init

.gitattributes

@@ -1,3 +1,4 @@
+*.index
 *.7z filter=lfs diff=lfs merge=lfs -text
 *.arrow filter=lfs diff=lfs merge=lfs -text
 *.bin filter=lfs diff=lfs merge=lfs -text

.gitignore

@@ -0,0 +1 @@
+*.index

app.py

@@ -0,0 +1,93 @@
+import faiss
+import gradio as gr
+import numpy as np
+import pandas as pd
+from sentence_transformers import SentenceTransformer
+import zipfile
+
+index_file = "faiss.index.zip"
+
+with zipfile.ZipFile(index_file, "r") as z:
+    z.extractall()
+
+pr_number = 14
+model = SentenceTransformer(
+    "intfloat/multilingual-e5-small",
+    revision=f"refs/pr/{pr_number}",
+    backend="openvino",
+)
+
+class FaissIndex:
+    def __init__(
+        self,
+        model: SentenceTransformer,
+        data_path: str = "faiss.lookup.csv",
+        index_path="faiss.index",
+    ):
+        self.model = model
+        self.df = pd.read_csv(data_path)
+        self.index = faiss.read_index(index_path)
+
+    def search(self, query, k=5):
+        query = np.array(query).astype("float32")
+        distances, indices = self.index.search(query, k)
+        return distances, indices
+
+    def extract_docs(self, indices, k):
+        indices = list(indices[0])
+        lookup = self.df.iloc[indices]
+        questions = lookup["question"].values
+        answers = lookup["answer"].values
+
+        pairs = list(zip(questions, answers))
+        # ensure we only have unique answers. The questions can be duplicates
+        filtered_pairs = []
+        seen = set()
+        for pair in pairs:
+            if pair[1] not in seen:
+                seen.add(pair[1])
+                filtered_pairs.append(pair)
+
+        # format pairs as: f"{answer}\n{kilde: {question}}"
+        formatted_pairs = []
+        for pair in filtered_pairs:
+            formatted_pairs.append(f"{pair[1]} (kilde: {pair[0]})")
+        return formatted_pairs
+
+    def search(self, query: str, k: int = 5):
+        query = "query: " + query
+        enc = self.model.encode([query])
+        emb = np.array(enc).astype("float32").reshape(1, -1)
+        _, indices = self.index.search(emb, k)
+        return self.extract_docs(indices, k)
+
+
+index = FaissIndex(model)
+
+
+def query_faiss_index(søketekst):
+    """
+    Queries the FAISS index with the provided search text and returns the top 5 results.
+    Args:
+        søketekst (str): The search text to query the FAISS index.
+    Returns:
+        str: A string containing the top 5 search results, separated by double newlines.
+    """
+
+    results = index.search(søketekst, k=5)
+    return "\n\n".join(results)
+
+
+# Create the Gradio interface
+iface = gr.Interface(
+    fn=query_faiss_index,
+    inputs=gr.Textbox(lines=2, placeholder="Søk etter info i SIKT", interactive=True),
+    outputs=gr.Textbox(label="Søkeresultater", type="text", lines=15),
+    title="SIKT-FAQ",
+    description="Semantisk søk i SIKT med Openvino.",
+    live=True
+)
+
+# Launch the Gradio app
+if __name__ == "__main__":
+    iface.launch()

faiss.index.zip

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a85d3ff0d5a335201c3b92fb3466df469c4f3fb9569a087166d393ca7801527e
+size 6895269

prep.py

@@ -0,0 +1,5 @@
+import zipfile
+
+index_file = "faiss.index"
+with zipfile.ZipFile(index_file + ".zip", "w") as z:
+    z.write(index_file)