Move LM clients to LiteLMM

opentofu/main.tf

@@ -80,6 +80,7 @@ resource "oci_database_autonomous_database" "default_adb" {
 module "vm" {
   count                 = var.infrastructure == "VM" ? 1 : 0
   source                = "./modules/vm"
+  optimizer_version     = var.optimizer_version
   label_prefix          = local.label_prefix
   tenancy_id            = var.tenancy_ocid
   compartment_id        = local.compartment_ocid

opentofu/modules/vm/iam.tf

@@ -34,6 +34,10 @@ resource "oci_identity_policy" "identity_node_policies" {
       "allow dynamic-group %s to read objects in compartment id %s",
       oci_identity_dynamic_group.compute_dynamic_group.name, var.compartment_id
     ),
+    format(
+      "allow dynamic-group %s to use generative-ai-family in compartment id %s",
+      oci_identity_dynamic_group.compute_dynamic_group.name, var.compartment_id
+    ),
   ]
   provider = oci.home_region
 }

opentofu/modules/vm/locals.tf

@@ -4,12 +4,13 @@
 
 locals {
   cloud_init_compute = templatefile("${path.module}/templates/cloudinit-compute.tpl", {
-    tenancy_id     = var.tenancy_id
-    compartment_id = var.compartment_id
-    oci_region     = var.region
-    db_name        = var.adb_name
-    db_password    = var.adb_password
-    install_ollama = var.vm_is_gpu_shape ? true : false
+    tenancy_id        = var.tenancy_id
+    compartment_id    = var.compartment_id
+    oci_region        = var.region
+    db_name           = var.adb_name
+    db_password       = var.adb_password
+    optimizer_version = var.optimizer_version
+    install_ollama    = var.vm_is_gpu_shape ? true : false
   })
 
   cloud_init_database = templatefile("${path.module}/templates/cloudinit-database.tpl", {

opentofu/modules/vm/templates/cloudinit-compute.tpl

@@ -57,11 +57,14 @@ write_files:
     permissions: '0755'
     content: |
       #!/bin/bash
-      # Setup for Instance Principles
-
       # Download/Setup Source Code
-      curl -L https://github.com/oracle/ai-optimizer/releases/latest/download/ai-optimizer-src.tar.gz \
-      | tar -xz -C /app
+      if [ "${optimizer_version}" = "main" ]; then
+          URL="https://github.com/oracle/ai-optimizer/archive/refs/heads/main.tar.gz"
+          curl -L "$URL" | tar -xz -C /app --strip-components=2 ai-optimizer-main/src
+      else
+          URL="https://github.com/oracle/ai-optimizer/releases/latest/download/ai-optimizer-src.tar.gz"
+          curl -L "$URL" | tar -xz -C /app
+      fi
       cd /app
       python3.11 -m venv .venv
       source .venv/bin/activate

opentofu/modules/vm/variables.tf

@@ -2,6 +2,10 @@
 # All rights reserved. The Universal Permissive License (UPL), Version 1.0 as shown at http://oss.oracle.com/licenses/upl
 # spell-checker: disable
 
+variable "optimizer_version" {
+  type = string
+}
+
 variable "tenancy_id" {
   type = string
 }

opentofu/schema.yaml

@@ -28,6 +28,7 @@ variableGroups:
 
   - title: "Hidden (Defaults)"
     variables:
+      - optimizer_version
       - adb_version
       - k8s_run_cfgmgt
     visible: false

opentofu/variables.tf

@@ -3,6 +3,16 @@
 # spell-checker: disable
 
 // Standard Default Vars
+variable "optimizer_version" {
+  description = "Determines if latest release or main is used"
+  type        = string
+  default     = "latest"
+  validation {
+    condition     = var.optimizer_version == "latest" || var.optimizer_version == "main"
+    error_message = "optimizer_version must be either 'latest' or 'main'."
+  }
+}
+
 variable "tenancy_ocid" {
   description = "The Tenancy ID of the OCI Cloud Account in which to create the resources."
   type        = string

src/client/content/api_server.py

@@ -14,18 +14,15 @@
 import streamlit as st
 from streamlit import session_state as state
 
-import client.utils.client as client
-import client.utils.api_call as api_call
-import common.logging_config as logging_config
+from client.utils import client, api_call
+from common import logging_config
 
 logger = logging_config.logging.getLogger("client.content.api_server")
 
 try:
     import launch_server
-
-    REMOTE_SERVER = False
 except ImportError:
-    REMOTE_SERVER = True
+    pass
 
 
 #####################################################
@@ -53,7 +50,7 @@ def server_restart() -> None:
     state.server["key"] = os.getenv("API_SERVER_KEY")
 
     launch_server.stop_server(state.server["pid"])
-    state.server["pid"] = launch_server.start_server(state.server["port"])
+    _, state.server["pid"] = launch_server.start_server(state.server["port"])
     time.sleep(10)
     state.pop("server_client", None)
 
@@ -72,16 +69,16 @@ async def main() -> None:
         key="user_server_port",
         min_value=1,
         max_value=65535,
-        disabled=REMOTE_SERVER,
+        disabled=state.server["remote"],
     )
     right.text_input(
         "API Server Key:",
         value=state.server["key"],
         key="user_server_key",
         type="password",
-        disabled=REMOTE_SERVER,
+        disabled=state.server["remote"],
     )
-    if not REMOTE_SERVER:
+    if not state.server["remote"]:
         st.button("Restart Server", type="primary", on_click=server_restart)
 
     st.header("Server Settings", divider="red")

src/client/content/chatbot.py

@@ -16,13 +16,9 @@
 from streamlit import session_state as state
 
 from client.content.config.tabs.models import get_models
-
-import client.utils.st_common as st_common
-import client.utils.api_call as api_call
-
+from client.utils import st_common, api_call, client
 from client.utils.st_footer import render_chat_footer
-import client.utils.client as client
-import common.logging_config as logging_config
+from common import logging_config
 
 logger = logging_config.logging.getLogger("client.content.chatbot")
 
@@ -143,13 +139,7 @@ async def main() -> None:
             # Stream until we hit the end then refresh to replace with history
             st.rerun()
         except Exception:
-            logger.error("Exception:", exc_info=1)
-            st.chat_message("ai").write(
-                """
-                I'm sorry, something's gone wrong.  Please try again.
-                If the problem persists, please raise an issue.
-                """
-            )
+            message_placeholder.markdown("An unexpected error occurred, please retry your request.")
             if st.button("Retry", key="reload_chatbot"):
                 st_common.clear_state_key("user_client")
                 st.rerun()

src/client/content/config/tabs/databases.py

@@ -13,22 +13,25 @@
 import streamlit as st
 from streamlit import session_state as state
 
-import client.utils.api_call as api_call
-import client.utils.st_common as st_common
-
-import common.logging_config as logging_config
+from client.utils import api_call, st_common
+from common import logging_config
 
 logger = logging_config.logging.getLogger("client.content.config.tabs.database")
 
 
 #####################################################
 # Functions
 #####################################################
-def get_databases(force: bool = False) -> None:
+def get_databases(validate: bool = False, force: bool = False) -> None:
     """Get Databases from API Server"""
     if force or "database_configs" not in state or not state.database_configs:
         try:
             logger.info("Refreshing state.database_configs")
+            # Validation will be done on currently configured client database
+            # validation includes new vector_stores, etc.
+            if validate:
+                client_database = state.client_settings.get("database", {}).get("alias", {})
+                _ = api_call.get(endpoint=f"v1/databases/{client_database}")
             state.database_configs = api_call.get(endpoint="v1/databases")
         except api_call.ApiError as ex:
             logger.error("Unable to populate state.database_configs: %s", ex)
@@ -63,7 +66,7 @@ def patch_database(name: str, supplied: dict, connected: bool) -> bool:
 def drop_vs(vs: dict) -> None:
     """Drop a Vector Storage Table"""
     api_call.delete(endpoint=f"v1/embed/{vs['vector_store']}")
-    get_databases(force=True)
+    get_databases(validate=True, force=True)
 
 
 def select_ai_profile() -> None:
@@ -221,7 +224,7 @@ def display_databases() -> None:
                     column_config={
                         "enabled": st.column_config.CheckboxColumn(label="Enabled", help="Toggle to enable or disable")
                     },
-                    use_container_width=True,
+                    width="stretch",
                     hide_index=True,
                 )
                 if st.button("Apply SelectAI Changes", type="secondary"):