llama.cpp: Embedding mit -ub 4096 (ub 8192 braucht 5 GB Rechenpuffer auf der P2000)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

deploy/compose/gpu-llamacpp/docker-compose.yml

@@ -39,7 +39,7 @@ services:
     container_name: llamacpp-embed
     profiles: ["embed"]
     command: ["-m", "/models/${EMBED_GGUF:-Qwen3-Embedding-0.6B-Q8_0.gguf}", "--embedding",
-              "--pooling", "${EMBED_POOLING:-last}", "-c", "8192", "-b", "8192", "-ub", "8192", "-np", "1",
+              "--pooling", "${EMBED_POOLING:-last}", "-c", "8192", "-b", "4096", "-ub", "4096", "-np", "1",   # ub 8192 sprengt die P2000 (Task 6.1)
               "-ngl", "99", "--alias", "embed-default", "--metrics", "--no-slots", "--no-webui",
               "--host", "0.0.0.0", "--port", "8081"]
     deploy: