Hugging Face's logo

Datasets:
memecaps
/
memecaps-sample

Modalities:
Image
Text
Formats:
parquet
Size:
10K - 100K
Libraries:
Datasets
Dask
License:
Dataset card Viewer Files Community
1
memecaps-sample / README.md
pszemraj's picture
pszemraj
Adding prompt rewrite with groq model llama-3.1-70b-versatile
0e965ac verified
preview code
|
raw
history blame contribute delete
5.17 kB
metadata
license: odc-by
dataset_info:
  - config_name: captioned-detail-large
    features:
      - name: image
        dtype: image
      - name: caption
        dtype: string
      - name: prompt_rewrite
        dtype: string
    splits:
      - name: train
        num_bytes: 747445700.625
        num_examples: 3603
    download_size: 746084015
    dataset_size: 747445700.625
  - config_name: captioned-detail-large-gemma2-9b
    features:
      - name: image
        dtype: image
      - name: caption
        dtype: string
      - name: prompt_rewrite
        dtype: string
    splits:
      - name: train
        num_bytes: 747521994.625
        num_examples: 3603
    download_size: 746174675
    dataset_size: 747521994.625
  - config_name: captioned-detail-large-gp
    features:
      - name: image
        dtype: image
      - name: caption
        dtype: string
      - name: prompt_rewrite
        dtype: string
    splits:
      - name: train
        num_bytes: 747947300.625
        num_examples: 3603
    download_size: 746239690
    dataset_size: 747947300.625
  - config_name: captioned-detail-large-llama-3.1-70b
    features:
      - name: image
        dtype: image
      - name: caption
        dtype: string
      - name: prompt_rewrite
        dtype: string
    splits:
      - name: train
        num_bytes: 747465594.625
        num_examples: 3603
    download_size: 746095364
    dataset_size: 747465594.625
  - config_name: captioned-detail-large-llama-3.1-8b
    features:
      - name: image
        dtype: image
      - name: caption
        dtype: string
      - name: prompt_rewrite
        dtype: string
    splits:
      - name: train
        num_bytes: 747841533.625
        num_examples: 3603
    download_size: 746168591
    dataset_size: 747841533.625
  - config_name: captioned-detail-large-llama3-groq-70b-8192-tool-use
    features:
      - name: image
        dtype: image
      - name: caption
        dtype: string
      - name: prompt_rewrite
        dtype: string
    splits:
      - name: train
        num_bytes: 747358805.625
        num_examples: 3603
    download_size: 746039675
    dataset_size: 747358805.625
  - config_name: captioned-detail-large-mixtral-8x7b
    features:
      - name: image
        dtype: image
      - name: caption
        dtype: string
      - name: prompt_rewrite
        dtype: string
    splits:
      - name: train
        num_bytes: 747357344.625
        num_examples: 3603
    download_size: 746055347
    dataset_size: 747357344.625
  - config_name: captioned-detail-large-prompt2
    features:
      - name: image
        dtype: image
      - name: caption
        dtype: string
      - name: text
        dtype: string
    splits:
      - name: train
        num_bytes: 747434044.61
        num_examples: 3603
    download_size: 746102794
    dataset_size: 747434044.61
  - config_name: captioned-large
    features:
      - name: image
        dtype: image
      - name: caption
        dtype: string
    splits:
      - name: train
        num_bytes: 745851296.625
        num_examples: 3603
    download_size: 745056031
    dataset_size: 745851296.625
  - config_name: captioned-short-large
    features:
      - name: image
        dtype: image
      - name: caption
        dtype: string
    splits:
      - name: train
        num_bytes: 745171818.625
        num_examples: 3603
    download_size: 744727719
    dataset_size: 745171818.625
  - config_name: default
    features:
      - name: image
        dtype: image
    splits:
      - name: train
        num_bytes: 440542498.746
        num_examples: 3606
    download_size: 429650880
    dataset_size: 440542498.746
  - config_name: train-format
    features:
      - name: image
        dtype: image
      - name: text
        dtype: string
    splits:
      - name: train
        num_bytes: 745669893.61
        num_examples: 3603
    download_size: 745084778
    dataset_size: 745669893.61
configs:
  - config_name: captioned-detail-large
    data_files:
      - split: train
        path: captioned-detail-large/train-*
  - config_name: captioned-detail-large-gemma2-9b
    data_files:
      - split: train
        path: captioned-detail-large-gemma2-9b/train-*
  - config_name: captioned-detail-large-gp
    data_files:
      - split: train
        path: captioned-detail-large-gp/train-*
  - config_name: captioned-detail-large-llama-3.1-70b
    data_files:
      - split: train
        path: captioned-detail-large-llama-3.1-70b/train-*
  - config_name: captioned-detail-large-llama-3.1-8b
    data_files:
      - split: train
        path: captioned-detail-large-llama-3.1-8b/train-*
  - config_name: captioned-detail-large-llama3-groq-70b-8192-tool-use
    data_files:
      - split: train
        path: captioned-detail-large-llama3-groq-70b-8192-tool-use/train-*
  - config_name: captioned-detail-large-mixtral-8x7b
    data_files:
      - split: train
        path: captioned-detail-large-mixtral-8x7b/train-*
  - config_name: captioned-detail-large-prompt2
    data_files:
      - split: train
        path: captioned-detail-large-prompt2/train-*
  - config_name: captioned-large
    data_files:
      - split: train
        path: captioned-large/train-*
  - config_name: captioned-short-large
    data_files:
      - split: train
        path: captioned-short-large/train-*
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
  - config_name: train-format
    data_files:
      - split: train
        path: train-format/train-*

memecaps-sample

this is a prelim sample of images testing the effectiveness of various things.

All input images are always the same, the different configurations are either

  1. Different captioning levels of detail (Florence 2 large)
  2. different prompt for llama3-70b to rewrite