mjaliz/siglip-text-encoder

Text Encoder extracted from mjaliz/vision-text-dual-encoder-v1

This is the text encoder component extracted from the VisionTextDualEncoder model mjaliz/vision-text-dual-encoder-v1.

Model Details

• Model type: XLMRobertaModel
• Source model: mjaliz/vision-text-dual-encoder-v1
• Includes projection: False

Usage

from transformers import AutoModel, AutoTokenizer

# Load text encoder
model = AutoModel.from_pretrained("mjaliz/siglip-text-encoder")
tokenizer = AutoTokenizer.from_pretrained("mjaliz/siglip-text-encoder")

# Encode text
texts = ["Hello world", "How are you?"]
inputs = tokenizer(texts, padding=True, truncation=True, return_tensors="pt")
outputs = model(**inputs)

# Get embeddings (pooler output or mean of last hidden state)
if hasattr(outputs, "pooler_output") and outputs.pooler_output is not None:
    embeddings = outputs.pooler_output
else:
    embeddings = outputs.last_hidden_state.mean(dim=1)

print(embeddings.shape)

Citation

If you use this model, please cite the original dual encoder model.