{
  "model": "ormandj/GLM-5.3-Flash-W4A16-NVFP4-K32-Experts-FP8-WO",
  "revision": "c3cbb9891b67c741bcbf6b176dd7af9265b069db",
  "image_digest": "sha256:ec4243f940a179a27fea21895077efd47cd050501f99a1d2a5fecf7df2e7be71",
  "runtime": "v0.4.3",
  "served_model": "glm-5.3-flash",
  "hardware": "2x RTX PRO 6000 Blackwell Max-Q 96 GB",
  "platform": "native Linux",
  "tp": 2,
  "shared_pool": 524288,
  "context_limit": 524288,
  "concurrency": 4,
  "mamba_slots": 28,
  "prefill": 4096,
  "kv": "fp8_e4m3",
  "recurrent": "bfloat16",
  "speculation": "adaptive EAGLE [3,5]",
  "hicache": false,
  "normal_output_limit": 4096,
  "template_source_sha256": "34d5ee66b12fa6446cdae131c352b8f68cd85369e0e6fda115583805fada3891",
  "template_derived_sha256": "3406cc91800b56c06439e65aaea1929e6bc803cdb62f7f995f25ab082efd3408",
  "ipc": "host",
  "initial_qualification_ipc": "private",
  "docker_shm_size": "64g",
  "effective_host_shm_gib_approx": 45,
  "recipe": "recipe.toml",
  "recipe_digest": "d188c858df5475321408353214083720af7ea226c8803ee86f6ad5fbd27e4975",
  "registry_digest": "07296e4c33dd4bb9d970cddf26d2ac488651c36596422dd0d939d279422de0c4"
}