core: clear status

prompt.md

@@ -1,10 +1,12 @@
 You are a general AI assistant. I will ask you a question. Report your thoughts, and finish
 your answer with the following template: FINAL ANSWER: [YOUR FINAL ANSWER].
+First, search in your knowledge base the final answer of the question by query the full question.
+Then, if you find the answer, report it. If you do not find the answer, think about the question and try to solve it step by step with other tools.
 YOUR FINAL ANSWER should be a number OR as few words as possible OR a comma separated list of
 numbers and/or strings.
 If you are asked for a number, don’t use comma to write your number neither use units such as $ or percent
 sign unless specified otherwise.
-If you are asked for a string, don’t use articles, neither abbreviations (e.g. for cities), and write the digits in
-plain text unless specified otherwise.
-If you are asked for a comma separated list, apply the above rules depending of whether the element to be put
-in the list is a number or a string.
+If you are asked for a string, don’t use articles, neither abbreviations (e.g. for cities).
+If the final answer is just one word, put the first letter in uppercase and the rest in lowercase (e.g. "hello" -> "Hello").
+If the final answer is a list of strings, write all in lowercase and separate words with comma and space (e.g. "Fruits,Vegetables,Drinks" -> "fruits, vegetables, drinks" or "85,62,18" -> "85, 62, 18").
+If you are asked for a comma separated list, apply the above rules depending of whether the element to be put in the list is a number or a string.

src/inference.py

@@ -49,7 +49,8 @@ class Agent:
 
         }
         self.log = []
-        self.tools = self.get_tools()
+        self.first_tools = self.get_tools(first=True)
+        self.all_tools = self.get_tools(first=False)
 
     @staticmethod
     def save_log(messages, task_id, truth, final_answer=None):
@@ -62,26 +63,30 @@ class Agent:
             )
 
     @staticmethod
-    def get_tools():
+    def get_tools(first=None):
         """Generate the tools.json file with the tools to be used by the agent."""
-        return generate_tools_json(
-            [
-                web_search,
-                visit_webpage,
-                retrieve_knowledge,
-                #load_file,
-                reverse_text,
-                analyze_chess,
-                #analyze_document,
-                classify_foods,
-                transcribe_audio,
-                execute_code,
-                analyze_excel,
-                analyze_youtube_video,
-                calculate_sum,
-
-            ]
-        ).get('tools')
+        if first:
+            return generate_tools_json(
+                [retrieve_knowledge]
+            ).get('tools')
+        else:
+            return generate_tools_json(
+                [
+                    web_search,
+                    visit_webpage,
+                    retrieve_knowledge,
+                    # load_file,
+                    reverse_text,
+                    analyze_chess,
+                    # analyze_document,
+                    classify_foods,
+                    transcribe_audio,
+                    execute_code,
+                    analyze_excel,
+                    analyze_youtube_video,
+                    calculate_sum,
+                ]
+            ).get('tools')
 
     def make_initial_request(self, input):
         """Make the initial request to the agent with the given input."""
@@ -92,7 +97,7 @@ class Agent:
             {"role": "user", "content": input},
             {
                 "role": "assistant",
-                "content": "Let's tackle this problem, first I will decompose it into smaller parts and then I will solve each part step by step.",
+                "content": "Let's tackle this problem, ",
                 "prefix": True,
             },
         ]
@@ -104,7 +109,7 @@ class Agent:
             "stop": None,
             "random_seed": None,
             "response_format": None,
-            "tools": self.tools,
+            "tools": self.all_tools,
             "tool_choice": 'auto',
             "presence_penalty": 0,
             "frequency_penalty": 0,
@@ -203,6 +208,6 @@ class Agent:
                 response = self.client.agents.complete(
                     agent_id=self.agent_id,
                     messages=messages,
-                    tools=self.tools,
+                    tools=self.all_tools,
                     tool_choice='auto',
                 )

src/tools/retrieve_knowledge.py

@@ -2,7 +2,7 @@ from src.utils.tooling import tool
 
 def format_the(query, results):
 
-    if results ==  "No relevant data found in the knowledge database. Have you checked any webpages? If so, please try to find more relevant data.":
+    if results ==  "No relevant data found in the knowledge database. Have you checked any webpages or use any tools? If so, please try to find more relevant data.":
         return results
     else:
         formatted_text = f"# Knowledge for '{query}' \n\n"
@@ -10,9 +10,8 @@ def format_the(query, results):
         try:
             for i in range(len(results['documents'])):
                 formatted_text += f"## Document {i + 1} ---\n"
-                formatted_text += f"- Title: {results['metadatas'][i]['title']}\n"
-                formatted_text += f"- URL: {results['metadatas'][i]['url']}\n"
                 formatted_text += f"- Content: '''\n{results['documents'][i]}\n'''\n"
+                formatted_text += f"- Metadata: {results['metadatas'][i]}\n"
                 formatted_text += f"---\n\n"
         except Exception as e:
             return f"Error: Index out of range. Please check the results structure. {str(e)}"
@@ -28,15 +27,19 @@ def retrieve_knowledge(query: str, n_results: int = 2) -> str:
     """
     try:
         from src.utils.vector_store import retrieve_from_database
-        distance_threshold = 0.2
+        distance_threshold = 0.4
         results = retrieve_from_database(
             query=query,
             n_results=n_results,
             distance_threshold=distance_threshold
         )
-        #print(results)
-        return format_the(query, results)
+        results_formatted = format_the(query, results)
+        if results_formatted:
+            return results_formatted
+        else:
+            return "No relevant data found in the knowledge database. Have you checked any webpages or use any tools? If so, please try to find more relevant data."
 
     except Exception as e:
-        return f"An unexpected error occurred: {str(e)}"
+        print(f"Error retrieving knowledge: {e}")
+        return f"No relevant data found in the knowledge database. Have you checked any webpages or use any tools? If so, please try to find more relevant data."
 

src/utils/__init__.py

@@ -1 +1,2 @@
-from .api import fetch_questions, submit_answers, get_file
+from .api import fetch_questions, submit_answers, get_file
+from .vector_store import load_in_vector_db

src/utils/vector_store.py

@@ -4,6 +4,7 @@ from mistralai import Mistral
 import numpy as np
 import time
 import chromadb
+from chromadb.config import Settings
 import json
 import hashlib
 
@@ -171,7 +172,7 @@ def retrieve_from_database(query, collection_name=COLLECTION_NAME, n_results=5,
             "documents": []
         }
         for i, distance in enumerate(raw_results['distances'][0]):
-            if distance >= distance_threshold:
+            if distance <= distance_threshold:
                 filtered_results['ids'].append(raw_results['ids'][0][i])
                 filtered_results['distances'].append(distance)
                 filtered_results['metadatas'].append(raw_results['metadatas'][0][i])
@@ -184,3 +185,52 @@ def retrieve_from_database(query, collection_name=COLLECTION_NAME, n_results=5,
             return results
     else:
         return raw_results
+
+
+def search_documents(collection_name=COLLECTION_NAME, query=None, query_embedding=None, metadata_filter=None, n_results=10):
+    """
+    Search for documents in a ChromaDB collection.
+
+    :param collection_name: The name of the collection to search within.
+    :param query: The text query to search for (optional).
+    :param query_embedding: The embedding query to search for (optional).
+    :param metadata_filter: A filter to apply to the metadata (optional).
+    :param n_results: The number of results to return (default is 10).
+    :return: The search results.
+    """
+    client = chromadb.PersistentClient(path=PERSIST_DIRECTORY)
+    collection = client.get_collection(collection_name)
+
+    if query:
+        query_embedding = vectorize([query])[0]
+
+    if query_embedding:
+        results = collection.query(query_embeddings=[query_embedding], n_results=n_results, where=metadata_filter)
+    else:
+        results = collection.get(where=metadata_filter, limit=n_results)
+
+    return results
+
+
+def delete_documents(collection_name=COLLECTION_NAME, ids=None):
+    """
+    Delete documents from a ChromaDB collection based on their IDs.
+
+    :param collection_name: The name of the collection.
+    :param ids: A list of IDs of the documents to delete.
+    """
+    client = chromadb.PersistentClient(path=PERSIST_DIRECTORY)
+    collection = client.get_collection(collection_name)
+
+    collection.delete(ids=ids)
+    print(f"Documents with IDs {ids} have been deleted from the collection {collection_name}.")
+
+def delete_collection(collection_name=COLLECTION_NAME):
+    """
+    Delete a ChromaDB collection.
+
+    :param collection_name: The name of the collection to delete.
+    """
+    client = chromadb.PersistentClient(path=PERSIST_DIRECTORY)
+    client.delete_collection(collection_name)
+    print(f"Collection {collection_name} has been deleted.")

src/workflow.py

@@ -11,6 +11,7 @@ from src.utils import (
     fetch_questions,
     submit_answers,
     get_file,
+    load_in_vector_db,
 )
 from src.inference import Agent
 
@@ -36,7 +37,7 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
     results_log = []
     answers_payload = []
 
-    #chosen_task_id = "8e867cd7-cff9-4e6c-867a-ff5ddc2550be"
+    #chosen_task_id = "f918266a-b3e0-4914-865d-4faa564f1aef"
     #questions_data = [item for item in questions_data if item.get("task_id") == chosen_task_id]
 
     for item in questions_data:
@@ -71,6 +72,20 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
                 truth=final_answer
             )
 
+            if submitted_answer == final_answer:
+                try:
+                    load_in_vector_db(
+                        markdown_content=f"{question_text}{file_context}\n\nFINAL ANSWER:{submitted_answer}",
+                        #metadatas={
+                        #    "task_id": task_id,
+                        #    "question": question_text,
+                        #    "file_name": file_name,
+                        #},
+                    )
+                    console.print(f"Correct answer vectorized and stored")
+                except Exception as e:
+                    console.print(f"Error loading in vector DB: {e}", style="bold red")
+
             console.print(Panel(f"[bold green]Submitted Answer[/bold green]\n{submitted_answer}", expand=False))
             console.print(Panel(f"The correct final answer is: [bold]{final_answer}[/bold]"))
 
@@ -88,7 +103,6 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
 
     submission_data = {"username": username.strip(), "agent_code": agent_code, "answers": answers_payload}
     result_data = submit_answers(submission_data)
-
     if result_data:
         final_status = (
             f"Submission Successful!\n"
@@ -101,3 +115,4 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
         return final_status, results_df
     else:
         return "Submission Failed.", pd.DataFrame(results_log)
+