Initial commit: Qwen3.6-MoE-35B-A3B server configuration and documentation

run_bge_m3_embedding_server.sh

@@ -0,0 +1,81 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+# Konfiguration
+HF_HOME="${HF_HOME:-/home/dschlueter/nvme2n1p7_home/huggingface}"
+MODEL_REL_PATH="models/embeddings/bge-m3-q8_0.gguf"
+IMAGE="ghcr.io/ggml-org/llama.cpp:server-cuda"
+CONTAINER_NAME="qwen-embeddings"
+HOST_PORT=8001
+CONTAINER_PORT=8001
+
+echo "[*] Verwende HF_HOME = $HF_HOME"
+if [ ! -f "$HF_HOME/$MODEL_REL_PATH" ]; then
+  echo "[!] Embedding-Modell-Datei nicht gefunden: $HF_HOME/$MODEL_REL_PATH" >&2
+  exit 1
+fi
+
+# Optional: altes gleichnamiges Container-Exemplar stoppen
+if docker ps -a --format '{{.Names}}' | grep -q "^${CONTAINER_NAME}\\$"; then
+  echo "[*] Stoppe existierenden Container $CONTAINER_NAME ..."
+  docker rm -f "$CONTAINER_NAME" >/dev/null 2>&1 || true
+fi
+
+echo "[*] Starte llama.cpp-Embedding-Server-Container ($IMAGE) ..."
+
+docker run -d --gpus '"device=0"' \
+  --name "$CONTAINER_NAME" \
+  -e HF_HOME="/hf_home" \
+  -v "$HF_HOME:/hf_home:ro" \
+  -p "${HOST_PORT}:${CONTAINER_PORT}" \
+  "$IMAGE" \
+    --embedding \
+    -m "/hf_home/${MODEL_REL_PATH}" \
+    -c 8192 \
+    -ngl 999 \
+    -fa on \
+    --batch-size 1024 \
+    --ubatch-size 512 \
+    --host 0.0.0.0 \
+    --port "$CONTAINER_PORT"
+
+echo "[*] Container gestartet: $CONTAINER_NAME"
+echo "[*] Warte, bis HTTP-Port ${HOST_PORT} antwortet ..."
+
+READY=0
+for i in {1..60}; do
+  if curl -s "http://localhost:${HOST_PORT}/" >/dev/null 2>&1; then
+    echo "[*] Server antwortet auf http://localhost:${HOST_PORT}/"
+    READY=1
+    break
+  fi
+  echo "[*] Warte (${i}/60) ..."
+  sleep 2
+done
+
+if [ "$READY" -ne 1 ]; then
+  echo "[!] Embedding-Server wurde nicht rechtzeitig erreichbar." >&2
+  echo "[*] Letzte Container-Logs:"
+  docker logs --tail 200 "$CONTAINER_NAME" || true
+  exit 1
+fi
+
+sleep 3
+
+echo "[*] Sende Test-Embedding-Request an /v1/embeddings ..."
+
+RESPONSE="$(curl -s -X POST "http://localhost:${HOST_PORT}/v1/embeddings" \
+  -H "Content-Type: application/json" \
+  -d '{
+    "model": "bge-m3-q8_0",
+    "input": "Dies ist ein kurzer Testtext für den Embedding-Server."
+  }')"
+
+echo
+echo "[*] Antwort vom Server:"
+echo "$RESPONSE"
+
+echo
+echo "[*] Zum Stoppen des Servers:"
+echo "    docker rm -f $CONTAINER_NAME"
+