TAUR-dev/rankalign-v6-gemma-2-2b-d0.15-e2-hc-b2d-dbl-all-ln-vlo-fsx

Hugging Face
TEXT GENERATIONConcurrency Cost:1Model Size:2.6BQuant:BF16Ctx Length:8kPublished:Apr 6, 2026Architecture:Transformer Warm

TAUR-dev/rankalign-v6-gemma-2-2b-d0.15-e2-hc-b2d-dbl-all-ln-vlo-fsx is a 2.6 billion parameter language model fine-tuned from Google's Gemma-2-2b base model. This model is specifically optimized for hypernym prediction tasks, utilizing a rankalign training approach with length normalization and validator log-odds. It is designed to excel in identifying hierarchical relationships between concepts, making it suitable for semantic understanding and knowledge graph applications. The model's training focused on a 'hypernym-concat-bananas-to-dogs-double-all' task, indicating a specialized focus on specific semantic categories.

Loading preview...

Model Overview

This model, TAUR-dev/rankalign-v6-gemma-2-2b-d0.15-e2-hc-b2d-dbl-all-ln-vlo-fsx, is a specialized fine-tuned checkpoint derived from the Google Gemma-2-2b base model. It is part of the rankalign project, which focuses on improving the alignment and ranking of language model outputs.

Key Training Details

  • Base Model: google/gemma-2-2b
  • Parameter Count: 2.6 billion
  • Context Length: 8192 tokens
  • Version: v6 of the rankalign project.
  • Task: Specifically fine-tuned for hypernym-concat-bananas-to-dogs-double-all tasks, indicating a focus on identifying hypernym relationships across a defined set of concepts.
  • Training Epochs: 2
  • Delta: 0.15
  • Length Normalization: Enabled during training.
  • Validator Log-Odds: Utilizes validator log-odds for improved performance in its specialized task.

Use Cases

This model is particularly well-suited for:

  • Hypernym Prediction: Identifying broader categories or superordinates for given concepts.
  • Semantic Relationship Extraction: Applications requiring the understanding of hierarchical semantic links.
  • Knowledge Graph Construction: Assisting in the automated building or enrichment of knowledge graphs by identifying 'is-a' relationships.
  • Specialized NLP Research: Researchers working on fine-grained semantic understanding and ranking tasks, especially within the domain of hypernymy.

Finetunes
1 models