❤️

You can now run Black Forest Labs FLUX.1 schnell & dev as serverless endpoints on RunPod with ComfyUI via the 3.1.0 release of runpod-worker-comfy. ➡️ Quickstart github.com/blib-la/runpod…

Instructions for finetuning (FFT!) Mistral Large 123B on RunPod's MI300X compute, along with an axolotl config file to get you started. Should cost you about ~$500 for ~60M tokens. huggingface.co/anthracite-org…

For those unaware, MI300x AMD GPUs are the most cost effective way to fully finetune a large (~123b) model (for a group of individuals, at least), by far. We've published the shell setup to get AMD training working, as well as the axolotl config used (same for the datasets).

Why Amazon Web Services does not have 80 GB VRAM GPU? there is a huge gap either you get 24 GB or 320 G. Well, let me give it a shot to RunPod

dstack AMD RunPod While testing, we were blown away by the MI300X’s memory capacity—it’s perfect for LLMs! Also, check out the model playground UI, which you automatically get when you run a model via dstack. Tested it with Llama 3.1 70B running on a single node at full precision! 🤯

Wow, dstack is on Hacker News! Give us a shout-out! 🚀

RunFlux - a Google Colab Notebook to spin up a FLUX1-dev LoRA training at RunPod - Upload images+captions - Choose a GPU - Click Run, lean back and watch LoRAs and sample images pushed to your Hugging Face repo Colab Notebook: colab.research.google.com/github/geronim…

A typical train-on-my-face LoRA run is ~1 USD For example, training on 14 images for 3000 steps (rank 16, batch size 1) took - 4hrs at 22c/hr on a RTX 3090 - 2hrs at 44c/hr on a RTX 4090

Thanks, Philadelphia Business Journal, for highlighting RunPod's inclusion on Forbes' Next Billion-Dollar Startups list. Keep reading to see how the company's globally distributed GPU cloud platform for developing and deploying AI enhances developers' daily lives shorturl.at/QeuXg

It's great to see RunPod, Pecan AI, MinIO, and Domino Data Lab shortlisted in the inaugural 2024 A.I. Awards! Check out the full list of honorees broken down by categories here: cloud-awards.com/2024-ai-awards…

🦋 CogVideoX-5B: Text-to-Video Diffusion Models with An Expert Transformer 📽 Jupyter Notebook 🥳 + 🥪 🥪 Tost AI + 🍇 RunPod serverless Thanks to Zhuoyi Yang ❤ Jiayan Teng ❤ Wendi Zheng ❤ Ming Ding ❤ Shiyu Huang ❤ Jiazheng Xu ❤ Yuanming Yang ❤ Wenyi Hong ❤ Xiaohan

🌟A Huge Thank You to Our Incredible Sponsors!🌟 We’re grateful to have the support of these amazing companies at The AI Conference 2024. Their contributions are driving the future of AI innovation and making this event possible. From tech giants to innovative startups, our

Self-Hosting LLaMA 3.1 70 on RunPod with vLLM Inference Engine For a 70B parameter model, to deploy in 16-bit floating point precision we’ll need ~140GB of memory, and for something like 4-bit (INT4) we only need ~45GB And you need additional memory for - Context

🔥 Advanced LivePortrait ComfyUI 🍇 RunPod template: runpod.io/console/deploy…

turns out Gamma wasn't the right fit for my slides because it couldn’t generate a theme and update all slides at once, so I ended up using v0 again ¯\_(ツ)_/¯ here's what I did: 1. created 4 different slides in v0 2. manually recreated them in Google slides i