{% raw %}# List available recipes
[private]
default:
    @just --list

# Show currently running vLLM server
status:
    @pgrep -a -f "vllm serve" || echo "No vLLM server running"

# Tail the vLLM log
logs:
    @tail -f {% endraw %}{{ vllm_home }}/vllm.log{% raw %}

# Stop any running vLLM server and wait for VRAM to free
stop:
    #!/usr/bin/env bash
    set -euo pipefail
    if pgrep -f "vllm serve" > /dev/null; then
        echo "Stopping vLLM..."
        pkill -TERM -f "vllm serve" || true
        sleep 2
        pkill -KILL -f "vllm serve" 2>/dev/null || true
    fi
    echo "Waiting for VRAM to release..."
    for i in $(seq 1 30); do
        used=$(rocm-smi --showmeminfo vram 2>/dev/null | awk '/VRAM Total Used Memory/ {print $NF}')
        total=$(rocm-smi --showmeminfo vram 2>/dev/null | awk '/VRAM Total Memory \(B\)/ {print $NF}')
        if [ -n "$used" ] && [ -n "$total" ] && [ "$total" -gt 0 ]; then
            pct=$(( used * 100 / total ))
            echo "  VRAM: ${pct}%"
            if [ "$pct" -lt 10 ]; then
                echo "VRAM free."
                exit 0
            fi
        fi
        sleep 2
    done
    echo "Warning: VRAM did not fully release after 60s"
{% endraw %}
{% for model in vllm_models_list %}
# Serve {{ model.name }}
{{ model.recipe }}: stop
    #!/usr/bin/env bash
    source {{ vllm_home }}/.bashrc
    nohup {{ vllm_home }}/vllm-venv/bin/vllm serve {{ vllm_models }}/{{ model.dir }} \
        --served-model-name {{ model.name }} \
        --host 0.0.0.0 \
        --port {{ vllm_port }} \
        --api-key ${VLLM_API_KEY} \
        --dtype auto \
        --max-model-len {{ model.max_model_len }} \
        --gpu-memory-utilization {{ vllm_gpu_memory_utilization }}{% if model.tool_call_parser is defined %} \
        --enable-auto-tool-choice \
        --tool-call-parser {{ model.tool_call_parser }}{% endif %} \
        > {{ vllm_home }}/vllm.log 2>&1 &
    echo "Started {{ model.name }} (pid $!). Run 'just logs' to follow."

{% endfor %}