Initial commit

.gitattributes

@@ -3,6 +3,7 @@
 *.bin filter=lfs diff=lfs merge=lfs -text
 *.bin.* filter=lfs diff=lfs merge=lfs -text
 *.bz2 filter=lfs diff=lfs merge=lfs -text
+*.csv filter=lfs diff=lfs merge=lfs -text
 *.ftz filter=lfs diff=lfs merge=lfs -text
 *.gz filter=lfs diff=lfs merge=lfs -text
 *.h5 filter=lfs diff=lfs merge=lfs -text
@@ -10,6 +11,7 @@
 *.lfs.* filter=lfs diff=lfs merge=lfs -text
 *.model filter=lfs diff=lfs merge=lfs -text
 *.msgpack filter=lfs diff=lfs merge=lfs -text
+*.npy filter=lfs diff=lfs merge=lfs -text
 *.onnx filter=lfs diff=lfs merge=lfs -text
 *.ot filter=lfs diff=lfs merge=lfs -text
 *.parquet filter=lfs diff=lfs merge=lfs -text

README.md

@@ -1,13 +1,13 @@
 ---
-title: Semanticsearch
-emoji: 👁
+title: Semantic Search
+emoji: 📖
 colorFrom: purple
-colorTo: indigo
+colorTo: red
 sdk: streamlit
 sdk_version: 1.2.0
 app_file: app.py
 pinned: false
-license: cc-by-4.0
+license: cc-by-nc-4.0
 ---
 
 Check out the configuration reference at https://huggingface.co/docs/hub/spaces#reference

app.py

@@ -0,0 +1,147 @@
+import time
+import re
+import pandas as pd
+import numpy as np
+import torch
+import torch.nn.functional as F
+from transformers import AutoTokenizer, AutoModel
+from tokenizers import Tokenizer, AddedToken
+import streamlit as st
+from st_click_detector import click_detector
+
+DEVICE = "cpu"
+MODEL_OPTIONS = ["msmarco-distilbert-base-tas-b", "all-mpnet-base-v2"]
+DESCRIPTION = """
+# Semantic search
+
+**Enter your query and hit enter**
+
+Built with 🤗 Hugging Face's [transformers](https://huggingface.co/transformers/) library, [SentenceBert](https://www.sbert.net/) models, [Streamlit](https://streamlit.io/) and 44k movie descriptions from the Kaggle [Movies Dataset](https://www.kaggle.com/rounakbanik/the-movies-dataset)
+"""
+
+
+@st.cache(
+    show_spinner=False,
+    hash_funcs={
+        AutoModel: lambda _: None,
+        AutoTokenizer: lambda _: None,
+        dict: lambda _: None,
+    },
+)
+def load():
+    models, tokenizers, embeddings = [], [], []
+    for model_option in MODEL_OPTIONS:
+        tokenizers.append(
+            AutoTokenizer.from_pretrained(f"sentence-transformers/{model_option}")
+        )
+        models.append(
+            AutoModel.from_pretrained(f"sentence-transformers/{model_option}").to(
+                DEVICE
+            )
+        )
+    embeddings.append(np.load("embeddings.npy"))
+    embeddings.append(np.load("embeddings2.npy"))
+    df = pd.read_csv("movies.csv")
+    return tokenizers, models, embeddings, df
+
+
+tokenizers, models, embeddings, df = load()
+
+
+def pooling(model_output):
+    return model_output.last_hidden_state[:, 0]
+
+
+def compute_embeddings(texts):
+    encoded_input = tokenizers[0](
+        texts, padding=True, truncation=True, return_tensors="pt"
+    ).to(DEVICE)
+
+    with torch.no_grad():
+        model_output = models[0](**encoded_input, return_dict=True)
+
+    embeddings = pooling(model_output)
+
+    return embeddings.cpu().numpy()
+
+
+def pooling2(model_output, attention_mask):
+    token_embeddings = model_output[0]
+    input_mask_expanded = (
+        attention_mask.unsqueeze(-1).expand(token_embeddings.size()).float()
+    )
+    return torch.sum(token_embeddings * input_mask_expanded, 1) / torch.clamp(
+        input_mask_expanded.sum(1), min=1e-9
+    )
+
+
+def compute_embeddings2(list_of_strings):
+    encoded_input = tokenizers[1](
+        list_of_strings, padding=True, truncation=True, return_tensors="pt"
+    ).to(DEVICE)
+    with torch.no_grad():
+        model_output = models[1](**encoded_input)
+    sentence_embeddings = pooling2(model_output, encoded_input["attention_mask"])
+    return F.normalize(sentence_embeddings, p=2, dim=1).cpu().numpy()
+
+
+@st.cache(
+    show_spinner=False,
+    hash_funcs={Tokenizer: lambda _: None, AddedToken: lambda _: None},
+)
+def semantic_search(query, model_id):
+    start = time.time()
+    if len(query.strip()) == 0:
+        return ""
+    if "[Similar:" not in query:
+        if model_id == 0:
+            query_embedding = compute_embeddings([query])
+        else:
+            query_embedding = compute_embeddings2([query])
+    else:
+        match = re.match(r"\[Similar:(\d{1,5}).*", query)
+        if match:
+            idx = int(match.groups()[0])
+            query_embedding = embeddings[model_id][idx : idx + 1, :]
+            if query_embedding.shape[0] == 0:
+                return ""
+        else:
+            return ""
+    indices = np.argsort(embeddings[model_id] @ np.transpose(query_embedding)[:, 0])[
+        -1:-11:-1
+    ]
+    if len(indices) == 0:
+        return ""
+    result = "<ol>"
+    for i in indices:
+        result += f"<li style='padding-top: 10px'><b>{df.iloc[i].title}</b> ({df.iloc[i].release_date}). {df.iloc[i].overview} "
+        result += f"<a id='{i}' href='#'>Similar movies</a></li>"
+    delay = "%.3f" % (time.time() - start)
+    return f"<p><i>Computation time: {delay} seconds</i></p>{result}</ol>"
+
+
+st.sidebar.markdown(DESCRIPTION)
+
+model_choice = st.sidebar.selectbox("Similarity model", options=MODEL_OPTIONS)
+model_id = 0 if model_choice == MODEL_OPTIONS[0] else 1
+
+if "query" in st.session_state:
+    query = st.text_input("", value=st.session_state["query"])
+else:
+    query = st.text_input("", value="time travel")
+
+clicked = click_detector(semantic_search(query, model_id))
+
+if clicked != "":
+    st.markdown(clicked)
+    change_query = False
+    if "last_clicked" not in st.session_state:
+        st.session_state["last_clicked"] = clicked
+        change_query = True
+    else:
+        if clicked != st.session_state["last_clicked"]:
+            st.session_state["last_clicked"] = clicked
+            change_query = True
+    if change_query:
+        st.session_state["query"] = f"[Similar:{clicked}] {df.iloc[int(clicked)].title}"
+        st.experimental_rerun()

embeddings.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:64495712bf1903dd04604cd5641f5b521912d8938339e9e9e3071dad8952b34a
+size 134876288

embeddings2.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:136aa7ffd5630d19dc88f1e779dbeb04011ef918ac3fba2148a8f5d58303d736
+size 134876288

movies.csv

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1da4fb07829b3f57bce3fa663641c50b3d3e65cdf949f6e6f340960a5acc1005
+size 16293996

requirements.txt

@@ -0,0 +1,5 @@
+torch
+transformers
+numpy
+pandas
+st-click-detector