jang_config.json

{
  "version": 2,
  "weight_format": "mxtq",
  "profile": "JANGTQ_K",
  "source_model": {
    "name": "MiniMax-M2.7",
    "org": "MiniMaxAI",
    "architecture": "minimax_m2"
  },
  "mxtq_seed": 42,
  "mxtq_bits": {
    "routed_expert": {
      "gate_proj": 2,
      "down_proj": 4,
      "up_proj": 2
    },
    "attention": 8,
    "shared_expert": 8,
    "embed_tokens": 8,
    "lm_head": 8,
    "norms_router_biases": 16
  },
  "quantization": {
    "method": "affine+mxtq",
    "group_size": 64,
    "bits_default": 4
  },
  "capabilities": {
    "reasoning_parser": "qwen3",
    "tool_parser": "minimax",
    "think_in_template": true,
    "supports_tools": true,
    "supports_thinking": true,
    "family": "minimax_m2",
    "modality": "text",
    "cache_type": "kv"
  },
  "routed_expert_layout": "prestacked"
}