src/streamlit_app.py

import streamlit as st
from sentence_transformers import SentenceTransformer
from sentence_transformers.quantization import quantize_embeddings
import numpy as np

st.set_page_config(page_title="MXBAI Embed Demo", layout="centered")
st.title("MRL + Binary Embedding Demo by DEJAN")

st.markdown(
    """
Enter any sentence (or multiple sentences separated by newline).  
Use the checkboxes on the left sidebar to toggle:

- **MRL** (truncate to 256 dimensions)  
- **Binary** (apply binary quantization)
"""
)

# Sidebar controls
st.sidebar.header("Options")
use_mrl = st.sidebar.checkbox("Enable MRL", value=False)
use_binary = st.sidebar.checkbox("Enable Binary", value=False)

# Cache the model loading so Streamlit doesn't reload on every interaction unnecessarily.
@st.cache_resource(show_spinner=False)
def load_model(mrl: bool) -> SentenceTransformer:
    """
    Load the sentence-transformers model.
    If mrl=True, load with truncate_dim=256. Otherwise, load default (512 dims).
    """
    if mrl:
        # Matryoshka Representation: cut down to 256 dims
        return SentenceTransformer("mixedbread-ai/mxbai-embed-large-v1", truncate_dim=256)
    else:
        # Full 512-dimension model
        return SentenceTransformer("mixedbread-ai/mxbai-embed-large-v1")

model = load_model(use_mrl)

# Text input
input_text = st.text_area(
    "Enter text here:", 
    placeholder="Type one or more sentences..."
)

if not input_text.strip():
    st.info("⏳ Please enter some text above to see its embedding.")
    st.stop()

if st.button("Compute Embedding"):
    # Split input into lines (one sentence per line if multiple)
    sentences = [line.strip() for line in input_text.split("\n") if line.strip()]
    if len(sentences) == 0:
        st.error("❗ Please enter at least one non-empty sentence.")
        st.stop()

    embeddings = model.encode(sentences)  # shape: (num_sentences, dim)

    if use_binary:
        # quantize_embeddings expects a 2D numpy array
        embeddings = quantize_embeddings(np.vstack(embeddings), precision="ubinary")
        st.subheader("Binary-Quantized Embeddings (dtype=bool)")
        st.write(embeddings)
    else:
        st.subheader(f"Floating‐Point Embeddings (shape = {embeddings.shape})")
        for idx, emb in enumerate(embeddings):
            st.markdown(f"**Sentence #{idx+1}:**")
            st.write(emb.tolist())

    st.success("✅ Done!")