[Release] Docs Agent version 0.1.4 (google#167)

What's changed:

- Bug fix: Update `docs_agent.py` to remove a no-longer-in-use error message variable.
- Clean up the prompt format in Docs Agent
  - Context is added to prompts as Markdown
  - Remove extra new lines in context
  - Add extra new lines after the instruction and before the question in prompts.
- Remove the warning message in the Chroma module, which as displayed
  when launching the chatbot.
- Update the embeddings diagrams in the main `README` file.
- Minor updates in the main `README` file.

Author: kyolee415

demos/palm/python/docs-agent/README.md

@@ -125,6 +125,10 @@ The following list summarizes the tasks and features of the Docs Agent sample ap
   the responses. (See the
   [Enabling users to submit a rewrite of a generated response][submit-a-rewrite] and
   [Enabling users to like generated responses][like-generate-responses] sections.)
+- **Convert Google Docs, PDF, and Gmail into Markdown files**: This feature uses
+  Apps Script to convert Google Docs, PDF, and Gmail into Markdown files, which then
+  can be used as input datasets for Docs Agent. For more information, see the
+  [`README`][apps-script-readme] file in the `apps_script` directory.
 
 ## Flow of events
 
@@ -205,10 +209,10 @@ by the PaLM model:
 - Additional condition (for fact-checking):
 
   ```
-  Can you compare the following text to the context provided in this prompt and write
-  a short message that warns the readers about which part of the text they should
-  consider fact-checking? (Please keep your response concise and focus on only
-  one important item.)
+  Can you compare the text below to the context provided
+  in this prompt above and write a short message that warns the readers about
+  which part of the text they should consider fact-checking? (Please keep your
+  response concise and focus on only one important item.)"
   ```
 
 - Previously generated response
@@ -597,20 +601,18 @@ To launch the Docs Agent chat app, do the following:
 
    ```
    $ poetry run ./chatbot/launch.sh
-   This script starts your flask app in a virtual environment
-   Installing all dependencies through pip...
-   Using the local vector database created at /home/alice/generative-ai-docs/demos/palm/python/docs-agent/vector_database
-   Using embedded DuckDB with persistence: data will be stored in: /home/alice/generative-ai-docs/demos/palm/python/docs-agent/vector_database
+   Reading the config file: /home/alice/docs-agent/config.yaml
+   Reading the config file: /home/alice/docs-agent/config.yaml
     * Serving Flask app 'chatbot'
     * Debug mode: on
-   WARNING: This is a development server. Do not use it in a production deployment. Use a production WSGI server instead.
+   INFO:werkzeug:WARNING: This is a development server. Do not use it in a production deployment. Use a production WSGI server instead.
     * Running on http://example.com:5000
-   Press CTRL+C to quit
-    * Restarting with stat
-   Using the local vector database created at /home/alice/generative-ai-docs/demos/palm/python/docs-agent/vector_database
-   Using embedded DuckDB with persistence: data will be stored in: /home/alice/generative-ai-docs/demos/palm/python/docs-agent/vector_database
-    * Debugger is active!
-    * Debugger PIN: 129-640-957
+   INFO:werkzeug:Press CTRL+C to quit
+   INFO:werkzeug: * Restarting with stat
+   Reading the config file: /home/alice/docs-agent/config.yaml
+   Reading the config file: /home/alice/docs-agent/config.yaml
+   WARNING:werkzeug: * Debugger is active!
+   INFO:werkzeug: * Debugger PIN: 825-594-989
    ```
 
    Notice the line that shows the URL of this server (`http://example.com:5000` in

demos/palm/python/docs-agent/chatbot/chatui.py

@@ -143,13 +143,13 @@ def ask_model(question):
     # 3. Add the custom condition text to the context.
     # 4. Send the prompt (condition + context + question) to the language model.
     query_result = docs_agent.query_vector_store(question)
-    context = markdown.markdown(query_result.fetch_formatted(Format.CONTEXT))
-    context_with_prefix = docs_agent.add_instruction_to_context(context)
-    response = docs_agent.ask_text_model_with_context(context_with_prefix, question)
+    context = query_result.fetch_formatted(Format.CONTEXT)
+    context_with_instruction = docs_agent.add_instruction_to_context(context)
+    response = docs_agent.ask_text_model_with_context(context_with_instruction, question)
 
     ### PROMPT 2: FACT-CHECK THE PREVIOUS RESPONSE.
     fact_checked_response = docs_agent.ask_text_model_to_fact_check(
-        context_with_prefix, response
+        context_with_instruction, response
     )
 
     ### PROMPT 3: GET 5 RELATED QUESTIONS.
@@ -176,10 +176,12 @@ def ask_model(question):
 
     ### PREPARE OTHER ELEMENTS NEEDED BY UI.
     # - Create a uuid for this request.
+    # - Convert the context returned from the database into HTML for rendering.
     # - Convert the first response from the model into HTML for rendering.
     # - Convert the fact-check response from the model into HTML for rendering.
     # - A workaround to get the server's URL to work with the rewrite and like features.
     new_uuid = uuid.uuid1()
+    context_in_html = markdown.markdown(context)
     response_in_html = markdown.markdown(response)
     fact_checked_response_in_html = markdown.markdown(fact_checked_response)
     server_url = request.url_root.replace("http", "https")
@@ -190,7 +192,7 @@ def ask_model(question):
     return render_template(
         "chatui/index.html",
         question=question,
-        context=context,
+        context_in_html=context_in_html,
         response=response,
         response_in_html=response_in_html,
         clickable_urls=clickable_urls,

demos/palm/python/docs-agent/chatbot/templates/chatui/result.html

@@ -22,7 +22,7 @@ <h2 class="handle">
     <label for="handle1">Context</label>
   </h2>
   <div class="content">
-    {{ context | safe }}
+    {{ context_in_html | safe }}
     <span class="reference-content">
       <h4>Reference:</h4>
       {{ clickable_urls | safe }}

demos/palm/python/docs-agent/chroma.py

@@ -69,9 +69,9 @@ def get_collection(self, name, embedding_function=None):
             )
         elif embedding_model is None or embedding_model == "palm/embedding-gecko-001":
             if embedding_model is None:
-                logging.warning(
-                    "Embedding model is not stored in the metadata of "
-                    "the collection %s. Using PaLM as default.",
+                logging.info(
+                    "Embedding model is not specified in the metadata of "
+                    "the collection %s. Using the default PaLM embedding model.",
                     name,
                 )
             palm = PaLM(embed_model="models/embedding-gecko-001", find_models=False)
@@ -138,7 +138,7 @@ def __init__(self, result: QueryResult, index: int) -> None:
 
     def format(self, format_type: Format, ref_index: int = None):
         d = {
-            "document": self.document,
+            "document": self.document.strip(),
             "ref_index": ref_index,
             "url": self.metadata.get("url", None),
             "distance": self.distance,

demos/palm/python/docs-agent/config.yaml

@@ -73,10 +73,10 @@ input:
 condition_text: "You are a helpful chatbot answering questions from users. Read
 the following context first and answer the question at the end:"
 
-fact_check_question: "Can you compare the following text to the context provided
-in this prompt and write a short message that warns the readers about which part
-of the text they should consider fact-checking? (Please keep your response
-concise and focus on only one important item.)"
+fact_check_question: "Can you compare the text below to the context provided
+in this prompt above and write a short message that warns the readers about
+which part of the text they should consider fact-checking? (Please keep your
+response concise and focus on only one important item.)"
 
 model_error_message: "PaLM is not able to answer this question at the
 moment. Rephrase the question and try asking again."

demos/palm/python/docs-agent/docs/images/docs-agent-embeddings-01.png

@@ -73,7 +73,7 @@ def __init__(self):
 
     # Use this method for talking to PaLM (Text)
     def ask_text_model_with_context(self, context, question):
-        new_prompt = f"{context}\nQuestion: {question}"
+        new_prompt = f"{context}\n\nQuestion: {question}"
         try:
             response = palm.generate_text(
                 prompt=new_prompt,
@@ -101,7 +101,7 @@ def ask_chat_model_with_context(self, context, question):
             return self.model_error_message
 
         if response.last is None:
-            return self.palm_none_response
+            return self.model_error_message
         return response.last
 
     # Use this method for asking PaLM (Text) for fact-checking
@@ -117,5 +117,5 @@ def query_vector_store(self, question):
     # Add specific instruction as a prefix to the context
     def add_instruction_to_context(self, context):
         new_context = ""
-        new_context += self.prompt_condition + "\n" + context
+        new_context += self.prompt_condition + "\n\n" + context
         return new_context

demos/palm/python/docs-agent/docs/images/docs-agent-embeddings-02.png

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "docs-agent"
-version = "0.1.3"
+version = "0.1.4"
 description = ""
 authors = ["Docs Agent contributors"]
 readme = "README.md"

demos/palm/python/docs-agent/docs_agent.py

@@ -43,11 +43,11 @@ def __init__(self):
             with open(INPUT_YAML, "r", encoding="utf-8") as inp_yaml:
                 self.config_values = yaml.safe_load(inp_yaml)
                 self.IS_CONFIG_FILE = True
-                print("Configuration defined in: " + INPUT_YAML)
+                print("Reading the config file: " + INPUT_YAML)
                 # Check that the required keys exist
                 self.validateKeys()
         except FileNotFoundError:
-            print("The file " + INPUT_YAML + " does not exist.")
+            print("The config file " + INPUT_YAML + " does not exist.")
             # Exits the scripts if there is no valid config file
             return sys.exit(1)