Updated UI and Ranker Agent

Agents/rankerAgent.py

@@ -2,84 +2,86 @@ import json
 import os
 from together import Together
 
-def rerank_best_answer(json_files, config_file='config.json', model="meta-llama/Llama-3-8b-chat-hf"):
-    # Load API key from configuration file
+# def rerank_best_answer(json_files, config_file='config.json', model="meta-llama/Meta-Llama-3.1-8B-Instruct-Turbo"):
+#     # Load API key from configuration file
 
 
-    together_ai_key = os.getenv("TOGETHER_AI")
-    if not together_ai_key:
-        raise ValueError("TOGETHER_AI environment variable not found. Please set it before running the script.")
+#     together_ai_key = os.getenv("TOGETHER_AI")
+#     if not together_ai_key:
+#         raise ValueError("TOGETHER_AI environment variable not found. Please set it before running the script.")
 
-    # Initialize Together client
-    client = Together(api_key=together_ai_key)
+#     # Initialize Together client
+#     client = Together(api_key=together_ai_key)
 
-    # Combine all JSON files into a single structure
-    combined_prompts = {}
-    for json_file in json_files:
-        with open(json_file, 'r') as file:
-            data = json.load(file)
+#     # Combine all JSON files into a single structure
+#     combined_prompts = {}
+#     for json_file in json_files:
+#         with open(json_file, 'r') as file:
+#             data = json.load(file)
         
-        # Format the input for the prompt
-        for item in data:
-            query_id = item['query_id']
-            if query_id not in combined_prompts:
-                combined_prompts[query_id] = {
-                    "question": item['input'],
-                    "answers": {}
-                }
-            combined_prompts[query_id]["answers"][json_file] = item['response']
-
-    responses = []
-
-    for query_id, prompt in combined_prompts.items():
-        # Generate the prompt text
-        prompt_text = f"""Input JSON:
-{json.dumps(prompt, indent=4)}
-
-For the above question, identify which model gave the best response based on accuracy. Ensure the chosen response is an answer and not a follow-up question. Provide the output in the format:
-{{
-    "best_model": "<model_name>",
-    "best_answer": "<answer>"
-}}
-Just output this JSON and nothing else.
-"""
-
-        # Generate response from Together API
-        response = client.chat.completions.create(
-            model=model,
-            messages=[{"role": "user", "content": prompt_text}],
-        )
-        response_content = response.choices[0].message.content 
-        # print(response_content)
-
-        prompt_text_extract_bestModel = f"""Input JSON:
-{json.dumps(response_content, indent=4)}
-
-Just Output the best_model from above JSON and nothing else.
-"""
-        prompt_text_extract_bestAnswer = f"""Input JSON:
-{json.dumps(response_content, indent=4)}
-
-Just Output the best_answer from above JSON and nothing else.
-"""
-        response_bestModel = client.chat.completions.create(
-            model=model,
-            messages=[{"role": "user", "content": prompt_text_extract_bestModel}],
-        )
-        response_bestAnswer = client.chat.completions.create(
-            model=model,
-            messages=[{"role": "user", "content": prompt_text_extract_bestAnswer}],
-        )
+#         # Format the input for the prompt
+#         for item in data:
+#             query_id = item['query_id']
+#             if query_id not in combined_prompts:
+#                 combined_prompts[query_id] = {
+#                     "question": item['input'],
+#                     "answers": {}
+#                 }
+#             combined_prompts[query_id]["answers"][json_file] = item['response']
+
+#     responses = []
+
+#     for query_id, prompt in combined_prompts.items():
+#         # Generate the prompt text
+#         prompt_text = f"""Input JSON:
+# {json.dumps(prompt, indent=4)}
+
+# For the above question, identify which model gave the best response based on accuracy. Ensure the chosen response is an answer and not a follow-up question. Provide the output in the format:
+# {{
+#     "best_model": "<model_name>",
+#     "best_answer": "<answer>"
+# }}
+# Just output this JSON and nothing else.
+# """
+
+#         # Generate response from Together API
+#         response = client.chat.completions.create(
+#             model=model,
+#             messages=[{"role": "user", "content": prompt_text}],
+#         )
+#         response_content = response.choices[0].message.content 
+#         # print(response_content)
+
+#         prompt_text_extract_bestModel = f"""Content:
+# {response_content}
+
+# Whats the best_model from above?
+# """
+#         prompt_text_extract_bestAnswer = f"""Content:
+# {response_content}
+
+# Whats the best_answer from above?
+# """
+#         print(prompt_text_extract_bestModel)
+#         print(prompt_text_extract_bestAnswer)
+#         response_bestModel = client.chat.completions.create(
+#             model=model,
+#             messages=[{"role": "user", "content": prompt_text_extract_bestModel}],
+#         )
+#         response_bestAnswer = client.chat.completions.create(
+#             model=model,
+#             messages=[{"role": "user", "content": prompt_text_extract_bestAnswer}],
+#         )
         
-        # print({"query_id": query_id, "question": prompt["question"], "Ranker_Output": response.choices[0].message.content})
-        responses.append({"query_id": query_id, "question": prompt["question"], "best_model": response_bestModel.choices[0].message.content, "best_answer": response_bestAnswer.choices[0].message.content})
+#         # print({"query_id": query_id, "question": prompt["question"], "Ranker_Output": response.choices[0].message.content})
+#         responses.append({"query_id": query_id, "question": prompt["question"], "best_model": response_bestModel.choices[0].message.content, "best_answer": response_bestAnswer.choices[0].message.content})
 
-        print(response_bestModel.choices[0].message.content)
+#         print(response_bestModel.choices[0].message.content)
 
-    return responses
+#     return responses
 
 
-def rankerAgent(prompt, config_file='config.json', model="meta-llama/Meta-Llama-3.1-8B-Instruct-Turbo"):
+def rankerAgent(prompt, model="meta-llama/Meta-Llama-3.1-8B-Instruct-Turbo"):
     # Load API key from configuration file
     together_ai_key = os.getenv("TOGETHER_AI")
     if not together_ai_key:
@@ -92,7 +94,7 @@ def rankerAgent(prompt, config_file='config.json', model="meta-llama/Meta-Llama-
     prompt_text = f"""Input JSON:
 {json.dumps(prompt, indent=4)}
 
-For the above question, identify which model gave the best response based on accuracy. Ensure the chosen response is an answer and not a follow-up question. Provide the output in the format:
+For the above question, identify which model gave the best response based on accuracy. Ensure the chosen response is an answer and not a follow-up question. The best_answer should be from the best_model only, as given in the above content. Provide the output in the format:
 {{
     "best_model": "<model_name>",
     "best_answer": "<answer>"
@@ -100,6 +102,7 @@ For the above question, identify which model gave the best response based on acc
 Just output this JSON and nothing else.
 """
 
+
     # Generate response from Together API
     response = client.chat.completions.create(
         model=model,
@@ -108,15 +111,15 @@ Just output this JSON and nothing else.
     response_content = response.choices[0].message.content 
     # print(response_content)
 
-    prompt_text_extract_bestModel = f"""Input JSON:
-{json.dumps(response_content, indent=4)}
+    prompt_text_extract_bestModel = f"""Content:
+{response_content}
 
-Just Output the best_model from above JSON and nothing else.
+Whats the best_model from above?
 """
-    prompt_text_extract_bestAnswer = f"""Input JSON:
-{json.dumps(response_content, indent=4)}
+    prompt_text_extract_bestAnswer = f"""Content:
+{response_content}
 
-Just Output the best_answer from above JSON and nothing else.
+Whats the best_answer from above?
 """
     response_bestModel = client.chat.completions.create(
         model=model,

app.py

@@ -51,10 +51,18 @@ def process_query(query):
         tf_idf_ranking_modified, bm25_ranking_modified, open_source_ranking_modified
     )
 
-    agent1_context = wiki_data[0]
+    try:
+        agent1_context = wiki_data[0]
+    except:
+        agent1_context = "Can't find a Wiki article for this query."
+        
     agent2_context = article
 
-    boolean_context = miniWikiCollectionDict[boolean_ranking[0]]
+    try:
+        boolean_context = miniWikiCollectionDict[boolean_ranking[0]]
+    except:
+        boolean_context = "Can't find a matching document for this query."
+
     tf_idf_context = miniWikiCollectionDict[tf_idf_ranking[0]]
     bm25_context = miniWikiCollectionDict[str(bm25_ranking[0])]
     vision_context = miniWikiCollectionDict[vision_ranking[0]]
@@ -136,29 +144,85 @@ def process_query(query):
         zeroShot, "Zero-shot doesn't have a context."
     )
 
+# CSS Styling for the fancy effects
+css = """
+#fancy-column {
+    background: linear-gradient(135deg, #1a242f, #2b3a44);  /* Dark blue-gray gradient background */
+    padding: 20px;
+    border-radius: 15px;
+}
+
+#query-input, #submit-button, #best-model-output, #best-answer-output {
+    border-radius: 10px;  /* Rounded corners */
+    box-shadow: 0 4px 6px rgba(0, 0, 0, 0.3);  /* Darker shadow for better contrast */
+    background-color: #34495e;  /* Dark background for inputs */
+    color: #ecf0f1;  /* Light text for good readability */
+}
+
+#query-input:focus, #submit-button:focus, #best-model-output:focus, #best-answer-output:focus {
+    outline: none;
+    border: 2px solid #7f8c8d;  /* Subtle accent border on focus */
+}
+
+#submit-button {
+    background-color: #16a085;  /* Muted teal color for button */
+    color: #ecf0f1;  /* Light text for button */
+    font-weight: bold;
+    padding: 10px;
+}
+
+#submit-button:hover {
+    background-color: #1abc9c;  /* Slightly lighter teal on hover */
+}
+
+#best-model-output, #best-answer-output {
+    background-color: #2c3e50;  /* Darker background for output boxes */
+}
+
+#best-model-output label, #best-answer-output label, #query-input label {
+    color: #ecf0f1;  /* Light text for labels */
+}
+"""
+
+
+
 # Interface creation
 def create_interface():
     with gr.Blocks() as interface:
-        query_input = gr.Textbox(label="Enter your query")
-        best_model_output = gr.Textbox(label="Best Model", interactive=False)
-        best_answer_output = gr.Textbox(label="Best Answer", interactive=False)
-
-        def create_answer_row(label):
+        with gr.Column(elem_id="fancy-column", scale=3):  # Fancy column with extra styling
             with gr.Row():
-                answer_textbox = gr.Textbox(label=f"{label} Answer", interactive=False)
-                context_button = gr.Button(f"Show {label} Context")
-                context_textbox = gr.Textbox(label=f"{label} Context", visible=False)
-
-                # Event to show the context
-                context_button.click(
-                    fn=lambda x: gr.update(visible=True, value=x),
-                    inputs=None,
-                    outputs=context_textbox
-                )
+                query_input = gr.Textbox(label="Enter your query", scale=3, elem_id="query-input")
+                submit_button = gr.Button("Submit", scale=1, elem_id="submit-button")
+            
+            # Adjusting the spacing between the output fields
+            with gr.Row():
+                best_model_output = gr.Textbox(label="Best Model", interactive=False, scale=1.5, elem_id="best-model-output")
+                best_answer_output = gr.Textbox(label="Best Answer", interactive=False, scale=1.5, elem_id="best-answer-output")
+
+        with gr.Column():
+            # Function to create a row for answers and contexts
+            def create_answer_row(label):
+                if label == "Agent 1":
+                    label = "Wiki Search"
+                elif label == "Agent 2":
+                    label = "Llama Context Generation"
+                elif label == "Open Source Answer":
+                    label = 'MiniLM Text Embedding model'
+                elif label == "Open Source (Modified)":
+                    label = 'MiniLM Text Embedding model (Modified)'
+                elif label == "TF-IDF + BM25 + Open RRF":
+                    label = "RRF (TF-IDF + BM25 + MiniLM)"
+                elif label == "TF-IDF + BM25 + Open RRF (Modified)":
+                    label = "RRF (TF-IDF + BM25 + MiniLM) (Modified)"
+                elif label == "TF-IDF + BM25 + Open RRF (Combined)":
+                    label = "RRF (TF-IDF + BM25 + MiniLM) (Combined)"
+                with gr.Row():
+                    answer_textbox = gr.Textbox(label=f"{label} Answer", interactive=False, scale=1.2, elem_id="best-model-output")
+                    context_textbox = gr.Textbox(label=f"{label} Context", scale=1.8, elem_id="best-answer-output")
+                
                 return answer_textbox, context_textbox
 
         agent1_output, agent1_context_output = create_answer_row("Agent 1")
-
         agent2_output, agent2_context_output = create_answer_row("Agent 2")
         boolean_output, boolean_context_output = create_answer_row("Boolean")
         tf_idf_output, tf_idf_context_output = create_answer_row("TF-IDF")
@@ -170,7 +234,7 @@ def create_interface():
         tf_idf_mod_output, tf_idf_mod_context_output = create_answer_row("TF-IDF (Modified)")
         bm25_mod_output, bm25_mod_context_output = create_answer_row("BM25 (Modified)")
         vision_mod_output, vision_mod_context_output = create_answer_row("Vision (Modified)")
-        open_source_mod_output, open_source_context_output = create_answer_row("Open Source (Modified)")
+        open_source_mod_output, open_source_mod_context_output = create_answer_row("Open Source (Modified)")
 
         tf_idf_rrf_output, tf_idf_rrf_context_output = create_answer_row("TF-IDF + BM25 + Open RRF")
         tf_idf_rrf_mod_output, tf_idf_rrf_mod_context_output = create_answer_row("TF-IDF + BM25 + Open RRF (Modified)")
@@ -178,7 +242,7 @@ def create_interface():
 
         zero_shot_output, zero_shot_context_output = create_answer_row("Zero Shot")
 
-        gr.Button("Submit").click(
+        submit_button.click(
             fn=process_query,
             inputs=query_input,
             outputs=[
@@ -195,7 +259,7 @@ def create_interface():
                 tf_idf_mod_output, tf_idf_mod_context_output,
                 bm25_mod_output, bm25_mod_context_output,
                 vision_mod_output, vision_mod_context_output,
-                open_source_mod_output, open_source_context_output,
+                open_source_mod_output, open_source_mod_context_output,
                 tf_idf_rrf_output, tf_idf_rrf_context_output,
                 tf_idf_rrf_mod_output, tf_idf_rrf_mod_context_output,
                 tf_idf_rrf_combined_output, tf_idf_rrf_combined_context_output,
@@ -208,4 +272,5 @@ def create_interface():
 # Launch the interface
 if __name__ == "__main__":
     interface = create_interface()
+    interface.css = css
     interface.launch()