Various bug fixes

charts/azimuth-chat/ci/ui-only-values.yaml

@@ -5,12 +5,3 @@ azimuth-llm:
     service:
       zenith:
         enabled: false
-    appSettings:
-      # Verify that we can set non-standard LLM params
-      llm_params:
-        max_tokens: 101
-        temperature: 0.1
-        top_p: 0.15
-        top_k: 1
-        presence_penalty: 0.9
-        frequency_penalty: 1

charts/azimuth-chat/ci/ui-param-overrides.yaml

@@ -0,0 +1,16 @@
+azimuth-llm:
+  api:
+    enabled: false
+  ui:
+    service:
+      zenith:
+        enabled: false
+    appSettings:
+      # Verify that we can set non-standard LLM params
+      llm_params:
+        max_tokens: 101
+        temperature: 0.1
+        top_p: 0.15
+        top_k: 1
+        presence_penalty: 0.9
+        frequency_penalty: 1

web-apps/chat/app.py

@@ -61,7 +61,6 @@ class PossibleSystemPromptException(Exception):
     streaming=True,
 )
 
-
 def inference(latest_message, history):
     # Allow mutating global variable
     global BACKEND_INITIALISED

web-apps/chat/defaults.yml

@@ -14,12 +14,12 @@ page_title: Large Language Model
 # See https://platform.openai.com/docs/api-reference/chat/create
 # and https://docs.vllm.ai/en/v0.6.0/serving/openai_compatible_server.html#extra-parameters
 llm_params:
-  max_tokens:
+  max_tokens: 1000
   temperature: 0
-  top_p:
-  top_k:
-  frequency_penalty:
-  presence_penalty:
+  top_p: 1
+  top_k: -1
+  frequency_penalty: 0
+  presence_penalty: 0
 
 # Gradio theme constructor parameters (e.g. 'primary_hue')
 # See https://www.gradio.app/guides/theming-guide

web-apps/utils/utils.py

@@ -95,7 +95,7 @@ def load_settings() -> dict:
             "Please check for typos"
         )
     settings = {**defaults, **overrides}
-    if "backend_url" not in settings or not settings["backend_url"]:
+    if "backend_url" not in settings or settings["backend_url"] == defaults["backend_url"]:
         # Try to detect in-cluster address
         in_cluster_backend = api_address_in_cluster()
         if not in_cluster_backend: