Ankh Base ConvBERT Q8

This model is a downstream Q8 secondary structure prediction head trained on frozen Ankh Base embeddings.

Architecture

  • Frozen backbone: Ankh Base
  • Downstream head: ConvBERTForMultiClassClassification
  • Input dim: 768
  • Hidden dim: 384
  • Attention heads: 4
  • Kernel size: 7
  • Dropout: 0.2
  • Number of labels: 8

Training Status

  • Last completed epoch: 5
  • Best validation Q8 accuracy: 0.39778582010894004

Important Note

This repository contains only the downstream head. For inference, load Ankh Base first, extract embeddings, then pass the embeddings to this head.

Downloads last month
11
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for joanitolopo/ankh-base-convbert-q8

Finetuned
(3)
this model

Dataset used to train joanitolopo/ankh-base-convbert-q8

Collection including joanitolopo/ankh-base-convbert-q8