Break the Danswer LLM logging from LiteLLM Verbose (#1795)

.vscode/env_template.txt

@@ -8,7 +8,7 @@ AUTH_TYPE=disabled
 
 # Always keep these on for Dev
 # Logs all model prompts to stdout
-LOG_ALL_MODEL_INTERACTIONS=True
+LOG_DANSWER_MODEL_INTERACTIONS=True
 # More verbose logging
 LOG_LEVEL=debug
 

.vscode/launch.template.jsonc

@@ -49,7 +49,7 @@
             "cwd": "${workspaceFolder}/backend",
             "envFile": "${workspaceFolder}/.env",
             "env": {
-                "LOG_ALL_MODEL_INTERACTIONS": "True",
+                "LOG_DANSWER_MODEL_INTERACTIONS": "True",
                 "LOG_LEVEL": "DEBUG",
                 "PYTHONUNBUFFERED": "1"
             },

backend/danswer/configs/app_configs.py

@@ -266,10 +266,14 @@ JOB_TIMEOUT = 60 * 60 * 6  # 6 hours default
 CURRENT_PROCESS_IS_AN_INDEXING_JOB = (
     os.environ.get("CURRENT_PROCESS_IS_AN_INDEXING_JOB", "").lower() == "true"
 )
-# Logs every model prompt and output, mostly used for development or exploration purposes
+# Sets LiteLLM to verbose logging
 LOG_ALL_MODEL_INTERACTIONS = (
     os.environ.get("LOG_ALL_MODEL_INTERACTIONS", "").lower() == "true"
 )
+# Logs Danswer only model interactions like prompts, responses, messages etc.
+LOG_DANSWER_MODEL_INTERACTIONS = (
+    os.environ.get("LOG_DANSWER_MODEL_INTERACTIONS", "").lower() == "true"
+)
 # If set to `true` will enable additional logs about Vespa query performance
 # (time spent on finding the right docs + time spent fetching summaries from disk)
 LOG_VESPA_TIMING_INFORMATION = (

backend/danswer/llm/chat_llm.py

@@ -23,6 +23,7 @@ from langchain_core.messages.tool import ToolCallChunk
 from langchain_core.messages.tool import ToolMessage
 
 from danswer.configs.app_configs import LOG_ALL_MODEL_INTERACTIONS
+from danswer.configs.app_configs import LOG_DANSWER_MODEL_INTERACTIONS
 from danswer.configs.model_configs import DISABLE_LITELLM_STREAMING
 from danswer.configs.model_configs import GEN_AI_API_ENDPOINT
 from danswer.configs.model_configs import GEN_AI_API_VERSION
@@ -316,7 +317,7 @@ class DefaultMultiLLM(LLM):
         tools: list[dict] | None = None,
         tool_choice: ToolChoiceOptions | None = None,
     ) -> BaseMessage:
-        if LOG_ALL_MODEL_INTERACTIONS:
+        if LOG_DANSWER_MODEL_INTERACTIONS:
             self.log_model_configs()
             self._log_prompt(prompt)
 
@@ -333,7 +334,7 @@ class DefaultMultiLLM(LLM):
         tools: list[dict] | None = None,
         tool_choice: ToolChoiceOptions | None = None,
     ) -> Iterator[BaseMessage]:
-        if LOG_ALL_MODEL_INTERACTIONS:
+        if LOG_DANSWER_MODEL_INTERACTIONS:
             self.log_model_configs()
             self._log_prompt(prompt)
 
@@ -361,7 +362,7 @@ class DefaultMultiLLM(LLM):
                 "The AI model failed partway through generation, please try again."
             )
 
-        if LOG_ALL_MODEL_INTERACTIONS and output:
+        if LOG_DANSWER_MODEL_INTERACTIONS and output:
             content = output.content or ""
             if isinstance(output, AIMessage):
                 if content:

deployment/docker_compose/docker-compose.dev.yml

@@ -83,7 +83,9 @@ services:
       # https://docs.danswer.dev/more/telemetry
       - DISABLE_TELEMETRY=${DISABLE_TELEMETRY:-}
       - LOG_LEVEL=${LOG_LEVEL:-info}  # Set to debug to get more fine-grained logs
-      - LOG_ALL_MODEL_INTERACTIONS=${LOG_ALL_MODEL_INTERACTIONS:-}  # Log all of the prompts to the LLM
+      - LOG_ALL_MODEL_INTERACTIONS=${LOG_ALL_MODEL_INTERACTIONS:-}  # LiteLLM Verbose Logging
+      # Log all of Danswer prompts and interactions with the LLM
+      - LOG_DANSWER_MODEL_INTERACTIONS=${LOG_DANSWER_MODEL_INTERACTIONS:-}
       # If set to `true` will enable additional logs about Vespa query performance
       # (time spent on finding the right docs + time spent fetching summaries from disk)
       - LOG_VESPA_TIMING_INFORMATION=${LOG_VESPA_TIMING_INFORMATION:-}
@@ -191,7 +193,9 @@ services:
       # https://docs.danswer.dev/more/telemetry
       - DISABLE_TELEMETRY=${DISABLE_TELEMETRY:-}
       - LOG_LEVEL=${LOG_LEVEL:-info}  # Set to debug to get more fine-grained logs
-      - LOG_ALL_MODEL_INTERACTIONS=${LOG_ALL_MODEL_INTERACTIONS:-}  # Log all of the prompts to the LLM
+      - LOG_ALL_MODEL_INTERACTIONS=${LOG_ALL_MODEL_INTERACTIONS:-}  # LiteLLM Verbose Logging
+      # Log all of Danswer prompts and interactions with the LLM
+      - LOG_DANSWER_MODEL_INTERACTIONS=${LOG_DANSWER_MODEL_INTERACTIONS:-}
       - LOG_VESPA_TIMING_INFORMATION=${LOG_VESPA_TIMING_INFORMATION:-}
 
       # Enterprise Edition stuff

deployment/docker_compose/docker-compose.gpu-dev.yml

@@ -79,7 +79,9 @@ services:
       # https://docs.danswer.dev/more/telemetry
       - DISABLE_TELEMETRY=${DISABLE_TELEMETRY:-}
       - LOG_LEVEL=${LOG_LEVEL:-info}  # Set to debug to get more fine-grained logs
-      - LOG_ALL_MODEL_INTERACTIONS=${LOG_ALL_MODEL_INTERACTIONS:-}  # Log all of the prompts to the LLM
+      - LOG_ALL_MODEL_INTERACTIONS=${LOG_ALL_MODEL_INTERACTIONS:-}  # LiteLLM Verbose Logging
+      # Log all of Danswer prompts and interactions with the LLM
+      - LOG_DANSWER_MODEL_INTERACTIONS=${LOG_DANSWER_MODEL_INTERACTIONS:-}
       # If set to `true` will enable additional logs about Vespa query performance
       # (time spent on finding the right docs + time spent fetching summaries from disk)
       - LOG_VESPA_TIMING_INFORMATION=${LOG_VESPA_TIMING_INFORMATION:-}
@@ -183,7 +185,9 @@ services:
       # https://docs.danswer.dev/more/telemetry
       - DISABLE_TELEMETRY=${DISABLE_TELEMETRY:-}
       - LOG_LEVEL=${LOG_LEVEL:-info}  # Set to debug to get more fine-grained logs
-      - LOG_ALL_MODEL_INTERACTIONS=${LOG_ALL_MODEL_INTERACTIONS:-}  # Log all of the prompts to the LLM
+      - LOG_ALL_MODEL_INTERACTIONS=${LOG_ALL_MODEL_INTERACTIONS:-}  # LiteLLM Verbose Logging
+      # Log all of Danswer prompts and interactions with the LLM
+      - LOG_DANSWER_MODEL_INTERACTIONS=${LOG_DANSWER_MODEL_INTERACTIONS:-}
       - LOG_VESPA_TIMING_INFORMATION=${LOG_VESPA_TIMING_INFORMATION:-}
 
       # Enterprise Edition only

deployment/helm/values.yaml

@@ -449,6 +449,7 @@ configMap:
   DISABLE_TELEMETRY: ""
   LOG_LEVEL: ""
   LOG_ALL_MODEL_INTERACTIONS: ""
+  LOG_DANSWER_MODEL_INTERACTIONS: ""
   LOG_VESPA_TIMING_INFORMATION: ""
   # Shared or Non-backend Related
   WEB_DOMAIN: "http://localhost:3000"  # for web server and api server

deployment/kubernetes/env-configmap.yaml

@@ -78,6 +78,7 @@ data:
   DISABLE_TELEMETRY: ""
   LOG_LEVEL: ""
   LOG_ALL_MODEL_INTERACTIONS: ""
+  LOG_DANSWER_MODEL_INTERACTIONS: ""
   LOG_VESPA_TIMING_INFORMATION: ""
   # Shared or Non-backend Related
   INTERNAL_URL: "http://api-server-service:80"  # for web server