DeBERTa-v3-base

This repository contains the official microsoft/deberta-v3-base model in safetensors format together with the complete tokenizer files for maximum compatibility.

Original Model

microsoft/deberta-v3-base

Files Included

• model.safetensors — model weights
• Full tokenizer (tokenizer.json, spm.model, tokenizer_config.json, etc.)
• config.json

Usage

from transformers import AutoModel, AutoTokenizer

model = AutoModel.from_pretrained("LiefOlsonMd/deberta-v3-base")
tokenizer = AutoTokenizer.from_pretrained("LiefOlsonMd/deberta-v3-base")

Model Details

• Architecture: DeBERTa-v3-base
• Backbone parameters: 86M
• Embedding parameters: ~98M (128k SentencePiece vocabulary)
• Total parameters: ≈ 184M
• Weights: model.safetensors (~371 MB)

License

MIT License (same as the original model)