Dispatch
Together AI open-sources a recipe to fine-tune a Jev-style classifier on Qwen3.5 4B for about $17 in 25 minutes
Together released together/Tev1-4B-experimental and a tev1 repo that trains it on 37,840 examples from eight sources: MultiNLI, BoolQ, Banking77, AG News, SST-5, 13,500 programmatic policy examples, 6,000 routing examples and 3,840 research-taxonomy examples. Training takes about 25 minutes on Together serverless. The post gives no accuracy comparison against hosted Jev, so treat it as a cost-of-entry data point, not a quality claim. A same-week parody, 'Jev in 25 Lines of Python' on Qwen3-0.6B (654 HN points), makes the same argument that the single-pass classifier layer is cheap to copy.
↳ Follow the thread