{
  "model_info": {
    "name": "RuvLTRA-Small-0.5B",
    "version": "1.0.0",
    "description": "ANE-optimized small language model based on Qwen2.5-0.5B-Instruct for edge deployment",
    "base_model": "Qwen/Qwen2.5-0.5B-Instruct",
    "license": "Apache-2.0",
    "created": "2026-01-19",
    "author": "RuvVector Team"
  },
  "architecture": {
    "model_type": "qwen2",
    "architecture_class": "Qwen2ForCausalLM",
    "hidden_size": 896,
    "intermediate_size": 4864,
    "num_hidden_layers": 24,
    "num_attention_heads": 14,
    "num_key_value_heads": 2,
    "vocab_size": 151936,
    "max_position_embeddings": 32768,
    "rope_theta": 1000000.0,
    "rms_norm_eps": 1e-6,
    "hidden_act": "silu",
    "attention_dropout": 0.0,
    "tie_word_embeddings": true,
    "use_sliding_window": false,
    "sliding_window": 32768,
    "max_window_layers": 21,
    "bos_token_id": 151643,
    "eos_token_id": 151645
  },
  "parameters": {
    "total": "0.49B",
    "total_exact": 494000000,
    "non_embedding": "0.36B",
    "non_embedding_exact": 360000000
  },
  "features": {
    "attention_mechanism": "grouped_query_attention",
    "positional_encoding": "rotary_position_embedding",
    "activation_function": "swiglu",
    "normalization": "rmsnorm",
    "has_qkv_bias": true,
    "multilingual": true,
    "supported_languages": 29
  },
  "optimizations": {
    "ane_optimized": true,
    "sona_enabled": true,
    "flash_attention_compatible": true,
    "continuous_batching": true,
    "speculative_decoding_ready": true,
    "target_hardware": [
      "apple_neural_engine",
      "metal_gpu",
      "cpu_arm64",
      "cpu_x86_64"
    ],
    "memory_optimizations": [
      "kv_cache_compression",
      "activation_checkpointing",
      "weight_sharing"
    ]
  },
  "quantization": {
    "recommended": "Q4_K_M",
    "targets": {
      "Q4_K_M": {
        "description": "4-bit quantization with K-means clustering (medium)",
        "size_mb": 491,
        "quality": "good",
        "speed": "fast",
        "memory_reduction": "75%",
        "recommended_for": ["mobile", "edge", "resource_constrained"]
      },
      "Q5_K_M": {
        "description": "5-bit quantization with K-means clustering (medium)",
        "size_mb": 522,
        "quality": "better",
        "speed": "fast",
        "memory_reduction": "68%",
        "recommended_for": ["balanced", "quality_conscious"]
      },
      "Q8_0": {
        "description": "8-bit quantization (round-to-nearest)",
        "size_mb": 676,
        "quality": "best",
        "speed": "moderate",
        "memory_reduction": "50%",
        "recommended_for": ["accuracy_critical", "development"]
      }
    },
    "additional_quantizations": {
      "Q2_K": {"size_mb": 415, "quality": "acceptable"},
      "Q3_K_M": {"size_mb": 432, "quality": "fair"},
      "Q4_0": {"size_mb": 429, "quality": "good"},
      "Q5_0": {"size_mb": 490, "quality": "better"},
      "Q6_K": {"size_mb": 650, "quality": "very_good"}
    }
  },
  "download_urls": {
    "official_gguf": {
      "base_url": "https://huggingface.co/Qwen/Qwen2.5-0.5B-Instruct-GGUF",
      "files": {
        "Q4_K_M": "https://huggingface.co/Qwen/Qwen2.5-0.5B-Instruct-GGUF/resolve/main/qwen2.5-0.5b-instruct-q4_k_m.gguf",
        "Q5_K_M": "https://huggingface.co/Qwen/Qwen2.5-0.5B-Instruct-GGUF/resolve/main/qwen2.5-0.5b-instruct-q5_k_m.gguf",
        "Q8_0": "https://huggingface.co/Qwen/Qwen2.5-0.5B-Instruct-GGUF/resolve/main/qwen2.5-0.5b-instruct-q8_0.gguf"
      }
    },
    "safetensors": {
      "base_url": "https://huggingface.co/Qwen/Qwen2.5-0.5B-Instruct",
      "format": "safetensors",
      "dtype": "bfloat16"
    },
    "third_party_gguf": {
      "bartowski": "https://huggingface.co/bartowski/Qwen2.5-0.5B-Instruct-GGUF",
      "tensorblock": "https://huggingface.co/tensorblock/Qwen2.5-0.5B-GGUF"
    }
  },
  "cli_download_commands": {
    "Q4_K_M": "huggingface-cli download Qwen/Qwen2.5-0.5B-Instruct-GGUF qwen2.5-0.5b-instruct-q4_k_m.gguf --local-dir . --local-dir-use-symlinks False",
    "Q5_K_M": "huggingface-cli download Qwen/Qwen2.5-0.5B-Instruct-GGUF qwen2.5-0.5b-instruct-q5_k_m.gguf --local-dir . --local-dir-use-symlinks False",
    "Q8_0": "huggingface-cli download Qwen/Qwen2.5-0.5B-Instruct-GGUF qwen2.5-0.5b-instruct-q8_0.gguf --local-dir . --local-dir-use-symlinks False"
  },
  "performance_targets": {
    "inference_latency_ms": {
      "first_token": 50,
      "per_token": 15
    },
    "throughput_tokens_per_sec": {
      "ane": 150,
      "metal_gpu": 120,
      "cpu": 40
    },
    "memory_usage_mb": {
      "Q4_K_M": 600,
      "Q5_K_M": 650,
      "Q8_0": 800
    },
    "context_window": {
      "default": 8192,
      "max": 32768
    }
  },
  "ruvllm_integration": {
    "loader": "gguf",
    "backend_priority": ["ane", "metal", "cpu"],
    "tokenizer": "qwen2",
    "chat_template": "qwen2_instruct",
    "system_prompt_support": true,
    "function_calling": false,
    "streaming": true
  },
  "benchmarks": {
    "mmlu": "pending",
    "humaneval": "pending",
    "gsm8k": "pending",
    "arc_challenge": "pending"
  }
}