Merge pull request #39 from zioproto/gpt4o

Make the completion model configurable

Author: jamesmcroft

infra/installation_script.tftpl

@@ -27,6 +27,7 @@ data:
     OPENAI_API_TYPE=azuread
     OPENAI_API_VERSION=2023-05-15
     OPENAI_API_BASE=${endpoint}
+    CHAT_MODEL_NAME=${chat_model_name}
 ---
 apiVersion: v1
 kind: Pod

infra/openai.tf

@@ -17,12 +17,12 @@ module "openai" {
   }
   deployment = {
     "chat_model" = {
-      name          = "gpt-35-turbo"
+      name          = var.chat_model_name
       model_format  = "OpenAI"
-      model_name    = "gpt-35-turbo"
-      model_version = "0301"
-      scale_type    = "Standard"
-      capacity      = 120
+      model_name    = var.chat_model_name
+      model_version = var.chat_model_version
+      scale_type    = var.scale_type
+      capacity      = 30
     },
     "embedding_model" = {
       name          = "text-embedding-ada-002"

infra/output.tf

@@ -1,11 +1,12 @@
 output "installation-script" {
   value = templatefile("installation_script.tftpl",
-    { resourceGroup = azurerm_resource_group.this.name,
-      aksName       = module.aks.aks_name,
-      registry      = azurerm_container_registry.acr.name,
-      endpoint      = module.openai.openai_endpoint,
-      clientid      = azurerm_user_assigned_identity.chatbot.client_id,
-      oidc_url      = module.aks.oidc_issuer_url,
+    { resourceGroup   = azurerm_resource_group.this.name,
+      aksName         = module.aks.aks_name,
+      registry        = azurerm_container_registry.acr.name,
+      endpoint        = module.openai.openai_endpoint,
+      clientid        = azurerm_user_assigned_identity.chatbot.client_id,
+      oidc_url        = module.aks.oidc_issuer_url,
+      chat_model_name = var.chat_model_name
     }
   )
 }

infra/variables.tf

@@ -1,4 +1,20 @@
 variable "region" {
   type    = string
   default = "eastus"
-}
+}
+
+variable "chat_model_name" {
+  type    = string
+  default = "gpt-4o"
+}
+
+variable "chat_model_version" {
+  type    = string
+  default = "2024-08-06"
+}
+
+variable "scale_type" {
+  type    = string
+  description = "values: GlobalStandard, Standard"
+  default = "GlobalStandard"
+}

sample-application/chatbot.py

@@ -8,6 +8,7 @@
     OPENAI_API_VERSION = 2023-05-15
     OPENAI_API_BASE = 'https://eastus.api.cognitive.microsoft.com/' # Replace with the URL of an Azure OpenAI
     OPENAI_API_KEY = '' # Replace with the corresponding API key
+    CHAT_MODEL_NAME = gpt-4o
 
 To run the application, use the following command:
 streamlit run chatbot.py
@@ -81,7 +82,7 @@ def send_click():
 Settings.context_window = 4096
 # Create the chat llm
 Settings.llm = AzureChatOpenAI(
-    deployment_name="gpt-35-turbo",
+    deployment_name=st.session_state.config["CHAT_MODEL_NAME"],
     openai_api_key=st.session_state.config["OPENAI_API_KEY"],
     openai_api_base=st.session_state.config["OPENAI_API_BASE"],
     openai_api_type=st.session_state.config["OPENAI_API_TYPE"],