TAUR-dev/rankalign-v6-gemma-2-2b-d0.15-e2-hc-b2d-dbl-all-tco-p0-nv1-ng1-vlo-fsx
The TAUR-dev/rankalign-v6-gemma-2-2b-d0.15-e2-hc-b2d-dbl-all-tco-p0-nv1-ng1-vlo-fsx model is a 2.6 billion parameter language model fine-tuned from Google's Gemma-2-2B base model. It is part of the rankalign project, specifically trained for hypernym-concat-bananas-to-dogs-double-all tasks. This model is optimized for tasks involving the identification and generation of hypernym relationships, utilizing a delta of 0.15 and online typicality correction. Its primary strength lies in its specialized training for semantic hierarchy understanding, particularly for hypernyms.
Loading preview...
Model Overview
This model, TAUR-dev/rankalign-v6-gemma-2-2b-d0.15-e2-hc-b2d-dbl-all-tco-p0-nv1-ng1-vlo-fsx, is a fine-tuned checkpoint derived from the Google Gemma-2-2B base model. It is developed within the rankalign project, focusing on specific linguistic tasks.
Key Training Details
- Base Model:
google/gemma-2-2b - Version: v6 of the rankalign project's fine-tuning.
- Task: Specialized in
hypernym-concat-bananas-to-dogs-double-all, indicating a focus on hypernym generation and validation. - Epochs: Trained for 2 epochs.
- Delta: A delta value of 0.15 was applied during training.
- Correction: Utilizes online typicality correction.
- Loss Weights: Features a preference loss weight of 0, NLL validator weight of 1, and NLL generator weight of 1.
- Validation: Employs validator log-odds and forces same-x conditions.
Intended Use Cases
This model is particularly suited for research and applications requiring precise understanding and generation of hypernym relationships. Its specific training on hypernym tasks suggests strong performance in semantic hierarchy extraction and related natural language understanding challenges. Developers can use the provided evaluation scripts to assess its performance across various hypernym-related datasets.