# llama-swap configuration for jersey detection testing
# ==================================================
# This configuration allows automatic model switching for testing
# different vision language models with the jersey detection test script.
#
# Usage:
#   llama-swap --config llama-swap-config.yaml --listen localhost:8080
#
# Then use the test script with --model-tag:
#   python test_jersey_detection.py ./images jersey_prompt.txt --model-tag "lfm2-vl-1.6b"
#
# llama-swap will automatically load the requested model and swap models
# as needed when you run tests with different --model-tag values.

models:
  # Small vision models (1-4B parameters)
  lfm2-vl-1.6b:
    name: "LiquidAI LFM2-VL 1.6B (F16)"
    cmd: llama-server --no-mmap -ngl 999 -fa on --host 0.0.0.0 --port ${PORT} -hf LiquidAI/LFM2-VL-1.6B-GGUF:F16

  gemma-3-4b:
    name: "Gemma 3 4B Instruct (F16)"
    cmd: llama-server --no-mmap -ngl 999 -fa on --host 0.0.0.0 --port ${PORT} -hf unsloth/gemma-3-4b-it-GGUF:F16

  kimi-vl-3b:
    name: "Kimi VL A3B Thinking (F16)"
    cmd: llama-server --no-mmap -ngl 999 -fa on --host 0.0.0.0 --port ${PORT} -hf ggml-org/Kimi-VL-A3B-Thinking-2506-GGUF:F16

  # Medium vision models (7-12B parameters)
  qwen2.5-vl-7b:
    name: "Qwen2.5-VL 7B Instruct (F16)"
    cmd: llama-server --no-mmap -ngl 999 -fa on --host 0.0.0.0 --port ${PORT} -hf unsloth/Qwen2.5-VL-7B-Instruct-GGUF:F16

  gemma-3-12b:
    name: "Gemma 3 12B Instruct (F16)"
    cmd: llama-server --no-mmap -ngl 999 -fa on --host 0.0.0.0 --port ${PORT} -hf unsloth/gemma-3-12b-it-GGUF:F16

  # Large models (24-27B parameters)
  mistral-small-24b-q8:
    name: "Mistral Small 3.2 24B Instruct (Q8_K_XL)"
    cmd: llama-server --no-mmap -ngl 999 -fa on --host 0.0.0.0 --port ${PORT} -hf unsloth/Mistral-Small-3.2-24B-Instruct-2506-GGUF:Q8_K_XL

  mistral-small-24b-q4:
    name: "Mistral Small 3.2 24B Instruct (Q4_K_XL)"
    cmd: llama-server --no-mmap -ngl 999 -fa on --host 0.0.0.0 --port ${PORT} -hf unsloth/Mistral-Small-3.2-24B-Instruct-2506-GGUF:Q4_K_XL

  gemma-3-27b:
    name: "Gemma 3 27B Instruct (Q8_0)"
    cmd: llama-server --no-mmap -ngl 999 -fa on --host 0.0.0.0 --port ${PORT} -hf unsloth/gemma-3-27b-it-GGUF:Q8_0

# Optional: Automatically unload models after 5 minutes of inactivity
# Uncomment to enable
# ttl: 300

# Optional: Preload a specific model on startup
# Uncomment to enable
# hooks:
#   onStartup:
#     - loadModel: qwen2.5-vl-7b