https://duoan.github.io/tags/siq-vl/ Recent content in SiQ-VL on Duo's Tech Blog https://duoan.github.io/images/papermod-cover.png https://duoan.github.io/images/papermod-cover.png Hugo -- 0.153.1 en-us Sun, 12 Jul 2026 00:00:00 +0000 https://duoan.github.io/posts/siq-vl-curriculum-under-compute-constraints/ Mon, 15 Dec 2025 00:00:00 +0000 https://duoan.github.io/posts/siq-vl-curriculum-under-compute-constraints/ <h1 id="siq-vl-a-curriculum-for-small-vlms-when-compute-is-the-hard-constraint">SiQ-VL: A Curriculum for Small VLMs When Compute Is the Hard Constraint</h1> <p>Most VLM writeups assume a cluster. SiQ-VL started from the opposite constraint: <strong>one (or few) GPUs</strong>, and the question was which design choices still buy capability when you cannot buy FLOPs.</p> <p>This post is the consolidated field guide for that project — architecture, token economics, staged training, and offline Chain-of-Thought (CoT) distillation — replacing three earlier notes that said the same thing three ways. Kernel-level throughput (how we pushed Stage-1 from ~15K to ~100K real tokens/s on Blackwell) lives in the companion post: <a href="https://duoan.github.io/posts/optimizing-vlm-training-on-one-gpu/">Optimizing VLM Training on One GPU</a>.</p>