a try! Phi 3.5 was just released, and is distilled from GPT4. Unsloth makes finetuning 2x faster, uses 70% less VRAM + has no accuracy degradations. We rewrite all backprop steps and reduce FLOPs and write everything in Triton (JIT low level CUDA). If you want to own the weights after fine-tuning, give Unsloth a spin! I have free Colabs and Kaggle notebooks as well at https://github.com/unslothai/unsloth
2024 · colab.research.google.com
21BH
Beating Hinton et al.'s capsule net with fewer params and less training▲14
Hello HN, I recently posted a work-in-progress paper, along with code necessary for replicating all its results, at: https://github.com/glassroom/heinsen_routing Among other things, the code in this repo outperforms Hinton et al.'s recent state-of-the-art result in visual recognition[0] while requiring fewer parameters and an order-of-magnitude fewer training epochs. Most of the original research we do at work tends to be either proprietary in nature or tightly coupled to internal code, so we cannot share it with the world. In this case, however, I was able to remove all…
2019
22AF
A free, GPU-accelerated Texas Hold'em GTO solver in C++/CUDA▲5
I built this mostly because I love the intersection of game AI, high-performance computing, and poker. I’d love for anyone interested in game theory or CUDA optimization to tear it apart, test the accuracy, and give me feedback. Happy to answer any questions about the algorithms, the transition from CPU to GPU, or poker AI in general!
Jul 2026 · bupticybee.github.io
23RG
Running Gemma-4 26B at 124 tokens/SEC on a CPU, no GPU▲10
I wanted to know how fast a 26B mixture-of-experts model could run on a desktop CPU with no GPU. Got ~40 tok/s single-stream (lossless) and ~124 batched. The surprising part was the byte budget: for this model you compress the output head (32% of per-token bytes), not the experts (16%). The writeup has the bandwidth roofline and the dead-ends; the repo has the reproducible recipe. Happy to answer questions. Repo: https://github.com/arun-prasath2005/gemma4-cpu-moe
Jun 2026 · apeg.dev
24PF
Psychedelia fine tuned Stable Diffusion model▲5
Hi guys! I did a hackathon this weekend for leapml.dev and built out a site that lets you generate images with stable diffusion in the browser: https://aiart.fm Each image gets tagged and stored in a database so you can go back and look at what prompts, seeds, step, models people have used to make nice images. There is also a rudeimntary voting system and the homepage is sorted by highest rated. But the most exciting bit is I added my own custom fine tuned model to it last night that I'm calling "Psychedelia". All of the top images one the homepage are using this model. It seems to…
2023 · aiart.fm

Ranked by how close each launch is in meaning, then by votes. Refine with a description →