Tune audio model GPU profiles

2026-06-09 12:52:13 +03:00
parent f49ba7abd5
commit cac8d89e64
1 changed files with 2 additions and 2 deletions
--- a/deploy/ai-server/docker-compose.audio.yml
+++ b/deploy/ai-server/docker-compose.audio.yml
@@ -36,7 +36,7 @@ services:
      - "--max-model-len"
      - "8192"
      - "--gpu-memory-utilization"
-      - "0.45"
+      - "0.25"
      - "--api-key"
      - "${VLLM_API_KEY}"
      - "--max-num-seqs"
@@ -95,7 +95,7 @@ services:
      - "--max-model-len"
      - "32768"
      - "--gpu-memory-utilization"
-      - "0.72"
+      - "0.62"
      - "--api-key"
      - "${VLLM_API_KEY}"
      - "--max-num-seqs"