buch of stuff

app.py

@@ -5,10 +5,10 @@ from sentence_transformers import SentenceTransformer
 import spaces
 
 @spaces.GPU
-def get_embeddings(query, task):
+def get_embeddings(queries, task):
     model = SentenceTransformer("Linq-AI-Research/Linq-Embed-Mistral", use_auth_token=os.getenv("HF_TOKEN"))
-    prompt = f"Instruct: {task}\nQuery: {query}" 
-    query_embeddings = model.encode([prompt])  
+    prompts = [f"Instruct: {task}\nQuery: {query}" for query in queries]
+    query_embeddings = model.encode(prompts)  
     return query_embeddings
 
 # Initialize a persistent Chroma client and retrieve collection
@@ -18,7 +18,7 @@ collection_en = client.get_collection(name="phil_en")
 authors_list_de = ["Ludwig Wittgenstein", "Sigmund Freud", "Marcus Aurelius", "Friedrich Nietzsche", "Epiktet", "Ernst Jünger", "Georg Christoph Lichtenberg", "Balthasar Gracian", "Hannah Arendt", "Erich Fromm", "Albert Camus"]
 authors_list_en = ["Friedrich Nietzsche", "Joscha Bach"]
 
-def query_chroma(collection, embeddings, authors, num_results=10):
+def query_chroma(collection, embeddings, authors):
     try:
         where_filter = {"author": {"$in": authors}} if authors else {}
 
@@ -26,7 +26,7 @@ def query_chroma(collection, embeddings, authors, num_results=10):
 
         results = collection.query(
             query_embeddings=[embeddings_list],
-            n_results=num_results,
+            n_results=10,
             where=where_filter,
             include=["documents", "metadatas", "distances"]
         )
@@ -53,59 +53,41 @@ def query_chroma(collection, embeddings, authors, num_results=10):
     except Exception as e:
         return {"error": str(e)}
 
-
-# Main function
-def perform_query(query, authors, num_results, database):
-    task = "Given a question, retrieve passages that answer the question"
-    embeddings = get_embeddings(query, task)
-    collection = collection_de if database == "German" else collection_en
-    results = query_chroma(collection, embeddings, authors, num_results)
-
-    if "error" in results:
-        return [gr.update(visible=True, value=f"Error: {results['error']}") for _ in range(max_textboxes * 2)]
-
-    updates = []
-    for res in results:
-        markdown_content = f"**{res['author']}, {res['book']}**\n\n{res['text']}"
-        updates.append(gr.update(visible=True, value=markdown_content))
-
-    updates += [gr.update(visible=False)] * (max_textboxes - len(results))
-
-    return updates
-
 def update_authors(database):
     return gr.update(choices=authors_list_de if database == "German" else authors_list_en)
 
-# Gradio interface
-max_textboxes = 30
 
-with gr.Blocks(css=".custom-markdown { border: 1px solid #ccc; padding: 10px; border-radius: 5px; }") as demo:
-    gr.Markdown("Enter your query, filter authors (default is all), click **Search** to search. Click **Flag** if a result is relevant to the query and interesting to you.")
-    with gr.Row():
-        with gr.Column():
-            database_inp = gr.Dropdown(label="Database", choices=["English", "German"], value="German")
-            inp = gr.Textbox(label="query", placeholder="Enter question...")
-            author_inp = gr.Dropdown(label="authors", choices=authors_list_de, multiselect=True)
-            num_results_inp = gr.Number(label="number of results", value=10, step=1, minimum=1, maximum=max_textboxes)
-            btn = gr.Button("Search")
 
-    components = []
+with gr.Blocks() as demo:
+    gr.Markdown("Enter your query, filter authors (default is all), click **Search** to search.")
+    database_inp = gr.Dropdown(label="Database", choices=["English", "German"], value="German")
+    author_inp = gr.Dropdown(label="Authors", choices=authors_list_de, multiselect=True)
+    inp = gr.Textbox(label="Query", placeholder="Enter questions separated by semicolons...")
+    btn = gr.Button("Search")
+
+    def perform_query(queries, authors, database):
+        queries = queries.split(';')
+        task = "Given a question, retrieve passages that answer the question"
+        embeddings = get_embeddings(queries, task)
+        collection = collection_de if database == "German" else collection_en
 
-    for _ in range(max_textboxes):
-        with gr.Column() as col:
-            text_out = gr.Markdown(visible=False, elem_classes="custom-markdown")
-            components.append(text_out)
+        results = [query_chroma(collection, embedding, authors) for embedding in embeddings]
+
+        for query, result in zip(queries, results):
+            with gr.Accordion(query):
+                markdown_contents = "\n".join(f"**{res['author']}, {res['book']}**\n\n{res['text']}" for res in result)
+                gr.Markdown(value=markdown_contents)
 
     btn.click(
-        fn=perform_query, 
-        inputs=[inp, author_inp, num_results_inp, database_inp],
-        outputs=components
+        perform_query,
+        inputs=[inp, author_inp, database_inp],
+        outputs=[]
     )
-    
+
     database_inp.change(
-    fn=update_authors,
-    inputs=database_inp,
-    outputs=author_inp
+        fn=lambda database: update_authors(database),
+        inputs=[database_inp],
+        outputs=[author_inp]
     )
 
-demo.launch()
+demo.launch()