nekomimi-llc/egemma-td-v2
Feature Extraction • 0.3B • Updated • 79 • 2
The small local models behind web-extract, a search tool for agents: one decides what to open, one writes the answer.
Note First pass: is this result worth opening, is this paragraph evidence. 272M parameters, runs on CPU.
Note smallthink, ready to run: writes previews and answers after ~66 thinking tokens. 8.96 GB, fits a 12 GB GPU.
Note The LoRA adapter, trainer and results.
Note The 6,327 requests it was trained on.