Project Greenhouse Trains Open-Source Search Reranker From Scratch on a Few GPUs

Jimmy Lin's team at Waterloo published Project Greenhouse, showing a competitive fully open-source pointwise reranker can be trained from scratch with modest compute—ClimbMix pretraining followed by supervised fine-tuning on RLHN relevance data—lowering the barrier to sovereign agentic search.

2026-10-09 ~ 2026-10-10 · 4 related posts

• Project Greenhouse: Jimmy Lin Team Trains Fully Open Reranker from Scratch on a Handful of GPUs — _reachsumit · 2026-10-09
• Project Greenhouse builds an open agentic search reranker from scratch on a few GPUs — lintool · 2026-10-10
• Not even that hard: Greenhouse's from-scratch reranker trained on just a few GPUs — lintool · 2026-10-10
• From-scratch open decoder-only reranker built with just a few GPUs — lintool · 2026-10-10