DeepSeek used SmolVLM to quality-filter interleaved pretraining data, researcher spots
eliebakouch · x · 2026-09-10
Hugging Face researcher Elie Bakouch spotted in DeepSeek's tech report that the company employed SmolVLM to run strict quality scoring on image-text content to extract high-quality interleaved pretraining data — a level of pretraining data care he says he hasn't seen in other models.
Related event: DeepSeek Found Using SmolVLM to Curate Pretraining Image-Text Data(3 posts)→
More from Research
- Deep Learning Course Adds Mechanistic Interpretability Lecture This Winter — CSProfKGD · 2026-10-26
- INCLUDE-2.0 benchmark evaluates regional knowledge across 91 languages and 86 countries — ABosselut · 2026-10-26
- CoRL paper: flow-matching VLAs can run 1-step Euler inference with no retraining — kaixhin · 2026-10-26
- CoVer at EMNLP 2026: Training Code LLMs to Generate and Self-Verify Simultaneously — algo_diver · 2026-10-26
- Robotic dexterity should be judged on speed, force control and tool use, not copying humans — DJiafei · 2026-10-26
- Turing-complete malware can run in PyTorch's backward pass via a zero-valued loss — retr0jirachi · 2026-10-26