Update start.sh

start.sh

@@ -1,16 +1,10 @@
 #!/bin/bash
 set -e
 
-# ==========================
-# 工作目錄
-# ==========================
 export WORK="/home/user/app"
-mkdir -p $WORK
+mkdir -p "$WORK"
 cd "$WORK"
 
-# ==========================
-# Python 環境
-# ==========================
 python3 -m venv venv
 source venv/bin/activate
 
@@ -18,27 +12,21 @@ echo "[PYTHON] Installing dependencies..."
 pip install --upgrade pip
 pip install gradio openai "setuptools>=70.0.0" wheel flask
 
-# ==========================
-# llama-server 與模型
-# ==========================
 wget -q -O model.gguf https://huggingface.co/LiquidAI/LFM2-8B-A1B-GGUF/resolve/main/LFM2-8B-A1B-Q4_K_M.gguf
 unzip -o llama_cpp_avx512.zip > /dev/null 2>&1
+
+echo "[LLAMA] 啟動伺服器..."
 nohup ./llama-server -m "$WORK/model.gguf" --port 8000 --host 0.0.0.0 --threads 2 --ctx-size 4096 --mlock --jinja --temp 1.1 --top-p 0.95 \
   > llama_server.log 2>&1 &
 
-# ==========================
-# 顯示 TPS 資訊
-# ==========================
-echo "[LLAMA] 正在監看 TPS..."
-(
-  tail -n0 -F "$WORK/llama_server.log" | \
-  grep --line-buffered -E "t/s\)|tokens/s" | \
-  while read -r line; do
-    echo "[TPS] $(echo "$line" | grep -oE '[0-9]+\.[0-9]+ t/s')"
-  done
-) &
-
-cd $WORK
-python3 app.py
+sleep 3
 
+echo "[LLAMA] 顯示最新日誌（含 TPS）..."
+tail -n20 -F "$WORK/llama_server.log" | while read -r line; do
+    if [[ "$line" =~ ([0-9]+\.[0-9]+)\ t/s ]]; then
+        echo "[TPS] ${BASH_REMATCH[1]} t/s"
+    fi
+done &
 
+cd "$WORK"
+python3 app.py