Hugging Face's logo

Datasets:
moeshawky
/
laion-qwen-siglip-pairs

Tasks:
Image-Text-to-Text
Languages:
English
Size:
10K<n<100K
Tags:
vision-language
contrastive
laion
qwen3
siglip2
License:
Dataset card Files
xet
 Community
Dataset Viewer
The dataset viewer is not available for this subset.
Cannot get the split names for the config 'default' of the dataset.
Exception:    SplitsNotFoundError
Message:      The split names could not be parsed from the dataset config.
Traceback:    Traceback (most recent call last):
                File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 286, in get_dataset_config_info
                  for split_generator in builder._split_generators(
                                         ~~~~~~~~~~~~~~~~~~~~~~~~~^
                      StreamingDownloadManager(base_path=builder.base_path, download_config=download_config)
                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                  )
                  ^
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/webdataset/webdataset.py", line 80, in _split_generators
                  raise ValueError(
                  ...<2 lines>...
                  )
              ValueError: The TAR archives of the dataset should be in WebDataset format, but the files in the archive don't share the same prefix or the same types.
              
              The above exception was the direct cause of the following exception:
              
              Traceback (most recent call last):
                File "/src/services/worker/src/worker/job_runners/config/split_names.py", line 68, in compute_split_names_from_streaming_response
                  for split in get_dataset_split_names(
                               ~~~~~~~~~~~~~~~~~~~~~~~^
                      path=dataset,
                      ^^^^^^^^^^^^^
                      config_name=config,
                      ^^^^^^^^^^^^^^^^^^^
                      token=hf_token,
                      ^^^^^^^^^^^^^^^
                  )
                  ^
                File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 340, in get_dataset_split_names
                  info = get_dataset_config_info(
                      path,
                  ...<6 lines>...
                      **config_kwargs,
                  )
                File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 291, in get_dataset_config_info
                  raise SplitsNotFoundError("The split names could not be parsed from the dataset config.") from err
              datasets.inspect.SplitsNotFoundError: The split names could not be parsed from the dataset config.

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

LAION-400M curated pairs for Qwen3 x SigLIP2 graft training

92,823 image-text pairs sampled deterministically from the public LAION-400M metadata parquet (Kaggle romainbeaumont/laion400m), fetched 2026-10-13, for training a frozen-tower vision-language alignment head (SigLIP2 vision tower projected into the Qwen3 text-embedding space).

Provenance

  • Sampling: 220,000 rows drawn from 192,434,199 stageable rows across all 32 parquet parts (seed 20260918; 46.5% gate yield: nsfw==UNLIKELY, similarity >= 0.3, height/width >= 256).
  • Fetch: two passes (48-thread + 500-concurrency async). LAION 2021 link rot measured at ~93% on the tail pool (404 27%, ConnectError 29%, 403 20%); 60% survival on the live prefix. Per-URL receipts: fetch_ledger.json (status, bytes, sha256, format).
  • Bridge: async-fetched files renamed to the {split}_{basename(url)} convention with sha256-identical dedupe; 174 ambiguous-provenance files skipped.
  • Split: deterministic 90/5/5 by sha256(url) — reproducible on any machine. PoC-overlap gate re-verified: 0 shared URLs with the PoC set.

Layout 

shard-00000.tar .. shard-00018.tar         19 deterministic GNU tar shards at
                                            repo root, 5,000 images each
                                            (tar members are flat:
                                            {split}_{basename(url)})
SHARDS.json   per-shard file count, sha256, byte size — verify before unpack
pairs.jsonl   one JSON object per row:
              {split, sample_id, file_name, caption, url, height, width, shard}
fetch_ledger.json   per-URL fetch receipts

Unpack + verify (receiving machine) 

hf download moeshawky/laion-qwen-siglip-pairs --repo-type dataset --local-dir ./data
python - <<'EOF'
import json, hashlib
for s in json.load(open("data/SHARDS.json")):
    h = hashlib.sha256(open(f"data/{s["file"]}","rb").read()).hexdigest()
    assert h == s["sha256"], s["file"]
print("shards verified")
EOF
mkdir -p data/images && for f in data/*.tar; do tar -xf "$f" -C data/images; done

file_name in pairs.jsonl then resolves directly under data/images/.

Content policy

Raw web-scraped captions and unfiltered web imagery — NO NSFW/aesthetics/ language filtering was applied (operator-mandated web preservation for corpus authenticity). Use accordingly.

Source metadata: https://huggingface.co/datasets/romainbeaumont/laion400m
Downloads last month
85
Total file size:
10.8 GB

Models trained or fine-tuned on moeshawky/laion-qwen-siglip-pairs