Image-text parquet dataset generated from: cc12m, datacomp12m, yfcc15m, mscoco.
Images are stored in the Hugging Face Image column format as encoded bytes.
Training rows come from --sources; validation/support rows come from
--eval-sources.
Generation arguments:
{
"cache_dir": null,
"chunk_size": 10000,
"compression": "zstd",
"download_max_retries": 5,
"download_num_proc": 8,
"eval_sources": [
"mscoco"
],
"hf_num_proc": 1,
"image_format":… See the full description on the dataset page:
https://huggingface.co/datasets/MVRL/WebMix-38M.
