TAUR-dev/rankalign-v6-gemma-2-2b-d0.15-e2-hc-b2d-dbl-all-ln-p0-nv1-ng1-vlo-fsx

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

The TAUR-dev/rankalign-v6-gemma-2-2b-d0.15-e2-hc-b2d-dbl-all-ln-p0-nv1-ng1-vlo-fsx model is a 2.6 billion parameter language model fine-tuned from the Google Gemma-2-2b base model. Developed as part of the rankalign project, this checkpoint is specifically optimized for hypernym prediction tasks, utilizing a unique training configuration that includes length normalization and validator log-odds. Its primary strength lies in accurately identifying hierarchical relationships between concepts, making it suitable for semantic understanding and knowledge graph construction.

Loading preview...

Model Overview

This model, rankalign-v6-gemma-2-2b-d0.15-e2-hc-b2d-dbl-all-ln-p0-nv1-ng1-vlo-fsx, is a fine-tuned checkpoint derived from the google/gemma-2-2b base model. It is part of the rankalign project, which focuses on enhancing language models for specific linguistic tasks.

Key Training Details

This particular version (v6) underwent a specialized training regimen with the following notable characteristics:

  • Base Model: google/gemma-2-2b
  • Task: hypernym-concat-bananas-to-dogs-double-all, indicating a focus on hypernym prediction across a diverse set of concepts.
  • Epochs: Trained for 2 epochs.
  • Delta: A delta value of 0.15 was applied during training.
  • Length Normalization: Enabled (True) to potentially improve consistency in output length.
  • Validator Log-Odds: Utilizes validator log-odds (True) for its NLL validator weight, suggesting a specific approach to evaluating and guiding generation.
  • Preference Loss Weight: Set to 0, implying the training primarily focused on NLL (Negative Log Likelihood) rather than direct preference optimization.

Intended Use Cases

Given its specialized training on hypernym prediction, this model is particularly well-suited for:

  • Semantic Relationship Extraction: Identifying 'is-a' relationships between words or concepts.
  • Knowledge Graph Construction: Populating or validating hierarchical structures within knowledge bases.
  • Taxonomy Generation: Assisting in the creation and organization of classification systems.
  • Linguistic Research: Exploring and evaluating methods for semantic understanding and hierarchical concept recognition.

Finetunes
1 models