← sources

LessThink-Qwen3-4B: the same model, with far less thinking [P]

source: reddit (reddit:MachineLearning)published 2026-09-30 07:19ingested 2026-10-01 02:40

retained text

I post-trained Qwen3-4B to spend 44% fewer tokens on reasoning, keeping its knowledge and answer style. The whole pipeline ran on one GPU.

folks, you can check it out on : https://5ivatej.com/lessthink/

document fetched 2026-10-01 00:30 · canonical https://reddit.com/r/MachineLearning/comments/1wtygav/lessthinkqwen34b_the_same_model_with_far_less