From 01a345979bdd2b9e5f2f228253b0a2c4b56c3628 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Sebastian=20Kr=C3=BCger?= <valknar@pivoine.art>
Date: Fri, 21 Nov 2025 18:46:33 +0100
Subject: [PATCH] fix: disable drop_params to preserve streaming metadata in
 LiteLLM

- Set drop_params: false in litellm_settings
- Set modify_params: false in litellm_settings
- Set drop_params: false in default_litellm_params
- Commented out LITELLM_DROP_PARAMS env var
- Removed --drop_params command flag

These settings were stripping critical streaming parameters causing
vLLM streaming responses to collapse into empty deltas
---
 ai/compose.yaml        | 5 ++---
 ai/litellm-config.yaml | 6 +++---
 2 files changed, 5 insertions(+), 6 deletions(-)

diff --git a/ai/compose.yaml b/ai/compose.yaml
index de2bb0c..aba3b63 100644
--- a/ai/compose.yaml
+++ b/ai/compose.yaml
@@ -100,7 +100,7 @@ services:
       ANTHROPIC_API_KEY: ${ANTHROPIC_API_KEY}
       LITELLM_MASTER_KEY: ${AI_LITELLM_API_KEY}
       DATABASE_URL: postgresql://${AI_DB_USER}:${AI_DB_PASSWORD}@ai_postgres:5432/litellm
-      LITELLM_DROP_PARAMS: 'true'
+      # LITELLM_DROP_PARAMS: 'true'  # DISABLED: Was breaking streaming
       NO_DOCS: 'true'
       NO_REDOC: 'true'
       # Performance optimizations
@@ -115,8 +115,7 @@ services:
         '--host',
         '0.0.0.0',
         '--port',
-        '4000',
-        '--drop_params'
+        '4000'
       ]
     depends_on:
       - ai_postgres
diff --git a/ai/litellm-config.yaml b/ai/litellm-config.yaml
index a790070..a2816ea 100644
--- a/ai/litellm-config.yaml
+++ b/ai/litellm-config.yaml
@@ -61,7 +61,7 @@ model_list:
       max_parallel_requests: 1
 
 litellm_settings:
-  drop_params: true
+  drop_params: false  # DISABLED: Was breaking streaming
   set_verbose: true  # Enable verbose logging for debugging streaming issues
   # Disable caching - it breaks streaming responses
   cache: false
@@ -73,7 +73,7 @@ litellm_settings:
   # Force strip specific parameters globally
   allowed_fails: 0
   # Modify params before sending to provider
-  modify_params: true
+  modify_params: false  # DISABLED: Was breaking streaming
   # Enable success and failure logging but minimize overhead
   success_callback: []  # Disable all success callbacks to reduce DB writes
   failure_callback: []  # Disable all failure callbacks
@@ -83,7 +83,7 @@ router_settings:
 
 # Drop unsupported parameters
 default_litellm_params:
-  drop_params: true
+  drop_params: false  # DISABLED: Was breaking streaming
 
 general_settings:
   disable_responses_id_security: true