What they found
The authors describe ArchSpace, a public platform for recording, reviewing and referencing research on LLM architecture. Each proposal is stored with its experiments and reasoning, reviewed by the community, and cited by later work. To judge ideas before expensive large-scale training, the platform uses scaling laws. These are fits that predict a model's loss at larger sizes from smaller experiments. The authors provide starting references: Qwen3.5 and OLMo3 baselines, fitted scaling laws, and a reusable training recipe. The abstract gives no experimental results, so it does not show how well the predictions hold up.
Why it matters
Model architecture affects how well SI systems scale, how efficiently they train, and what they cost to run. The authors say industry reports rarely record which ideas failed and why. They also say academic tests are often small and use inconsistent baselines. A shared record with common baselines could cut repeated work and make comparisons fairer.
How it changes the race
If it works, groups with little compute could test architecture ideas against shared baselines and get a credible read on larger scales. That could widen who contributes ideas. The abstract reports no results yet, so the effect is unproven.
The abstract, by the authors
Model architecture shapes the scaling behavior, training efficiency, and inference cost of LLMs, making it a central focus of both industry and academia. In industry, teams such as those behind DeepSeek, Qwen, and Kimi routinely introduce architectural innovations, yet public reports rarely document which ideas succeeded, which were abandoned, and why. In academia, limited compute and inconsistent baselines often restrict evaluations to small-scale, single-point comparisons, making it difficult to establish whether improvements hold at larger scales or translate into broader impact. ArchSpace addresses both gaps as a public platform for Recording, Reviewing, and Referencing Architecture Researc}: each proposal is recorded with its experiments and reasoning, reviewed by the community, and referenced by later work. To evaluate ideas before costly large-scale training, ArchSpace leverages scaling laws to predict model loss at larger scales from smaller experiments, a practice widely adopted in industry. We provide initial references that make the platform usable: Qwen3.5 and OLMo3 baselines, fitted scaling laws, and a reusable training recipe. Ideas validated through this review can in turn become new baselines, and ArchSpace thereby aims to broaden the range of ideas that shape the next generation of models.
Quoted from arXiv:2611.17332. The research and its words belong to its authors.
