Update Dockerfile for 4xL40S multi-GPU training

Dockerfile

@@ -14,11 +14,13 @@ RUN pip install --no-cache-dir -r requirements.txt
 
 # 学習スクリプト
 COPY train.py .
+COPY train_multi_gpu.py .
 
 # HFトークンは環境変数で渡す
 ENV HF_TOKEN=""
 ENV TRANSFORMERS_CACHE=/app/cache
 ENV HF_HOME=/app/cache
 
-# 学習実行
-CMD ["python", "train.py"]
+# マルチGPU学習 (4xL40S)
+# シングルGPUの場合は: CMD ["python", "train.py"]
+CMD ["accelerate", "launch", "--num_processes", "4", "train_multi_gpu.py"]