start.sh

#!/bin/bash
set -e

# Find llama-server binary
LLAMA_BIN=$(find /usr /app /llama.cpp /usr/local / -maxdepth 6 -name "llama-server" -type f 2>/dev/null | head -1)

if [ -z "$LLAMA_BIN" ]; then
    echo "ERROR: llama-server binary not found"
    exit 1
fi

echo "Found llama-server at: $LLAMA_BIN"

"$LLAMA_BIN" \
    -m /app/gemma-4-E2B-it-UD-Q5_K_XL.gguf \
    --mmproj /app/mmproj-BF16.gguf \
    --host 127.0.0.1 \
    --port 8080 \
    -t 2 \
    --cache-type-k q8_0 \
    --cache-type-v iq4_nl \
    -c 128000 \
    -n 38912 &

LLAMA_PID=$!
echo "llama-server started (PID $LLAMA_PID)"

# Wait up to 5 minutes for llama-server to be healthy
echo "Waiting for llama-server to be ready..."
for i in $(seq 1 150); do
    if curl -sf http://127.0.0.1:8080/health > /dev/null 2>&1; then
        echo "llama-server is ready"
        break
    fi
    if ! kill -0 "$LLAMA_PID" 2>/dev/null; then
        echo "ERROR: llama-server process died"
        exit 1
    fi
    if [ "$i" -eq 150 ]; then
        echo "ERROR: llama-server did not become ready in time"
        exit 1
    fi
    sleep 2
done

exec uvicorn main:app --host 0.0.0.0 --port 7860