#!/bin/bash
set -e

echo "=== Downloading model ==="
MODEL_PATH=$(python3 -c "
from huggingface_hub import hf_hub_download
path = hf_hub_download(
    repo_id='Jackrong/Qwen3.5-9B-Claude-4.6-Opus-Reasoning-Distilled-v2-GGUF',
    filename='Qwen3.5-9B.Q4_K_M.gguf',
)
print(path)
")

echo "Model path: $MODEL_PATH"

# Find llama-server and cd into its directory so it finds .so backends
LLAMA_SERVER=$(find /opt/llama -name "llama-server" -type f | head -1)
LLAMA_DIR=$(dirname "$LLAMA_SERVER")
echo "llama-server dir: $LLAMA_DIR"
echo "Files in dir:"
ls -la "$LLAMA_DIR"/*.so* 2>/dev/null || echo "no .so files here"
ls -la "$LLAMA_DIR"/../ 2>/dev/null

echo "=== Starting llama-server on port 7860 ==="
cd "$LLAMA_DIR"
exec ./llama-server \
    --model "$MODEL_PATH" \
    --host 0.0.0.0 \
    --port 7860 \
    --ctx-size 2048 \
    --threads 2 \
    --batch-size 256