From 659871118f998e25631553bd628498aaae117cd8 Mon Sep 17 00:00:00 2001
From: tkrmagid <tkrmagid@git.tkrmagid.kr>
Date: Thu, 21 May 2026 21:32:54 +0900
Subject: [PATCH] =?UTF-8?q?fix(compose-gpu):=20MODEL=5FDEVICE=20=EB=A5=BC?=
 =?UTF-8?q?=20.env=20=EB=A1=9C=20=EB=8D=AE=EC=96=B4=EC=93=B8=20=EC=88=98?=
 =?UTF-8?q?=20=EC=9E=88=EA=B2=8C=20=EB=B3=80=EA=B2=BD?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

이전: MODEL_DEVICE: cuda (하드코딩) → .env 에 MODEL_DEVICE=cpu 두어도
compose environment 가 env_file 보다 우선이라 무시됨.

수정: MODEL_DEVICE: ${MODEL_DEVICE:-cuda} — 셸 env 또는 .env 의 값이
있으면 그 값, 없으면 기본 cuda.

용도: FinBERT/Chronos 가 GPU 에서 'no kernel image is available for
execution on the device' (PyTorch cu121 / RTX 3070 Ti 호환 이슈) 가 날 때
.env 에 MODEL_DEVICE=cpu 두고 backend 만 recreate 하면 CPU 폴백.

  echo MODEL_DEVICE=cpu>> .env
  docker compose -f docker-compose.yml -f docker-compose.gpu.yml up -d backend
---
 docker-compose.gpu.yml | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

diff --git a/docker-compose.gpu.yml b/docker-compose.gpu.yml
index 21ae651..16f4efd 100644
--- a/docker-compose.gpu.yml
+++ b/docker-compose.gpu.yml
@@ -8,5 +8,8 @@ services:
               count: all
               capabilities: [gpu]
     environment:
-      MODEL_DEVICE: cuda
+      # MODEL_DEVICE 는 .env 로 덮어쓰기 가능. GPU 빌드라도 PyTorch/CUDA 호환 문제 (예:
+      # 'no kernel image is available for execution on the device') 발생 시 .env 에
+      # MODEL_DEVICE=cpu 를 두고 `docker compose ... up -d backend` 로 회피.
+      MODEL_DEVICE: ${MODEL_DEVICE:-cuda}
       NVIDIA_VISIBLE_DEVICES: all