fix: full runtime setup — clone, build, download, serve

start.sh

@@ -1,28 +1,49 @@
 #!/bin/bash
 set -e
 
-MODEL_DIR="/home/user/app/models"
-MODEL_PATH="$MODEL_DIR/ggml-model-i2_s.gguf"
-SERVER_BIN="/home/user/app/llama-server"
+WORK_DIR="/home/user/app"
+BITNET_DIR="$WORK_DIR/BitNet"
+MODEL_PATH="$WORK_DIR/BitNet/models/bitnet-b1.58-2B-4T-gguf/ggml-model-i2_s.gguf"
+SERVER_BIN="$WORK_DIR/BitNet/build/bin/llama-server"
 
-# Download model if not present (runtime download to avoid build timeout)
+# ─── Step 1: Clone and build bitnet.cpp if not already done ──────────────────
+if [ ! -f "$SERVER_BIN" ]; then
+    echo "=== First run: building bitnet.cpp ==="
+    echo "This takes ~5 minutes. Subsequent restarts will be fast."
+    echo ""
+
+    if [ ! -d "$BITNET_DIR" ]; then
+        echo "[1/3] Cloning bitnet.cpp..."
+        git clone --depth 1 --recursive https://github.com/microsoft/BitNet.git "$BITNET_DIR"
+        pip install --no-cache-dir -r "$BITNET_DIR/requirements.txt"
+    fi
+
+    echo "[2/3] Building with I2_S kernel..."
+    cd "$BITNET_DIR"
+    python setup_env.py --hf-repo microsoft/bitnet-b1.58-2B-4T-gguf -q i2_s
+    echo "Build complete!"
+    cd "$WORK_DIR"
+else
+    echo "bitnet.cpp already built, skipping..."
+fi
+
+# ─── Step 2: Verify model exists ─────────────────────────────────────────────
 if [ ! -f "$MODEL_PATH" ]; then
-    echo "Downloading BitNet b1.58 2B4T GGUF model (1.1 GB)..."
+    echo "[3/3] Downloading model..."
     python -c "
 from huggingface_hub import hf_hub_download
-import os
-path = hf_hub_download(
+hf_hub_download(
     repo_id='microsoft/bitnet-b1.58-2B-4T-gguf',
     filename='ggml-model-i2_s.gguf',
-    local_dir='$MODEL_DIR'
+    local_dir='$BITNET_DIR/models/bitnet-b1.58-2B-4T-gguf'
 )
-print(f'Downloaded to: {path}')
+print('Model downloaded!')
 "
-    echo "Model downloaded!"
 fi
 
-# Start llama-server in background
-echo "Starting bitnet.cpp llama-server..."
+# ─── Step 3: Start llama-server ──────────────────────────────────────────────
+echo ""
+echo "=== Starting bitnet.cpp llama-server ==="
 $SERVER_BIN \
     -m "$MODEL_PATH" \
     --host 127.0.0.1 \
@@ -33,20 +54,22 @@ $SERVER_BIN \
 
 SERVER_PID=$!
 
-# Wait for server to be ready
-echo "Waiting for server to start..."
+# Wait for server
+echo "Waiting for server..."
 for i in $(seq 1 120); do
     if curl -s http://127.0.0.1:8080/health > /dev/null 2>&1; then
-        echo "Server ready! (took ${i}s)"
+        echo "Server ready! (${i}s)"
         break
     fi
     if [ $i -eq 120 ]; then
-        echo "ERROR: Server failed to start after 120s"
+        echo "ERROR: Server failed to start"
+        # Try to show what went wrong
+        $SERVER_BIN -m "$MODEL_PATH" --host 127.0.0.1 --port 8081 -t 2 -c 512 2>&1 | head -20
         exit 1
     fi
     sleep 1
 done
 
-# Start Gradio app
+# ─── Step 4: Start Gradio ────────────────────────────────────────────────────
 echo "Starting Gradio app..."
 exec python app.py