Unedited
Qwen3.5-0.8B as released, the labels in its prompt
Each prompt, labelled.
hedwigaihedwigaihedwigai
Train a small model on your examples, no gradients
Give user prompts and the label or answer each should get. Qwen3.5-0.8B gets new neurons, fitted by least squares from forward passes, with no gradients. Then type any prompt. The post explains how it works.
Continue in
1. Pick a task
2. Train it
User promptLabel
Settings evaluated
Train it, and each setting tried appears here.
Loading the recorded tasks…
3. Ask it something
Sign in to run
Typing your own prompts is free once you are signed in.
4. Compare
Trained
The same model with new neurons, no gradients