# Model registry configuration
# Centralized registry of available models and their corresponding configuration files
#
# "reasoning_supported=true" means that there's a another llm yaml file with the same name but addidional "_think.yaml" suffix,
# which is tested to work as expected, otherwise you need to mamually modify the llm yaml to enable their reasoning mode.

llm_models:
  "nvidia/NVIDIA-Nemotron-Nano-9B-v2":
    yaml_id: "nemotron_nano_v2.yaml"
    reasoning_supported: false
    
  "meta-llama/Llama-3.1-8B-Instruct":
    yaml_id: "llama3.1-8B-instruct.yaml"
    reasoning_supported: false

  "Qwen/Qwen2.5-7B-Instruct":
    yaml_id: "qwen2.5-7B.yaml"
    reasoning_supported: false
    
  "Qwen/Qwen3-8B":
    yaml_id: "qwen3-8B.yaml"
    reasoning_supported: true

  "hf_llm_generic":
    yaml_id: "hf_llm_generic.yaml"
    reasoning_supported: false

tts_models:
  "fastpitch-hifigan":
    yaml_id: "nemo_fastpitch-hifigan.yaml"

  "hexgrad/Kokoro-82M":
    yaml_id: "kokoro_82M.yaml"

stt_models:
  "stt_en_fastconformer_hybrid_large_streaming_80ms":
    yaml_id: "nemo_cache_aware_streaming.yaml"
    type: "nemo"