data/lilac.yml

# Lilac project config.
# See https://lilacml.com/api_reference/index.html#lilac.Config for details.

datasets:
  - namespace: local
    name: glue
    source:
      dataset_name: glue
      config_name: ax
      source_name: huggingface
    embeddings:
      - path: premise
        embedding: gte-small
      - path: premise
        embedding: gte-base
      - path: hypothesis
        embedding: gte-small
    signals:
      - path: premise
        signal:
          signal_name: pii
      - path: hypothesis
        signal:
          signal_name: pii
      - path: premise
        signal:
          signal_name: text_statistics
    settings:
      ui:
        media_paths:
          - premise
        markdown_paths: []
  - namespace: local
    name: glue_ax
    source:
      dataset_name: glue
      config_name: ax
      source_name: huggingface
    embeddings:
      - path: hypothesis
        embedding: gte-small
    signals:
      - path: premise
        signal:
          signal_name: text_statistics
      - path: premise
        signal:
          signal_name: pii
      - path: premise
        signal:
          signal_name: near_dup
      - path: hypothesis
        signal:
          embedding: gte-small
          namespace: ''
          concept_name: ''
          signal_name: concept_score
      - path: hypothesis
        signal:
          embedding: gte-small
          namespace: lilac
          concept_name: positive-sentiment
          signal_name: concept_score
      - path: hypothesis
        signal:
          embedding: gte-small
          namespace: lilac
          concept_name: non-english
          signal_name: concept_score
    settings:
      ui:
        media_paths:
          - hypothesis
        markdown_paths: []
  - namespace: local
    name: imdb3
    source:
      dataset_name: imdb
      source_name: huggingface
    settings:
      ui:
        media_paths:
          - text
        markdown_paths: []
  - namespace: local
    name: imdb
    source:
      dataset_name: imdb
      source_name: huggingface
    embeddings:
      - path: text
        embedding: gte-small
    signals:
      - path: text
        signal:
          signal_name: pii
      - path: text
        signal:
          signal_name: text_statistics
    settings:
      ui:
        media_paths:
          - text
        markdown_paths: []
  - namespace: local
    name: imdb2
    source:
      dataset_name: imdb
      source_name: huggingface
    settings:
      ui:
        media_paths:
          - text
        markdown_paths: []
  - namespace: lilac
    name: OpenOrca-100k
    source:
      dataset_name: Open-Orca/OpenOrca
      sample_size: 100000
      source_name: huggingface
    embeddings:
      - path: question
        embedding: gte-small
      - path: response
        embedding: gte-small
    signals:
      - path: question
        signal:
          signal_name: near_dup
      - path: question
        signal:
          signal_name: pii
      - path: question
        signal:
          signal_name: lang_detection
      - path: question
        signal:
          embedding: gte-small
          namespace: lilac
          concept_name: positive-sentiment
          signal_name: concept_score
      - path: question
        signal:
          embedding: gte-small
          namespace: lilac
          concept_name: non-english
          signal_name: concept_score
      - path: question
        signal:
          embedding: gte-small
          namespace: lilac
          concept_name: toxicity
          signal_name: concept_score
      - path: question
        signal:
          embedding: gte-small
          namespace: lilac
          concept_name: question
          signal_name: concept_score
      - path: question
        signal:
          embedding: gte-small
          namespace: lilac
          concept_name: legal-termination
          signal_name: concept_score
      - path: question
        signal:
          embedding: gte-small
          namespace: lilac
          concept_name: source-code
          signal_name: concept_score
      - path: question
        signal:
          embedding: gte-small
          namespace: lilac
          concept_name: negative-sentiment
          signal_name: concept_score
      - path: question
        signal:
          embedding: gte-small
          namespace: lilac
          concept_name: profanity
          signal_name: concept_score
      - path: question
        signal:
          signal_name: text_statistics
      - path: response
        signal:
          signal_name: near_dup
      - path: response
        signal:
          signal_name: pii
      - path: response
        signal:
          signal_name: lang_detection
      - path: response
        signal:
          embedding: gte-small
          namespace: lilac
          concept_name: positive-sentiment
          signal_name: concept_score
      - path: response
        signal:
          embedding: gte-small
          namespace: lilac
          concept_name: non-english
          signal_name: concept_score
      - path: response
        signal:
          embedding: gte-small
          namespace: lilac
          concept_name: toxicity
          signal_name: concept_score
      - path: response
        signal:
          embedding: gte-small
          namespace: lilac
          concept_name: question
          signal_name: concept_score
      - path: response
        signal:
          embedding: gte-small
          namespace: lilac
          concept_name: legal-termination
          signal_name: concept_score
      - path: response
        signal:
          embedding: gte-small
          namespace: lilac
          concept_name: source-code
          signal_name: concept_score
      - path: response
        signal:
          embedding: gte-small
          namespace: lilac
          concept_name: negative-sentiment
          signal_name: concept_score
      - path: response
        signal:
          embedding: gte-small
          namespace: lilac
          concept_name: profanity
          signal_name: concept_score
      - path: response
        signal:
          signal_name: text_statistics
      - path: system_prompt
        signal:
          signal_name: pii
    settings:
      ui:
        media_paths:
          - question
          - response
        markdown_paths: []
  - namespace: local
    name: the_movies_dataset
    source:
      filepaths:
        - gs://lilac-data/datasets/the_movies_dataset/the_movies_dataset.csv
      names: []
      source_name: csv
    settings:
      ui:
        media_paths:
          - overview
        markdown_paths: []
  - namespace: local
    name: glue_ax_parquet
    source:
      filepaths:
        - gs://lilac-data/datasets/glue_ax_parquet/glue_ax.parquet
      source_name: parquet
    settings:
      ui:
        media_paths:
          - premise
        markdown_paths: []
  - namespace: lilac
    name: mmlu_professional_law
    source:
      dataset_name: cais/mmlu
      config_name: professional_law
      source_name: huggingface
    embeddings:
      - path: question
        embedding: gte-small
      - path:
          - choices
          - '*'
        embedding: gte-small
    signals:
      - path: question
        signal:
          signal_name: near_dup
      - path: question
        signal:
          signal_name: pii
      - path: question
        signal:
          signal_name: lang_detection
      - path: question
        signal:
          embedding: gte-small
          namespace: lilac
          concept_name: positive-sentiment
          signal_name: concept_score
      - path: question
        signal:
          embedding: gte-small
          namespace: lilac
          concept_name: non-english
          signal_name: concept_score
      - path: question
        signal:
          embedding: gte-small
          namespace: lilac
          concept_name: toxicity
          signal_name: concept_score
      - path: question
        signal:
          embedding: gte-small
          namespace: lilac
          concept_name: question
          signal_name: concept_score
      - path: question
        signal:
          embedding: gte-small
          namespace: lilac
          concept_name: legal-termination
          signal_name: concept_score
      - path: question
        signal:
          embedding: gte-small
          namespace: lilac
          concept_name: source-code
          signal_name: concept_score
      - path: question
        signal:
          embedding: gte-small
          namespace: lilac
          concept_name: negative-sentiment
          signal_name: concept_score
      - path: question
        signal:
          embedding: gte-small
          namespace: lilac
          concept_name: profanity
          signal_name: concept_score
      - path: question
        signal:
          signal_name: text_statistics
      - path:
          - choices
          - '*'
        signal:
          signal_name: near_dup
      - path:
          - choices
          - '*'
        signal:
          signal_name: pii
      - path:
          - choices
          - '*'
        signal:
          signal_name: lang_detection
      - path:
          - choices
          - '*'
        signal:
          embedding: gte-small
          namespace: lilac
          concept_name: positive-sentiment
          signal_name: concept_score
      - path:
          - choices
          - '*'
        signal:
          embedding: gte-small
          namespace: lilac
          concept_name: non-english
          signal_name: concept_score
      - path:
          - choices
          - '*'
        signal:
          embedding: gte-small
          namespace: lilac
          concept_name: toxicity
          signal_name: concept_score
      - path:
          - choices
          - '*'
        signal:
          embedding: gte-small
          namespace: lilac
          concept_name: question
          signal_name: concept_score
      - path:
          - choices
          - '*'
        signal:
          embedding: gte-small
          namespace: lilac
          concept_name: legal-termination
          signal_name: concept_score
      - path:
          - choices
          - '*'
        signal:
          embedding: gte-small
          namespace: lilac
          concept_name: source-code
          signal_name: concept_score
      - path:
          - choices
          - '*'
        signal:
          embedding: gte-small
          namespace: lilac
          concept_name: negative-sentiment
          signal_name: concept_score
      - path:
          - choices
          - '*'
        signal:
          embedding: gte-small
          namespace: lilac
          concept_name: profanity
          signal_name: concept_score
      - path:
          - choices
          - '*'
        signal:
          signal_name: text_statistics
    settings:
      ui:
        media_paths:
          - question
          - - choices
            - '*'
        markdown_paths: []
      preferred_embedding: gte-small
  - namespace: local
    name: deepset-prompt-inj
    source:
      dataset_name: deepset/prompt-injections
      source_name: huggingface
    embeddings:
      - path: text
        embedding: gte-small
    settings:
      ui:
        media_paths:
          - text
        markdown_paths: []
  - namespace: local
    name: jasper-prompt-inj
    source:
      dataset_name: JasperLS/prompt-injections
      source_name: huggingface
    embeddings:
      - path: text
        embedding: gte-small
    settings:
      ui:
        media_paths:
          - text
        markdown_paths: []
  - namespace: local
    name: mosaic-chat-v2
    source:
      dataset_name: sam-mosaic/chat-v2
      source_name: huggingface
    embeddings:
      - path: prompt
        embedding: gte-small
      - path: response
        embedding: gte-small
    signals:
      - path: prompt
        signal:
          signal_name: near_dup
      - path: prompt
        signal:
          signal_name: pii
      - path: prompt
        signal:
          signal_name: lang_detection
      - path: prompt
        signal:
          embedding: gte-small
          namespace: lilac
          concept_name: non-english
          signal_name: concept_score
      - path: prompt
        signal:
          embedding: gte-small
          namespace: lilac
          concept_name: toxicity
          signal_name: concept_score
      - path: prompt
        signal:
          embedding: gte-small
          namespace: lilac
          concept_name: source-code
          signal_name: concept_score
      - path: prompt
        signal:
          embedding: gte-small
          namespace: lilac
          concept_name: negative-sentiment
          signal_name: concept_score
      - path: prompt
        signal:
          embedding: gte-small
          namespace: lilac
          concept_name: profanity
          signal_name: concept_score
      - path: prompt
        signal:
          signal_name: text_statistics
      - path: response
        signal:
          signal_name: near_dup
      - path: response
        signal:
          signal_name: pii
      - path: response
        signal:
          signal_name: lang_detection
      - path: response
        signal:
          embedding: gte-small
          namespace: lilac
          concept_name: non-english
          signal_name: concept_score
      - path: response
        signal:
          embedding: gte-small
          namespace: lilac
          concept_name: toxicity
          signal_name: concept_score
      - path: response
        signal:
          embedding: gte-small
          namespace: lilac
          concept_name: source-code
          signal_name: concept_score
      - path: response
        signal:
          embedding: gte-small
          namespace: lilac
          concept_name: negative-sentiment
          signal_name: concept_score
      - path: response
        signal:
          embedding: gte-small
          namespace: lilac
          concept_name: profanity
          signal_name: concept_score
      - path: response
        signal:
          signal_name: text_statistics
    settings:
      ui:
        media_paths:
          - prompt
          - response
        markdown_paths: []
      preferred_embedding: gte-small
  - namespace: local
    name: databricks-dolly-15k-curated-en
    source:
      dataset_name: argilla/databricks-dolly-15k-curated-en
      source_name: huggingface
    embeddings:
      - path: original-context
        embedding: gte-small
      - path:
          - new-context
          - value
          - '*'
        embedding: gte-small
      - path: original-instruction
        embedding: gte-small
    signals:
      - path: original-instruction
        signal:
          signal_name: near_dup
      - path: original-instruction
        signal:
          signal_name: pii
      - path: original-instruction
        signal:
          signal_name: lang_detection
      - path: original-instruction
        signal:
          signal_name: text_statistics
      - path: original-context
        signal:
          signal_name: near_dup
      - path: original-context
        signal:
          signal_name: pii
      - path: original-context
        signal:
          signal_name: lang_detection
      - path: original-context
        signal:
          embedding: gte-small
          namespace: lilac
          concept_name: positive-sentiment
          signal_name: concept_score
      - path: original-context
        signal:
          embedding: gte-small
          namespace: lilac
          concept_name: non-english
          signal_name: concept_score
      - path: original-context
        signal:
          embedding: gte-small
          namespace: lilac
          concept_name: toxicity
          signal_name: concept_score
      - path: original-context
        signal:
          embedding: gte-small
          namespace: lilac
          concept_name: question
          signal_name: concept_score
      - path: original-context
        signal:
          embedding: gte-small
          namespace: lilac
          concept_name: legal-termination
          signal_name: concept_score
      - path: original-context
        signal:
          embedding: gte-small
          namespace: lilac
          concept_name: source-code
          signal_name: concept_score
      - path: original-context
        signal:
          embedding: gte-small
          namespace: lilac
          concept_name: negative-sentiment
          signal_name: concept_score
      - path: original-context
        signal:
          embedding: gte-small
          namespace: lilac
          concept_name: profanity
          signal_name: concept_score
      - path: original-context
        signal:
          signal_name: text_statistics
      - path: original-response
        signal:
          signal_name: near_dup
      - path: original-response
        signal:
          signal_name: pii
      - path: original-response
        signal:
          signal_name: lang_detection
      - path: original-response
        signal:
          signal_name: text_statistics
      - path:
          - new-instruction
          - value
          - '*'
        signal:
          signal_name: near_dup
      - path:
          - new-instruction
          - value
          - '*'
        signal:
          signal_name: pii
      - path:
          - new-instruction
          - value
          - '*'
        signal:
          signal_name: lang_detection
      - path:
          - new-instruction
          - value
          - '*'
        signal:
          signal_name: text_statistics
      - path:
          - new-context
          - value
          - '*'
        signal:
          signal_name: near_dup
      - path:
          - new-context
          - value
          - '*'
        signal:
          signal_name: pii
      - path:
          - new-context
          - value
          - '*'
        signal:
          signal_name: lang_detection
      - path:
          - new-context
          - value
          - '*'
        signal:
          embedding: gte-small
          namespace: lilac
          concept_name: positive-sentiment
          signal_name: concept_score
      - path:
          - new-context
          - value
          - '*'
        signal:
          embedding: gte-small
          namespace: lilac
          concept_name: non-english
          signal_name: concept_score
      - path:
          - new-context
          - value
          - '*'
        signal:
          embedding: gte-small
          namespace: lilac
          concept_name: toxicity
          signal_name: concept_score
      - path:
          - new-context
          - value
          - '*'
        signal:
          embedding: gte-small
          namespace: lilac
          concept_name: question
          signal_name: concept_score
      - path:
          - new-context
          - value
          - '*'
        signal:
          embedding: gte-small
          namespace: lilac
          concept_name: legal-termination
          signal_name: concept_score
      - path:
          - new-context
          - value
          - '*'
        signal:
          embedding: gte-small
          namespace: lilac
          concept_name: source-code
          signal_name: concept_score
      - path:
          - new-context
          - value
          - '*'
        signal:
          embedding: gte-small
          namespace: lilac
          concept_name: negative-sentiment
          signal_name: concept_score
      - path:
          - new-context
          - value
          - '*'
        signal:
          embedding: gte-small
          namespace: lilac
          concept_name: profanity
          signal_name: concept_score
      - path:
          - new-context
          - value
          - '*'
        signal:
          signal_name: text_statistics
      - path:
          - new-response
          - value
          - '*'
        signal:
          signal_name: near_dup
      - path:
          - new-response
          - value
          - '*'
        signal:
          signal_name: pii
      - path:
          - new-response
          - value
          - '*'
        signal:
          signal_name: lang_detection
      - path:
          - new-response
          - value
          - '*'
        signal:
          signal_name: text_statistics
      - path: original-instruction
        signal:
          signal_name: spacy_ner
    settings:
      ui:
        media_paths:
          - original-instruction
          - original-context
          - original-response
          - - new-instruction
            - value
            - '*'
          - - new-context
            - value
            - '*'
          - - new-response
            - value
            - '*'
        markdown_paths: []
      preferred_embedding: gte-small
  - namespace: local
    name: open-asssistant-conversations
    source:
      dataset_name: OpenAssistant/oasst1
      source_name: huggingface
    embeddings:
      - path: text
        embedding: gte-small
    signals:
      - path: text
        signal:
          signal_name: near_dup
      - path: text
        signal:
          signal_name: pii
      - path: text
        signal:
          signal_name: lang_detection
      - path: text
        signal:
          embedding: gte-small
          namespace: lilac
          concept_name: positive-sentiment
          signal_name: concept_score
      - path: text
        signal:
          embedding: gte-small
          namespace: lilac
          concept_name: non-english
          signal_name: concept_score
      - path: text
        signal:
          embedding: gte-small
          namespace: lilac
          concept_name: toxicity
          signal_name: concept_score
      - path: text
        signal:
          embedding: gte-small
          namespace: lilac
          concept_name: question
          signal_name: concept_score
      - path: text
        signal:
          embedding: gte-small
          namespace: lilac
          concept_name: legal-termination
          signal_name: concept_score
      - path: text
        signal:
          embedding: gte-small
          namespace: lilac
          concept_name: source-code
          signal_name: concept_score
      - path: text
        signal:
          embedding: gte-small
          namespace: lilac
          concept_name: negative-sentiment
          signal_name: concept_score
      - path: text
        signal:
          embedding: gte-small
          namespace: lilac
          concept_name: negative-sentiment
          signal_name: concept_score
      - path: text
        signal:
          embedding: gte-small
          namespace: lilac
          concept_name: profanity
          signal_name: concept_score
      - path: text
        signal:
          signal_name: text_statistics
    settings:
      ui:
        media_paths:
          - text
        markdown_paths: []
      preferred_embedding: gte-small
  - namespace: local
    name: enron-emails
    source:
      dataset_name: EleutherAI/pile
      config_name: enron_emails
      sample_size: 100000
      source_name: huggingface
    embeddings:
      - path: text
        embedding: gte-small
    signals:
      - path: text
        signal:
          signal_name: near_dup
      - path: text
        signal:
          signal_name: pii
      - path: text
        signal:
          signal_name: lang_detection
      - path: text
        signal:
          embedding: gte-small
          namespace: lilac
          concept_name: positive-sentiment
          signal_name: concept_score
      - path: text
        signal:
          embedding: gte-small
          namespace: lilac
          concept_name: non-english
          signal_name: concept_score
      - path: text
        signal:
          embedding: gte-small
          namespace: lilac
          concept_name: toxicity
          signal_name: concept_score
      - path: text
        signal:
          embedding: gte-small
          namespace: lilac
          concept_name: question
          signal_name: concept_score
      - path: text
        signal:
          embedding: gte-small
          namespace: lilac
          concept_name: legal-termination
          signal_name: concept_score
      - path: text
        signal:
          embedding: gte-small
          namespace: lilac
          concept_name: source-code
          signal_name: concept_score
      - path: text
        signal:
          embedding: gte-small
          namespace: lilac
          concept_name: negative-sentiment
          signal_name: concept_score
      - path: text
        signal:
          embedding: gte-small
          namespace: lilac
          concept_name: profanity
          signal_name: concept_score
      - path: text
        signal:
          signal_name: text_statistics
    settings:
      ui:
        media_paths:
          - text
        markdown_paths: []
      preferred_embedding: gte-small