diff --git a/deploy/ai-server/docker-compose.audio.yml b/deploy/ai-server/docker-compose.audio.yml index 9db527b..35e06a2 100644 --- 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"