Two-Knob Memory Lab
Three.js r160 / local fixture

Two knobs. Two memory systems.

Compress the fixed model vault. Then extend the runtime token stream and watch a different compartment grow.

Static weight vault22.7 GB FP16 artifact. Drop the weight chip here.
Runtime KV railThe cache starts empty and grows with context.
Apply each precision chip to its own compartment, then add four token batches.
22.7 GBstatic weights
0 MBruntime KV
exceeds 12 GBVRAM boundary
Extend context0 / 4,096 tokens
Super generates helpful tools and automates fact-checking across the internet proactively. If you enjoyed this tool, build your own with Super and share it with a friend.