models.yaml

# DXA-LM Model Configurations v0.2
# Designed for lightweight CPU inference on an 8 GB Intel Mac

default_model: baseline

models:
  baseline:
    id: baseline
    provider: local
    backend: baseline
    model_path: ""
    context_size: 4096
    temperature: 0.0
    max_tokens: 1024
    description: "Deterministic expert rule baseline engine (Zero-RAM overhead)"

  mock:
    id: mock
    provider: local
    backend: mock
    model_path: ""
    context_size: 2048
    temperature: 0.0
    max_tokens: 1024
    description: "Fast mock backend for unit testing and pipeline verification"

  dxa_tiny_0_5b:
    id: dxa_tiny_0_5b
    provider: local
    backend: llama_cpp
    model_path: "models/Qwen2.5-0.5B-Instruct-Q4_K_M.gguf"
    repo_id: "Qwen/Qwen2.5-0.5B-Instruct-GGUF"
    filename: "qwen2.5-0.5b-instruct-q4_k_m.gguf"
    download_url: "https://huggingface.co/Qwen/Qwen2.5-0.5B-Instruct-GGUF/resolve/main/qwen2.5-0.5b-instruct-q4_k_m.gguf"
    size_mb: 398
    context_size: 4096
    temperature: 0.0
    max_tokens: 1024
    threads: 4
    gpu_layers: 0
    description: "0.5B parameter quantized model for fast edge stack detection"

  dxa_core_1b:
    id: dxa_core_1b
    provider: local
    backend: llama_cpp
    model_path: "models/dxa-core-1b-q4_k_m.gguf"
    context_size: 4096
    temperature: 0.0
    max_tokens: 1024
    threads: 4
    gpu_layers: 0
    description: "1.0B parameter model for deployment planning and log parsing"

  dxa_expert_1_5b:
    id: dxa_expert_1_5b
    provider: local
    backend: llama_cpp
    model_path: "models/Qwen2.5-Coder-1.5B-Instruct-Q4_K_M.gguf"
    repo_id: "Qwen/Qwen2.5-Coder-1.5B-Instruct-GGUF"
    filename: "qwen2.5-coder-1.5b-instruct-q4_k_m.gguf"
    download_url: "https://huggingface.co/Qwen/Qwen2.5-Coder-1.5B-Instruct-GGUF/resolve/main/qwen2.5-coder-1.5b-instruct-q4_k_m.gguf"
    size_mb: 986
    context_size: 4096
    temperature: 0.0
    max_tokens: 1024
    threads: 4
    gpu_layers: 0
    description: "1.5B parameter specialized code model for multi-cause failure diagnosis"

  dxa_hybrid_1_5b:
    id: dxa_hybrid_1_5b
    provider: local
    backend: hybrid
    model_path: "models/Qwen2.5-Coder-1.5B-Instruct-Q4_K_M.gguf"
    base_llm: dxa_expert_1_5b
    context_size: 4096
    temperature: 0.0
    max_tokens: 1024
    threads: 4
    gpu_layers: 0
    description: "Hybrid architecture: Deterministic inspection/detection + LLM reasoning and repair"

  dxa_reasoner_3b:
    id: dxa_reasoner_3b
    provider: local
    backend: llama_cpp
    model_path: "models/dxa-reasoner-3b-q4_k_m.gguf"
    context_size: 8192
    temperature: 0.0
    max_tokens: 1024
    threads: 4
    gpu_layers: 0
    description: "3.0B parameter repair and configuration reconciliation model"