jang_config.json

{
  "version": 2,
  "weight_format": "mxtq",
  "profile": "JANGTQ2",
  "source_model": {
    "name": "Ling-2.6-flash",
    "org": "inclusionAI",
    "architecture": "bailing_hybrid"
  },
  "mxtq_seed": 42,
  "mxtq_bits": {
    "routed_expert": 2,
    "attention": 8,
    "shared_expert": 8,
    "dense_mlp": 8,
    "embed_tokens": 8,
    "lm_head": 8,
    "mtp_eh_proj": 8,
    "norms_router_biases": 16
  },
  "quantization": {
    "method": "affine+mxtq",
    "group_size": 64,
    "bits_default": 2
  },
  "capabilities": {
    "reasoning_parser": "deepseek_r1",
    "tool_parser": "deepseek",
    "think_in_template": false,
    "supports_tools": true,
    "supports_thinking": true,
    "family": "bailing_hybrid",
    "modality": "text",
    "cache_type": "hybrid"
  },
  "routed_expert_layout": "prestacked"
}