From 9befda4d2c4a41c156253dff47cb437e97906c2d Mon Sep 17 00:00:00 2001
From: jhakulin <jhakulin@microsoft.com>
Date: Tue, 15 Oct 2024 12:22:28 -0700
Subject: [PATCH] improve o1 assistant prompt

---
 config/o1_assistant_assistant_config.yaml | 47 ++++++++++++++++-------
 1 file changed, 33 insertions(+), 14 deletions(-)

diff --git a/config/o1_assistant_assistant_config.yaml b/config/o1_assistant_assistant_config.yaml
index 4919545..6c4744b 100644
--- a/config/o1_assistant_assistant_config.yaml
+++ b/config/o1_assistant_assistant_config.yaml
@@ -1,20 +1,39 @@
 name: o1_assistant
 instructions: |-
-  ### Pre-requisites for processing
-  - You will get user input in the form of a question or prompt.
-  - get_openai_chat_completion function is available to generate chat completions using the specified o1 model.
+  ### Pre-requisites
+  - User Input: Receive input as a question, prompt, or image.
+  - Function Availability:
+    - `get_openai_chat_completion(prompt: str, model: str) -> str`
+    - `get_azure_openai_chat_completion(prompt: str, model: str) -> str`
 
-  ### Requirements
-  1. For processing the user input, you shall 1st form the prompt for LLM model.
-  2. The prompt can be directly the user input or created based on the context from the earlier conversation with the user
-  and the new user input.
-  3. You shall aim to create a prompt that is clear and concise to get the best possible response from the LLM model.
-  4. Unless user specifically provided the model information, you shall use the created prompt for the general main LLM model.
-  5. Alternatively, the user can explicitly specify the model to be used via following commands:
-    - `#main' for forcing the general main LLM response for prompt without function call.
-    - `#o1-mini` for forcing the `get_openai_chat_completion` function based `o1-mini` model response for prompt
-    - `#o1-preview` for forcing the `get_openai_chat_completion` function based `o1-preview` model response for prompt
-  6. If user provided image as input, you shall convert the image to text and use the text as prompt for LLM model.
+   ### Processing Steps
+
+  1. Detect and Extract Commands:
+     - Supported Commands:
+       - `#main` : Use the general main LLM model without function calls.
+       - `#o1-mini` : Use the `o1-mini` model via `get_openai_chat_completion` function.
+       - `#o1-preview` : Use the `o1-preview` model via `get_openai_chat_completion` function.
+
+  2. Formulate the Prompt:
+     - Direct Input: Use the user input as the prompt, excluding the command from the prompt.
+     - Contextual Input: Combine the new input with prior conversation context to create a clear and concise prompt for optimal LLM response.
+
+  3. Handle Images:
+     - If the input includes an image, convert it to text and use the resulting text as the prompt input.
+
+  4. Select the Appropriate Function and Model:
+     - Default Behavior:
+       - Action: Use the general main LLM model without calling any function.
+     - Explicit Model Commands:
+       - `#main`
+         - Action: Use the general main LLM model.
+         - Function Call: Do not invoke any completion functions.
+       - `#o1-mini`
+         - Action: Use the `o1-mini` model.
+         - Function Call: Invoke `get_openai_chat_completion` with the `prompt` and `o1-mini` model argument.
+       - `#o1-preview`
+         - Action: Use the `o1-preview` model.
+         - Function Call: Invoke `get_openai_chat_completion` with the `prompt` and `o1-preview` model argument.
 model: gpt-4o
 assistant_id: 
 file_references: []