Parameter budgeting01
LoRA
Trade adapter rank for attention-matrix coverage without changing the parameter budget.
In the lab
Set rank to 4 with Q/V, then rank to 2 with Q/K/V/O. Watch the adapter count stay at 65,536.
An interactive companion for AI papers
Turn the paper you’re reading into a shareable interactive lab—with the model you choose.
Generate a shareable interactive explainer from an arXiv paper, or start with a currently public hands-on example.
The public labs shown below open without an account or API key.
Do more than summarize
Manipulate the mechanism.
Open a currently public lab, change the controls, and compare the explanation with its linked paper. Only public, ready labs appear here.
For the reader
The paper stays in the loop.
Flash Papers is for ML and research engineers, advanced learners and journal clubs, and paper authors who want a shareable interactive companion.
The explainer is generated by the model you choose. It can miss nuance or get a mechanism wrong, so every lab should be checked against its linked source. It is a learning aid, not a substitute for the paper.
Read the launch note →Your paper. Your model. A lab you can share.