Noetra engineer's first project: building an image encoder from scratch for multimodal LLMs
HirokatuKataoka · x · 2026-11-04
A Noetra engineer shares her first project at the company—a tech blog on building an image encoder from scratch for multimodal LLMs. Substance matches the team's post: a fully in-house encoder matching SigLIP2 with stronger Japanese and Chinese benchmarks.
More from Research
- Deep Principle open-sources Hydrogen-1.0, a science model trained on failed lab experiments, scoring 93% on GPQA Diamond — alex_verem · 2026-11-04
- Exactly solvable theory shows when Muon beats Adam: steep spectra, hard targets — CatAstro_Piyush · 2026-11-04
- UC Berkeley's open-source LEANN shrinks a 201GB vector index to 6GB for laptop RAG — thisguyknowsai · 2026-11-04
- Tsinghua & NVIDIA propose PrismWAM, splitting semantics, physics and actions in world models — jiqizhixin · 2026-11-04
- BAAI releases CS-Dialogue, a 104-hour Mandarin-English code-switching speech dataset — AdinaYakup · 2026-11-04
- KierstonBench simulates Allentown, PA to test LLMs on crime investigation—only frontier models hold up — Andrew Semenov · 2026-11-04