BootsofLagrangian/commoncatalog-cc-by-recap-qwen3p5-35b-a3b

CommonCatalog CC-BY recaptions with Qwen3.5-35B-A3B

Dataset commoncatalog-cc-by: 14.577 Million caption rows.

This public caption-only repository contains 14,576,560 generated captions for 14,576,558 image assets and no image payload. It includes 2 additional distinct caption variants. Rows match the public image-bearing common-canvas/commoncatalog-cc-by release at revision 80f50fe4a1ca937f37a11be3f8eee5199d776ff3 through Flickr photoid, represented here by asset_instance_id, and the source Parquet row locator in image_shard plus image_member. URL and content hashes are retained for independent reconciliation.

Original source text and our recap

The pinned upstream image record retains its own text fields, including caption, description, title, usertags, and machinetags. It also contains the upstream release's blip2_caption. Those fields belong to the original/upstream dataset and are not our output. This repository does not copy or rewrite them. Our only generated text field is the separate caption_text produced with Qwen3.5. It does not replace or claim to correct any upstream text field.

Load captions or pinned source images

from datasets import load_dataset

captions = load_dataset(
    "BootsofLagrangian/commoncatalog-cc-by-recap-qwen3p5-35b-a3b",
    split="train",
    streaming=True,
)
caption_row = next(iter(captions))
print(caption_row["image_shard"], caption_row["image_member"], caption_row["caption_text"])

source_images = load_dataset(
    "common-canvas/commoncatalog-cc-by",
    split="train",
    streaming=True,
    revision="80f50fe4a1ca937f37a11be3f8eee5199d776ff3",
)
source_row = next(iter(source_images))
image = source_row["jpg"]
print(
    source_row["caption"],
    source_row["description"],
    source_row["title"],
    source_row["usertags"],
    source_row["machinetags"],
    source_row["blip2_caption"],
)

The two examples intentionally load separate repositories. Caption rows retain the pinned source Parquet filename and row locator; do not assume an unpinned moving main has the same row layout.

Captions were produced with the same caption-generation method used in A Matched-Budget Audit Framework for Recaptioned Image-Text Supervision Distributions: Qwen/Qwen3.5-35B-A3B-FP8, one long-caption stage, a 512-token output limit, Qwen3.5 release sampling defaults, free-form output, and vLLM 0.17.1 serving. Thinking was disabled. Output received whitespace stripping only; no semantic post-processing was applied. See generation_config.yaml for the system prompt, user prompt, and reproduction hyperparameters.

This release is for research and audit reproducibility. It is not training-ready: users must perform semantic post-processing, safety and policy filtering, URL/content identity verification, and image-rights review before training use.

CC BY 4.0 applies only to the generated caption text in this repository. It does not license source images or third-party metadata.

@misc{oh2026matchedbudget,
  title={A Matched-Budget Audit Framework for Recaptioned Image-Text Supervision Distributions},
  author={Giyeong Oh and Junghun Park and Yuhan Bae and Youngjae Yu},
  year={2026}
}