martyn / safetensors-merge-supermario

Merge safetensor files using the technique described in "Language Models are Super Mario: Absorbing Abilities from Homologous Models as a Free Lunch"

Hi there 👋

Welcome to my profile, where I code, create, and collaborate. I'm a full stack engineer, a technology optimist, and an AI enthusiast.

About me: 🙋

• 🚀 I help startups and businesses launch their ideas and scale their products.
• 📚 I write articles to share how to use technology and AI for good.
• 💸 I like cryptocurrency and it's potential.

Open source models 🤖

I publish open source models to:

• 🖌️ Civit AI
• 🤗 HuggingFace

Open source 🔧

• 🔀 Safetensors Merge SuperMario DARE - A tool for merging LoRas, diffusion models, language models, and more using the DARE technique.
• 🎮 DeepFloyd IF server - a simple server for DeepFloyd with a runpod.io compatible Dockerfile.
• 🧮 GPT-echo - experiments using echo state networks with pretrained language models.
• ✨ NEAR monsters - NFT trading card game. NEAR boshacks hackathon entry winning 2nd place. Features custom rust smart contracts.
• 🎨 HyperGAN - a GAN framework in pytorch(previously tensorflow 1.0).

Follow me on other sites: 🎉

• 💼 LinkedIn - I write some articles here.
• 📝 Medium - I write some articles here too.
• 💬 Twitter - Mostly for sharing thoughts.

safetensors-merge-supermario's Issues

SD Turbo (SD 2.1) vs SDXL Turbo merging?

Is it possible to merge SD 2.1 models with SD Turbo model? Streamdiffusion only supports SD 2.1 hence I wanted to know if it's possible, thanks!

SD Next compatibility ?

Hi there :)
Is compatibility with other A1111 forks, such as the Vladmandic (Sd Next) one, being considered ?
Cheers !

Inference step

Hello,
thanks for sharing!
I wonder how you get the merged image? Inference for only One step or 50steps? It will be better to provide the prompt and seed!
Thanks