Project Greenhouse: fully open, sovereign agentic search models on a handful of GPUs
lintool · x · 2026-10-20
Jimmy Lin's team released the arXiv paper Project Greenhouse (2610.11922), arguing that fully open and sovereign agentic search models can be built with only modest compute.
- Method: a two-step recipe — pretraining from scratch followed by supervised fine-tuning, using only commonly available datasets; deliberately avoiding third-party open-weight backbones for end-to-end control of training.
- Result: a competitive pointwise decoder-only reranker, with the bulk of experiments done on no more than a handful of GPUs.
- Open artifacts: data, code, configs, and checkpoints for the Gaggle model family, enabling transparent, independent reproduction.
The authors frame this as a first step toward validating their broader thesis of sovereign agentic search on small compute.
More from Research
- Gojo project generates 108 ornament designs from 36 skeletons with hierarchical pattern-aware coloring — debreuil · 2026-10-20
- Paper: 'Before They Can Solve' predicts a base model's post-training potential early — hyunw_kim · 2026-10-20
- Base-model probes predict cross-benchmark coding-agent potential at ρ=0.976 — hyunw_kim · 2026-10-20
- Three probes track post-training SWE-bench rankings, BPB hits ρ=0.964 — hyunw_kim · 2026-10-20
- NVIDIA asks: must you post-train every base model to find the best coding agent? — hyunw_kim · 2026-10-20
- SaTML got 450 submissions, 2x last year — chairs early-rejected ~40% of them — m2saxon · 2026-10-20