run several at once

app.py

@@ -11,7 +11,15 @@ from jinja2 import Environment, FileSystemLoader
 
 from backend.query_llm import generate_hf, generate_openai
 from backend.semantic_search import retrieve
+import itertools
 
+emb_models = ["bge", "minilm"]
+splitters = ['ct', 'rct', 'nltk']
+chunk_sizes = ["500", "2000"]
+sub_vectors = ["8", "16", "32"]
+
+# Create all combinations of the provided arrays
+combinations = itertools.product(emb_models, splitters, chunk_sizes, sub_vectors)
 
 TOP_K = int(os.getenv("TOP_K", 4))
 
@@ -54,17 +62,31 @@ def bot(history, api_kind,
     prompt = template.render(documents=documents, query=query)
     prompt_html = template_html.render(documents=documents, query=query)
 
+
     if api_kind == "HuggingFace":
          generate_fn = generate_hf
     elif api_kind == "OpenAI":
          generate_fn = generate_openai
     else:
          raise gr.Error(f"API {api_kind} is not supported")
+    
 
     history[-1][1] = ""
-    for character in generate_fn(prompt, history[:-1]):
-        history[-1][1] = character
-        yield history, prompt_html
+    # for character in generate_fn(prompt, history[:-1]):
+    #     history[-1][1] = character
+    #     yield history, prompt_html
+    
+    for model_name, doc, size, sub_vector in combinations:
+        documents_i = retrieve(query, int(num_docs), model_name, sub_vector, size, doc)
+        prompt_i = template.render(documents=documents_i, query=query)
+        prompt_html = template_html.render(documents=documents, query=query)
+        
+        hist_chunk = ""
+        prev_hist = history[-1][1] + f"\nmodel {model_name}, splitter {doc}, size {size}, sub vector {sub_vector}\n"
+        for character in generate_fn(prompt_i, history[:-1]):
+            hist_chunk = character
+            history[-1][1] = prev_hist + hist_chunk
+            yield history, prompt_html
 
 
 with gr.Blocks() as demo:
@@ -90,12 +112,12 @@ with gr.Blocks() as demo:
 
     with gr.Row():
         num_docs = gr.Slider(1, 20, label="number of docs", step=1, value=4)
-        model_kind = gr.Radio(choices=["bge", "minilm"], value="bge", label="embedding model")
-        sub_vector_size = gr.Radio(choices=["8", "16", "32"], value="32", label="sub-vector size")
+        model_kind = gr.Radio(choices=emb_models, value="bge", label="embedding model")
+        sub_vector_size = gr.Radio(choices=sub_vectors, value="32", label="sub-vector size")
     with gr.Row():
         api_kind = gr.Radio(choices=["HuggingFace", "OpenAI"], value="HuggingFace", label="Chat model engine")
-        chunk_size = gr.Radio(choices=["500", "2000"], value="2000", label="chunk size")
-        splitter_type = gr.Radio(choices=["ct", "rct","nltk"], value="nltk", label="splitter")
+        chunk_size = gr.Radio(choices=chunk_sizes, value="2000", label="chunk size")
+        splitter_type = gr.Radio(choices=splitters, value="nltk", label="splitter")
 
     prompt_html = gr.HTML()
     # Turn off interactivity while generating if you click