DCVLM-Baseline (200B tokens)

DCVLM-Baseline (200B tokens) 🚨🚨🚨 NOTE: We are still uploading the dataset currently, we expect everything to be updated by 25th July, 2026. DCVLM-Baseline is the reference training mixture from our DataComp-VLM paper. It is a pre-mixed, decontaminated, ready-to-train multimodal pretraining dataset, materialized as flat WebDataset tar shards so it can be consumed by any training stack. This is the full-scale 200B-token (large-pool) release consisting of 103,985,276 samples.… See the full description on the dataset page:

Source: Hugging Face Hub (mlfoundations/dcvlm-baseline-200b). Metadata imported from the dataset’s Hub tags.