datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
longvideogen_wavespeed_compact_3_trial3Compact_VLM_filter_data
Filtration-Oriented Image-Caption Dataset
This dataset is created to train a small Vision-Language Model (VLM) that learns in-context criteria to filter noisy web-scale image-caption pairs.
We used the base Qwen2-VL-2B model to fine-tune a filtration-oriented variant, optimized to assess and filter large datasets efficiently. The goal is to build a lightweight VLM that can be deployed locally, reducing dependency on large-scale APIs and minimizing both compute costs and latency.… See the full description on the dataset page: https://huggingface.co/datasets/Dauka-transformers/Compact_VLM_filter_data.judge_prompt_dataset_v1_dedup_compactcord-v2-compactfood-kb-compactjudge_prompt_dataset_v1_dedup_compact_flatFineTree_V2-aggresive-compact-approved-no-bbox-validationFineTree_V2-aggresive-compact-approved-no-bbox-minimal-instruction-validationFineTree_V2-aggresive-compact-approved-trainFineTree_V2-aggresive-compact-approved-minimal-instruction-validationFineTree_V2-aggresive-compact-approved-no-bbox-trainFineTree_V2-aggresive-compact-approved-validationFineTree_V2-aggresive-compact-approved-minimal-instruction-trainFineTree_V2-aggresive-compact-approved-no-bbox-minimal-instruction-train