niltheory's picture
Update README.md
8ea11e4
|
raw
history blame
1.56 kB
metadata
model-index:
  - name: Existence Analysis
    results:
      - task:
          type: text-classification
        dataset:
          name: niltheory/ExistenceTypes
          type: parquet
        metrics:
          - name: AI2 Reasoning Challenge (25-Shot)
            type: AI2 Reasoning Challenge (25-Shot)
            value: 64.59
        source:
          name: Open LLM Leaderboard
          url: https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard
language:
  - en
metrics:
  - accuracy

Existence Analysis Model (EAM)

Created for: Compendium Terminum, IP
Base Model: bert-large-cased-whole-word-masking

Iterative Development

Iteration #1:

  • Initial Model: Utilized distilBert for foundational training.
  • Dataset Size: 96 entries.
  • Outcome: Established baseline for accuracy metrics.

Iteration #2:

  • Model Upgrade: Transitioned to bert-base-uncased from distilbert-base-uncased.
  • Dataset Expansion: Increased from 96 to 296 entries.
  • Performance: Improved accuracy scores; identified edge cases for refinement.

Iteration #3:

  • Model Upgrade: Transitioned to bert-large-cased-whole-word-masking from bert-base-uncased.
  • Advancements: Enhanced contextual sensitivity and accuracy.
  • Results: Demonstrated more nuanced understanding and sensitivity in predictions.

Observations

  • Each iteration has contributed to the model's evolving sophistication, leading to improved interpretive performance and accuracy.
  • Continuous evaluation, especially in complex or ambiguous cases, is pivotal for future enhancements.