0neisall/tiny-embedding-test

Tiny embedding test model

Randomly initialized 1-layer BERT with mean pooling and L2 normalization. Outputs 16-dimensional embeddings. Maximum sequence length: 128 tokens.

For upload/loading tests only. Not trained; no useful semantic quality. The tiny vocabulary maps most words to [UNK].

from sentence_transformers import SentenceTransformer

# Replace with your uploaded repository ID.
model = SentenceTransformer("YOUR_USERNAME/tiny-embedding-test")
embeddings = model.encode(["hello world", "你好世界"])
print(embeddings.shape)  # (2, 16)