zip2zip: Inference-Time Adaptive Vocabularies for Language Models via Token Compression
Paper • 2506.01084 • Published • 7
How to use nathanrchn/zip2zip-test with Transformers:
# Load model directly
from transformers import AutoModel
model = AutoModel.from_pretrained("nathanrchn/zip2zip-test", dtype="auto")Base model
microsoft/Phi-3.5-mini-instruct