feat: vdb model

.gitignore

@@ -1 +1,2 @@
 env/
+data/

app.py

@@ -1,20 +1,39 @@
 import gradio as gr
+import json
+import faiss_vdb as vdb
 
-def my_inference_function(name):
-  return "Hello " + name + "!"
+def upsert_role(roleId, data):
+    """
+    body: { meta, content }
+    """
+    text = json.dumps(data['content'])
+    meta = data['meta']
+    meta['id'] = roleId
+    vdb.upsert(text, meta)
+
+    return 'true'
 
-gradio_interface = gr.Interface(
-  fn = my_inference_function,
-  inputs = "text",
-  outputs = "text"
-)
-def image_classifier(inp):
-    return {'cat': 0.3, 'dog': 0.7}
 
-demo0 = gr.Interface(fn=image_classifier, inputs="image", outputs="label")
+i_upsert = gr.Interface(fn=upsert_role, inputs=["text", "json"],
+                       outputs="text")
 # gradio_interface.launch()
 
-demo = gr.TabbedInterface([gradio_interface, demo0], ["gradio_interface", "demo"])
+
+def search_roles(content, size):
+    if size is None: size = 4
+    text = json.dumps(content)
+    docs = vdb.search(text, size)
+    return docs
+
+
+i_search = gr.Interface(
+    fn=search_roles,
+    inputs=["json", "number"],
+    outputs="json"
+)
+
+demo = gr.TabbedInterface([i_upsert, i_search], [
+                          "upsert role", "search roles"])
 
 if __name__ == "__main__":
-    demo.launch()
+    demo.launch()

faiss_vdb.py

@@ -0,0 +1,28 @@
+from langchain.vectorstores import FAISS
+from langchain.embeddings import HuggingFaceEmbeddings
+import os
+"""
+for development only.
+"""
+
+db_path = './data'
+
+embeddings = HuggingFaceEmbeddings(model_name='all-mpnet-base-v2')
+db = FAISS.load_local(db_path, embeddings) if os.path.exists(
+    db_path) else FAISS.from_texts([''], embeddings)
+
+
+def upsert(text, meta):
+    """
+    Args:
+            text: string
+            meta: dict, must contain id.
+    """
+    db.aadd_texts([text], metadatas=[meta], ids=[meta['id']])
+    db.save_local(db_path)
+
+
+def search(text, size=4):
+    docs = db.similarity_search(text, size)
+    data = [doc.metadata for doc in docs if ('id' in doc.metadata)]
+    return data