Correction: the SmolVLM data-filtering trick is from DeepSeek's earlier tech report

eliebakouch · x · 2026-09-10

Elie Bakouch clarifies his earlier post: the meticulous pretraining data handling — using SmolVLM to score image-text quality and extract high-quality interleaved data — comes from DeepSeek's earlier tech report.

Related event: DeepSeek Found Using SmolVLM to Curate Pretraining Image-Text Data(3 posts)→

Original post →

More from Research

Research channel →