1/n align deserialize() to library design

## What

Exclude Previous Run Outputs when performing config.resolve(), aka, model parser `.deserialize()`

## Why

When resolving (deserializing) a prompt, if the prompt as executed before, the output from the previous execution is currently included in the deserialized state. While this was the expected behbaiour in the past, it has become apparent that including previous outputs is counterintuitive and not user-friendly

python/src/aiconfig/default_parsers/openai.py

@@ -209,7 +209,7 @@ async def deserialize(
                 )
 
         # Add in the latest prompt
-        add_prompt_as_message(prompt, aiconfig, completion_params["messages"], params)
+        add_prompt_as_message(prompt, aiconfig, completion_params["messages"], params, is_last_prompt=True)
         await aiconfig.callback_manager.run_callbacks(
             CallbackEvent(
                 "on_deserialize_complete", __name__, {"output": completion_params}
@@ -452,7 +452,7 @@ def refine_chat_completion_params(model_settings):
 
 
 def add_prompt_as_message(
-    prompt: Prompt, aiconfig: "AIConfigRuntime", messages: List, params=None
+    prompt: Prompt, aiconfig: "AIConfigRuntime", messages: List, params=None, is_last_prompt: bool = False
 ):
     """
     Converts a given prompt to a message and adds it to the specified messages list.
@@ -491,7 +491,8 @@ def add_prompt_as_message(
         messages.append(message_data)
 
     output = aiconfig.get_latest_output(prompt)
-    if output:
+    # Avoid deserializing the last prompt's output. The output from the previous execution should not be included.
+    if output and is_last_prompt is not True:
         if output.output_type == "execute_result":
             output_message = output.data
             if output_message["role"] == "assistant":

typescript/lib/parsers/openai.ts

@@ -454,16 +454,19 @@ export class OpenAIChatModelParser extends ParameterizedModelParser<Chat.ChatCom
             aiConfig.getModelName(currentPrompt) ===
             aiConfig.getModelName(prompt)
           ) {
+            if (currentPrompt.name === prompt.name) {
+              // Add Last Prompt. We have reached the end of the chat history
+              this.addPromptAsMessage(currentPrompt, aiConfig, messages, params, true);
+              break;
+            }
             this.addPromptAsMessage(currentPrompt, aiConfig, messages, params);
           }
 
-          if (currentPrompt.name === prompt.name) {
-            // If this is the current prompt, then we have reached the end of the chat history
-            break;
-          }
+
         }
       } else {
-        this.addPromptAsMessage(prompt, aiConfig, messages, params);
+        // If we don't want to remember the chat context, then we only need to add the latest prompt as a message
+        this.addPromptAsMessage(prompt, aiConfig, messages, params, true);
       }
 
       // Update the completion params with the resolved messages
@@ -641,7 +644,8 @@ export class OpenAIChatModelParser extends ParameterizedModelParser<Chat.ChatCom
     prompt: Prompt,
     aiConfig: AIConfigRuntime,
     messages: Chat.ChatCompletionMessageParam[],
-    params?: JSONObject
+    params?: JSONObject,
+    isLastPrompt: Boolean = false,
   ) {
     // Resolve the prompt with the given parameters, and add it to the messages array
     const promptTemplate = this.getPromptTemplate(prompt, aiConfig);
@@ -668,7 +672,8 @@ export class OpenAIChatModelParser extends ParameterizedModelParser<Chat.ChatCom
     }
 
     const output = aiConfig.getLatestOutput(prompt);
-    if (output != null) {
+    // Avoid deserializing the last prompt's output. The output from the previous execution should not be included.
+    if (output != null && isLastPrompt !== true) {
       if (output.output_type === "execute_result") {
         const outputMessage =
           output.data as unknown as Chat.ChatCompletionMessageParam;