Frankenstein Composer

build a custom LLM by stitching two models, then chat with the result. SLERP paper E1 negative GitHub
What this does: drag two 8B language models onto the skeleton, blend their weights with a recipe, and chat with the resulting hybrid. Pre-built defaults activate instantly. Custom blend params trigger a live merge (~3 min, ~16 GB).
quick start →

Skeleton

Drag a base model into Left and another into Right for every row. Recipe + params at the bottom control how the two blend.
Left organmodel A
Spinelayer block
Right organmodel B

What you're building

Skeleton is empty. Drag a base model from the palette, or click a quick-start chip above.

Blend params

Compute pool
loading...
how to contribute compute (5 min, one curl command)
No model active. Build a skeleton, then click Combine.
Once a model is active, type a prompt below. CPU inference: 2-5 min per 100 tokens.
try: math (17+25) reasoning chain poem about ice