{
  "schema_version": "k3_deb_e2e_chat_summary.v1",
  "base": "http://192.168.100.233:8090",
  "chat": {
    "http": 200,
    "body": {
      "admission_trace": {
        "admission": "accepted",
        "api_surface": "scheduler",
        "deadline_ms": 30000,
        "estimated_runtime_ms": 3200,
        "eta_ms": 3200,
        "model": "llm-chat",
        "normalized_task": "kb.chat.complete",
        "priority": 96,
        "queue_capacity": 8,
        "queue_depth": 0,
        "reason": "sync_admitted",
        "resource_domain": "ACCEL.chat",
        "resource_key": "ACCEL-SVC",
        "resource_queue_capacity": 8,
        "resource_queue_depth": 0,
        "resource_queue_oldest_wait_ms": 0,
        "route": "POST /kb/tasks/kb.chat.complete",
        "runtime_estimate_samples": 0,
        "runtime_estimate_source": "static",
        "scheduled_as": "sync",
        "schema_version": "openai_scheduler_unified_admission.v1",
        "service_class": "realtime_answer"
      },
      "answer_profile": "normal_answer",
      "backend": "llama_http",
      "cache_hit": false,
      "cache_hit_source": "scheduler_prefix_hint",
      "cache_key": "fnv1a64:8c7e906dfb71cf8a",
      "cache_revision": 1,
      "cached_tokens": 0,
      "cold_start_wait_ms": 0.0,
      "decode_ms": 678.301,
      "device_used": "ACCEL",
      "finish_reason": "stop",
      "latency_ms": 1845.97838,
      "max_output_tokens": 96,
      "model": "llm-chat",
      "model_residency": "hot_resident",
      "output_tokens": 23,
      "outputs": {
        "choices": [
          {
            "finish_reason": "stop",
            "index": 0,
            "message": {
              "content": "当前测试配置中，llm-chat 使用 MiniCPM5-1B Q4_K_M。",
              "role": "assistant"
            }
          }
        ],
        "created": 1786497190,
        "id": "chatcmpl-sC3NtIIpWhspXDMtghMh97uuT89UAg6l",
        "model": "MiniCPM5-1B-Q4_K_M.gguf",
        "object": "chat.completion",
        "system_fingerprint": "b9488-21379f9d7",
        "timings": {
          "cache_n": 0,
          "predicted_ms": 678.301,
          "predicted_n": 23,
          "predicted_per_second": 33.90825017212123,
          "predicted_per_token_ms": 29.491347826086958,
          "prompt_ms": 1148.512,
          "prompt_n": 274,
          "prompt_per_second": 238.56955782786773,
          "prompt_per_token_ms": 4.191649635036496
        },
        "usage": {
          "completion_tokens": 23,
          "prompt_tokens": 274,
          "prompt_tokens_details": {
            "cached_tokens": 0
          },
          "total_tokens": 297
        }
      },
      "prefill_ms": 1148.512,
      "prompt_cache_key": "fnv1a64:8c7e906dfb71cf8a",
      "prompt_tokens": 274,
      "queue_wait_ms": 200.325794,
      "scheduled_as": "sync",
      "schema_version": "kb_task.v1",
      "service_class": "realtime_answer",
      "startup_state": "hot_resident",
      "startup_wait_ms": 0.357272,
      "task": "kb.chat.complete",
      "telemetry": {
        "answer_profile": "normal_answer",
        "backend": "llama_http",
        "cache_hit": false,
        "cache_hit_source": "scheduler_prefix_hint",
        "cache_key": "fnv1a64:8c7e906dfb71cf8a",
        "cache_revision": 1,
        "cached_tokens": 0,
        "cold_start_wait_ms": 0.0,
        "decode_ms": 678.301,
        "finish_reason": "stop",
        "latency_ms": 1845.97838,
        "max_output_tokens": 96,
        "model_residency": "hot_resident",
        "output_tokens": 23,
        "prefill_ms": 1148.512,
        "prompt_cache_key": "fnv1a64:8c7e906dfb71cf8a",
        "prompt_tokens": 274,
        "queue_ms": 200.325794,
        "startup_state": "hot_resident",
        "startup_wait_ms": 0.357272,
        "trace_id": "k3-deb-e2e-chat-minicpm",
        "truncated": false,
        "worker_id": "llm-chat@ACCEL:3358660",
        "worker_pid": 3358660
      },
      "trace_id": "k3-deb-e2e-chat-minicpm",
      "truncated": false,
      "worker_id": "llm-chat@ACCEL:3358660",
      "worker_pid": 3358660
    }
  },
  "summary_submit": {
    "http": 202,
    "body": {
      "admission_trace": {
        "admission": "accepted",
        "api_surface": "scheduler",
        "deadline_ms": 20000,
        "estimated_runtime_ms": 15000,
        "eta_ms": 85000,
        "model": "llm-summary",
        "normalized_task": "kb.document.summary",
        "priority": 55,
        "queue_capacity": 8,
        "queue_depth": 0,
        "reason": "explicit_async",
        "resource_domain": "ACCEL.aux.session",
        "resource_key": "ACCEL-SVC",
        "resource_queue_capacity": 8,
        "resource_queue_depth": 0,
        "resource_queue_oldest_wait_ms": 0,
        "route": "POST /kb/tasks/kb.document.summary:async",
        "runtime_estimate_samples": 0,
        "runtime_estimate_source": "static",
        "scheduled_as": "async",
        "schema_version": "openai_scheduler_unified_admission.v1",
        "service_class": "user_async"
      },
      "eta_ms": 85000,
      "job_id": "job_ba57c5d23a5d607cfa9284976ad446e",
      "model": "llm-summary",
      "reason": "explicit_async",
      "scheduled_as": "async",
      "schema_version": "kb_task.v1",
      "service_class": "user_async",
      "status": "queued",
      "task": "kb.document.summary",
      "trace_id": "k3-deb-e2e-summary-qwen30"
    }
  },
  "summary_final": {
    "backend": "llama_http",
    "cache_hit": false,
    "cache_hit_source": "scheduler_prefix_hint",
    "cache_key": "fnv1a64:3acd825aa8463f7e",
    "cache_revision": 2,
    "cached_tokens": 0,
    "capability_revision": 62,
    "cold_start_wait_ms": 103940.282664,
    "contract_revision": 62,
    "contract_version": "edge-scheduler-v2",
    "decode_ms": 8643.886,
    "device_used": "ACCEL",
    "finish_reason": "length",
    "job_id": "job_ba57c5d23a5d607cfa9284976ad446e",
    "latency_ms": 10949.090934,
    "max_output_tokens": 96,
    "model": "llm-summary",
    "model_residency": "cold_start",
    "model_revision": 62,
    "output_tokens": 96,
    "outputs": {
      "choices": [
        {
          "finish_reason": "length",
          "index": 0,
          "message": {
            "content": "**摘要：**\n\n本技术文档描述了 K3 调度器的 deb e2e 配置，主要内容包括：\n\n- 将 `llm-chat` 绑定至 **MiniCPM5-1B Q4_K_M** 模型；\n- 将 `llm-summary` 绑定至 **Qwen3-30B-A3B Q4_0** 模型；\n- `llm-summary` ",
            "role": "assistant"
          }
        }
      ],
      "created": 1786497306,
      "id": "chatcmpl-VXBYmjPgyehjaR0oxQOOPuoHlzn9JDRH",
      "model": "Qwen3-30B-A3B-Q4_0.gguf",
      "object": "chat.completion",
      "system_fingerprint": "b9488-21379f9d7",
      "timings": {
        "cache_n": 0,
        "predicted_ms": 8643.886,
        "predicted_n": 96,
        "predicted_per_second": 11.106115929802868,
        "predicted_per_token_ms": 90.04047916666667,
        "prompt_ms": 2270.652,
        "prompt_n": 103,
        "prompt_per_second": 45.3614204202141,
        "prompt_per_token_ms": 22.04516504854369
      },
      "usage": {
        "completion_tokens": 96,
        "prompt_tokens": 103,
        "prompt_tokens_details": {
          "cached_tokens": 0
        },
        "total_tokens": 199
      }
    },
    "phase": "done",
    "prefill_ms": 2270.652,
    "prompt_cache_key": "fnv1a64:3acd825aa8463f7e",
    "prompt_tokens": 103,
    "queue_wait_ms": 0.0,
    "reason": "explicit_async",
    "recoverable": false,
    "revision": 62,
    "scheduled_as": "async",
    "schema_version": "job_status.v2",
    "service_class": "user_async",
    "source": "kb_task",
    "startup_state": "cold_start",
    "startup_wait_ms": 103940.282664,
    "status": "done",
    "task": "kb.document.summary",
    "telemetry": {
      "backend": "llama_http",
      "cache_hit": false,
      "cache_hit_source": "scheduler_prefix_hint",
      "cache_key": "fnv1a64:3acd825aa8463f7e",
      "cache_revision": 2,
      "cached_tokens": 0,
      "cold_start_wait_ms": 103940.282664,
      "decode_ms": 8643.886,
      "finish_reason": "length",
      "latency_ms": 10949.090934,
      "max_output_tokens": 96,
      "model_residency": "cold_start",
      "output_tokens": 96,
      "prefill_ms": 2270.652,
      "prompt_cache_key": "fnv1a64:3acd825aa8463f7e",
      "prompt_tokens": 103,
      "queue_ms": 0.0,
      "startup_state": "cold_start",
      "startup_wait_ms": 103940.282664,
      "trace_id": "k3-deb-e2e-summary-qwen30",
      "truncated": true,
      "truncation_reason": "length",
      "worker_id": "llm-summary@ACCEL:3359667",
      "worker_pid": 3359667
    },
    "terminal_age_ms": 2632.795703,
    "terminal_sequence": 2,
    "trace_id": "k3-deb-e2e-summary-qwen30",
    "truncated": true,
    "truncation_reason": "length",
    "ttl_ms": 900000,
    "worker_id": "llm-summary@ACCEL:3359667",
    "worker_pid": 3359667
  },
  "post_summary_recovered_models": {
    "capability_revision": 72,
    "contract_revision": 72,
    "contract_version": "edge-scheduler-v2",
    "model_revision": 72,
    "models": [
      {
        "adapter_ref": "llama_http",
        "artifact": {
          "availability": "local",
          "path": "/root/models/spacemit-ai/llm/Qwen3-30B-A3B-Q4_0.gguf",
          "quantization": "Q4_0",
          "runtime": "llama_http",
          "status": "present"
        },
        "backend": "RVV/IME",
        "backend_profile": {
          "accelerator_path": "ime2",
          "api": "/v1/chat/completions",
          "artifact_source": "spacemit_model_zoo",
          "model_family": "Qwen3-30B-A3B",
          "quantization": "Q4_0",
          "requires_no_think": true,
          "role": "document_summary_and_long_context_organization",
          "runtime": "llama.cpp-tools-spacemit",
          "runtime_binary": "/usr/bin/llama-server"
        },
        "cold_start_eta_ms": 70000,
        "context_window_tokens": 8192,
        "devices": {
          "ACCEL": {
            "in_flight": 0,
            "role": "primary",
            "status": "cold"
          }
        },
        "dispatchable": "DISPLACED",
        "embedding_dim": 0,
        "estimated_runtime_ms": 15000,
        "eta_ms": 0,
        "exclusive": true,
        "hot_state": "cold",
        "last_latency_ms": 11149.048622,
        "last_success_ts": "2026-08-12T01:15:06Z",
        "last_successful_inference_ts": "2026-08-12T01:15:06Z",
        "last_worker_error": "",
        "last_worker_error_reason": "",
        "last_worker_error_ts": "",
        "lifecycle": "COLD",
        "max_output_tokens": 1024,
        "memory_watermark": {
          "kv_cache_per_1k_tokens_gb": 0.75,
          "per_worker_dram_gb": 18.0,
          "resident_model_gb": 18.0
        },
        "model_residency": "cold",
        "name": "llm-summary",
        "p50_latency_ms": 11149.048622,
        "quality_profile": {},
        "quantization": "Q4_0",
        "quarantine_scope": "none",
        "queue_capacity": 8,
        "queue_depth": 0,
        "replicas": 1,
        "resource_domain": "ACCEL.aux.session",
        "resource_key": "ACCEL-SVC",
        "resource_lease_amount": 1,
        "resource_lease_type": "exclusive_session",
        "runtime_adapter": "llama_http",
        "runtime_status": {
          "adapter_ref": "llama_http",
          "last_successful_inference_ts": "2026-08-12T01:15:06Z",
          "last_worker_error_reason": "",
          "lifecycle": "COLD",
          "ready": false,
          "worker_kind": "llama_http"
        },
        "service_class": "realtime_answer",
        "state": "UNAVAILABLE",
        "state_revision": 68,
        "worker_kind": "llama_http"
      },
      {
        "adapter_ref": "llama_http",
        "artifact": {
          "availability": "local",
          "path": "/root/models/spacemit-ai/llm/MiniCPM5-1B/MiniCPM5-1B-Q4_K_M.gguf",
          "quantization": "Q4_K_M",
          "runtime": "llama_http",
          "status": "present"
        },
        "backend": "RVV/IME",
        "backend_profile": {
          "accelerator_path": "ime2",
          "api": "/v1/chat/completions",
          "artifact_source": "vlm-llm-benchmark:minicpm-probe",
          "bench_evidence": "docs/minicpm-k3-rag-replacement.zh.md",
          "chat_template_kwargs": {
            "enable_thinking": false
          },
          "model_family": "MiniCPM5-1B",
          "quantization": "Q4_K_M",
          "recommended_stop": [
            "\n\n",
            "请注意",
            "注：",
            "停止。",
            "以上。"
          ],
          "requires_no_think": true,
          "role": "rag_answer_candidate",
          "runtime": "llama.cpp-tools-spacemit",
          "runtime_binary": "/usr/bin/llama-server"
        },
        "cold_start_eta_ms": 15000,
        "context_window_tokens": 8192,
        "devices": {
          "ACCEL": {
            "in_flight": 0,
            "role": "primary",
            "status": "ready"
          }
        },
        "dispatchable": "FREE",
        "embedding_dim": 0,
        "estimated_runtime_ms": 3200,
        "eta_ms": 0,
        "exclusive": true,
        "hot_state": "resident",
        "last_latency_ms": 1846.55135,
        "last_success_ts": "2026-08-12T01:13:10Z",
        "last_successful_inference_ts": "2026-08-12T01:13:10Z",
        "last_worker_error": "",
        "last_worker_error_reason": "",
        "last_worker_error_ts": "",
        "lifecycle": "READY",
        "max_output_tokens": 1024,
        "memory_watermark": {
          "kv_cache_per_1k_tokens_gb": 0.08,
          "per_worker_dram_gb": 1.2,
          "resident_model_gb": 1.2
        },
        "model_residency": "resident",
        "name": "llm-chat",
        "p50_latency_ms": 1846.55135,
        "quality_profile": {},
        "quantization": "Q4_K_M",
        "quarantine_scope": "none",
        "queue_capacity": 8,
        "queue_depth": 0,
        "replicas": 1,
        "resource_domain": "ACCEL.chat",
        "resource_key": "ACCEL-SVC",
        "resource_lease_amount": 1,
        "resource_lease_type": "exclusive_session",
        "runtime_adapter": "llama_http",
        "runtime_status": {
          "adapter_ref": "llama_http",
          "last_successful_inference_ts": "2026-08-12T01:13:10Z",
          "last_worker_error_reason": "",
          "lifecycle": "READY",
          "ready": true,
          "worker_kind": "llama_http"
        },
        "service_class": "realtime_answer",
        "state": "READY_FAST",
        "state_revision": 72,
        "worker_kind": "llama_http"
      },
      {
        "adapter_ref": "llama_http",
        "artifact": {
          "availability": "local",
          "path": "/root/models/spacemit-ai/rerank/Bge-Reranker-V2-M3-Q4_0.gguf",
          "quantization": "Q4_0",
          "runtime": "llama_http",
          "status": "present"
        },
        "backend": "RVV/IME",
        "backend_profile": {
          "accelerator_path": "ime2",
          "api": "/v1/rerank",
          "artifact_source": "spacemit_model_zoo",
          "embedding_dim": 1024,
          "pooling": "rank",
          "quantization": "Q4_0",
          "runtime": "llama.cpp-tools-spacemit",
          "runtime_binary": "/usr/bin/llama-server"
        },
        "cold_start_eta_ms": 8000,
        "context_window_tokens": 0,
        "devices": {
          "ACCEL": {
            "in_flight": 0,
            "role": "primary",
            "status": "cold"
          }
        },
        "dispatchable": "DISPLACED",
        "embedding_dim": 1024,
        "estimated_runtime_ms": 145,
        "eta_ms": 0,
        "exclusive": false,
        "hot_state": "cold",
        "last_successful_inference_ts": "",
        "last_worker_error": "",
        "last_worker_error_reason": "",
        "last_worker_error_ts": "",
        "lifecycle": "COLD",
        "max_output_tokens": 0,
        "memory_watermark": {
          "kv_cache_per_1k_tokens_gb": 0.0,
          "per_worker_dram_gb": 0.8,
          "resident_model_gb": 0.8
        },
        "model_residency": "cold",
        "name": "reranker-int8",
        "quality_profile": {},
        "quantization": "Q4_0",
        "quarantine_scope": "none",
        "queue_capacity": 16,
        "queue_depth": 0,
        "replicas": 1,
        "resource_domain": "ACCEL.aux",
        "resource_key": "ACCEL-BATCH",
        "resource_lease_amount": 1,
        "resource_lease_type": "counting",
        "runtime_adapter": "llama_http",
        "runtime_status": {
          "adapter_ref": "llama_http",
          "last_successful_inference_ts": "",
          "last_worker_error_reason": "",
          "lifecycle": "COLD",
          "ready": false,
          "worker_kind": "llama_http"
        },
        "service_class": "realtime_retrieval",
        "state": "UNAVAILABLE",
        "state_revision": 0,
        "worker_kind": "llama_http"
      },
      {
        "adapter_ref": "llama_http",
        "artifact": {
          "availability": "local",
          "path": "/root/models/spacemit-ai/embed/Bge-Small-Zh-V1.5-Q4_K_M.gguf",
          "quantization": "Q4_K_M",
          "runtime": "llama_http",
          "status": "present"
        },
        "backend": "RVV/IME",
        "backend_profile": {
          "accelerator_path": "ime2",
          "api": "/v1/embeddings",
          "artifact_source": "spacemit_model_zoo",
          "embedding_dim": 1024,
          "pooling": "mean",
          "quantization": "Q4_K_M",
          "runtime": "llama.cpp-tools-spacemit",
          "runtime_binary": "/usr/bin/llama-server"
        },
        "cold_start_eta_ms": 8000,
        "context_window_tokens": 0,
        "devices": {
          "ACCEL": {
            "in_flight": 0,
            "role": "primary",
            "status": "ready"
          }
        },
        "dispatchable": "FREE",
        "embedding_dim": 1024,
        "estimated_runtime_ms": 25,
        "eta_ms": 0,
        "exclusive": false,
        "hot_state": "resident",
        "last_latency_ms": 9.527033,
        "last_success_ts": "2026-08-12T01:13:07Z",
        "last_successful_inference_ts": "2026-08-12T01:13:07Z",
        "last_worker_error": "",
        "last_worker_error_reason": "",
        "last_worker_error_ts": "",
        "lifecycle": "READY",
        "max_output_tokens": 0,
        "memory_watermark": {
          "kv_cache_per_1k_tokens_gb": 0.0,
          "per_worker_dram_gb": 0.7,
          "resident_model_gb": 2.8
        },
        "model_residency": "resident",
        "name": "embedding-int8",
        "p50_latency_ms": 9.527033,
        "quality_profile": {},
        "quantization": "Q4_K_M",
        "quarantine_scope": "none",
        "queue_capacity": 32,
        "queue_depth": 0,
        "replicas": 4,
        "resource_domain": "ACCEL.retrieval",
        "resource_key": "ACCEL-RETRIEVAL",
        "resource_lease_amount": 1,
        "resource_lease_type": "counting",
        "runtime_adapter": "llama_http",
        "runtime_status": {
          "adapter_ref": "llama_http",
          "last_successful_inference_ts": "2026-08-12T01:13:07Z",
          "last_worker_error_reason": "",
          "lifecycle": "READY",
          "ready": true,
          "worker_kind": "llama_http"
        },
        "service_class": "realtime_retrieval",
        "state": "READY_FAST",
        "state_revision": 22,
        "worker_kind": "llama_http"
      },
      {
        "adapter_ref": "openai_http",
        "artifact": {
          "availability": "external",
          "path": "sherpa-onnx-matcha-tts",
          "runtime": "openai_http",
          "status": "unknown"
        },
        "backend": "RVV/IME",
        "backend_profile": {
          "adapter": "openai_http",
          "api": "/v1/audio/speech",
          "artifact_source": "image_preinstalled",
          "legacy_contract_model": "/root/models/spacemit-ai/tts/synthetic-token-ids-v1.onnx",
          "model_family": "sherpa-onnx-matcha-tts",
          "not_ready_code": "native_tts_model_unavailable",
          "quality_scope": "native product TTS PCM16 WAV generation",
          "required_model_dir": "/opt/rv-sherpa-onnx/models/candidates/current/tts-matcha",
          "runtime": "rv-sherpa-onnx-runtime"
        },
        "cold_start_eta_ms": 7000,
        "context_window_tokens": 0,
        "devices": {
          "CPU": {
            "in_flight": 0,
            "role": "primary",
            "status": "cold"
          }
        },
        "dispatchable": "DISPLACED",
        "embedding_dim": 0,
        "estimated_runtime_ms": 800,
        "eta_ms": 0,
        "exclusive": true,
        "hot_state": "cold",
        "last_successful_inference_ts": "",
        "last_worker_error": "",
        "last_worker_error_reason": "",
        "last_worker_error_ts": "",
        "lifecycle": "COLD",
        "max_output_tokens": 0,
        "memory_watermark": {
          "kv_cache_per_1k_tokens_gb": 0.0,
          "per_worker_dram_gb": 0.5,
          "resident_model_gb": 0.5
        },
        "model_residency": "cold",
        "name": "tts-default",
        "quality_profile": {},
        "quantization": "unknown",
        "quarantine_scope": "none",
        "queue_capacity": 4,
        "queue_depth": 0,
        "replicas": 1,
        "resource_domain": "CPU",
        "resource_key": "CPU",
        "resource_lease_amount": 2,
        "resource_lease_type": "counting",
        "runtime_adapter": "openai_http",
        "runtime_status": {
          "adapter_ref": "openai_http",
          "last_successful_inference_ts": "",
          "last_worker_error_reason": "",
          "lifecycle": "COLD",
          "ready": false,
          "worker_kind": "openai_http"
        },
        "service_class": "realtime_cpu",
        "state": "UNAVAILABLE",
        "state_revision": 0,
        "worker_kind": "openai_http"
      },
      {
        "adapter_ref": "openai_http",
        "artifact": {
          "availability": "external",
          "path": "sherpa-onnx-streaming-zipformer",
          "runtime": "openai_http",
          "status": "unknown"
        },
        "backend": "RVV/IME",
        "backend_profile": {
          "accepts": [
            "audio_base64",
            "chunk_mel"
          ],
          "adapter": "openai_http",
          "api": "/v1/audio/transcriptions",
          "artifact_source": "image_preinstalled",
          "legacy_frontend_probe": "/root/models/spacemit-ai/vlm/qwen3-asr-0.6B/Qwen3-ASR-0.6B-encoder-frontend.dynq.onnx",
          "model_dir": "/opt/rv-sherpa-onnx/models/candidates/current/asr-streaming-zipformer",
          "model_family": "sherpa-onnx-streaming-zipformer-bilingual-zh-en",
          "runtime": "rv-sherpa-onnx-runtime",
          "stage": "native_audio_transcription"
        },
        "cold_start_eta_ms": 0,
        "context_window_tokens": 0,
        "devices": {
          "CPU": {
            "in_flight": 0,
            "role": "primary",
            "status": "cold"
          }
        },
        "dispatchable": "DISPLACED",
        "embedding_dim": 0,
        "estimated_runtime_ms": 500,
        "eta_ms": 0,
        "exclusive": true,
        "hot_state": "cold",
        "last_successful_inference_ts": "",
        "last_worker_error": "",
        "last_worker_error_reason": "",
        "last_worker_error_ts": "",
        "lifecycle": "COLD",
        "max_output_tokens": 0,
        "memory_watermark": {
          "kv_cache_per_1k_tokens_gb": 0.0,
          "per_worker_dram_gb": 0.4,
          "resident_model_gb": 0.4
        },
        "model_residency": "cold",
        "name": "asr-whisper",
        "quality_profile": {},
        "quantization": "unknown",
        "quarantine_scope": "none",
        "queue_capacity": 24,
        "queue_depth": 0,
        "replicas": 1,
        "resource_domain": "CPU",
        "resource_key": "CPU",
        "resource_lease_amount": 4,
        "resource_lease_type": "counting",
        "runtime_adapter": "openai_http",
        "runtime_status": {
          "adapter_ref": "openai_http",
          "last_successful_inference_ts": "",
          "last_worker_error_reason": "",
          "lifecycle": "COLD",
          "ready": false,
          "worker_kind": "openai_http"
        },
        "service_class": "realtime_cpu",
        "state": "UNAVAILABLE",
        "state_revision": 0,
        "worker_kind": "openai_http"
      },
      {
        "adapter_ref": "llama_http",
        "artifact": {
          "availability": "local",
          "path": "/root/models/spacemit-ai/vlm/Qwen3.5-2B/qwen3_5_2b-text-q41.gguf",
          "quantization": "Q4_1",
          "runtime": "llama_http",
          "status": "present"
        },
        "backend": "RVV/IME",
        "backend_profile": {
          "accelerator_path": "ime2+smt",
          "api": "/v1/chat/completions",
          "artifact_source": "spacemit_model_zoo",
          "media_backend": "smt",
          "model_family": "Qwen3.5-2B SMT",
          "quantization": "Q4_1",
          "requires_no_think": true,
          "runtime": "llama.cpp-tools-spacemit",
          "runtime_binary": "/usr/bin/llama-server",
          "smt_config_dir": "/root/models/spacemit-ai/vlm/Qwen3.5-2B"
        },
        "cold_start_eta_ms": 25000,
        "context_window_tokens": 4096,
        "devices": {
          "ACCEL": {
            "in_flight": 0,
            "role": "primary",
            "status": "cold"
          }
        },
        "dispatchable": "DISPLACED",
        "embedding_dim": 0,
        "estimated_runtime_ms": 11000,
        "eta_ms": 0,
        "exclusive": true,
        "hot_state": "cold",
        "last_successful_inference_ts": "",
        "last_worker_error": "",
        "last_worker_error_reason": "",
        "last_worker_error_ts": "",
        "lifecycle": "COLD",
        "max_output_tokens": 1024,
        "memory_watermark": {
          "kv_cache_per_1k_tokens_gb": 0.125,
          "per_worker_dram_gb": 5.5,
          "resident_model_gb": 5.5
        },
        "model_residency": "cold",
        "name": "vlm",
        "quality_profile": {
          "benchmark": "vlm_document_extraction",
          "field_accuracy": 1.0,
          "model_family": "Qwen3.5-2B SMT",
          "strict_json_rate": 0.967
        },
        "quantization": "Q4_1",
        "quarantine_scope": "none",
        "queue_capacity": 4,
        "queue_depth": 0,
        "replicas": 1,
        "resource_domain": "ACCEL.aux.session",
        "resource_key": "ACCEL-SVC",
        "resource_lease_amount": 1,
        "resource_lease_type": "exclusive_session",
        "runtime_adapter": "llama_http",
        "runtime_status": {
          "adapter_ref": "llama_http",
          "last_successful_inference_ts": "",
          "last_worker_error_reason": "",
          "lifecycle": "COLD",
          "ready": false,
          "worker_kind": "llama_http"
        },
        "service_class": "realtime_vlm_compact",
        "state": "UNAVAILABLE",
        "state_revision": 0,
        "worker_kind": "llama_http"
      },
      {
        "adapter_ref": "ort_ep",
        "artifact": {
          "availability": "local",
          "path": "/root/models/spacemit-ai/vision/ppocr/PP-OCRv5_mobile_rec.onnx",
          "runtime": "ort_ep",
          "status": "present"
        },
        "backend": "NPU",
        "backend_profile": {
          "ocr_ingest": {
            "adaptive_render_dpi": [
              200,
              100,
              80
            ],
            "degradation_policy": "poll_scheduler_job_to_terminal_before_metadata_only",
            "recommended_document_timeout_ms": 180000,
            "recommended_poll_timeout_ms": 180000,
            "recommended_timeout_ms": 120000,
            "schema_version": "ocr_ingest_contract.v1"
          },
          "ocr_pipeline": {
            "det_model_path": "/root/models/spacemit-ai/vision/ppocr/PP-OCRv5_mobile_det.onnx",
            "dictionary_path": "/root/models/spacemit-ai/vision/ppocr/ppocrv5_dict.txt",
            "max_lines": 256,
            "max_side": 960,
            "mode": "recognize"
          }
        },
        "cold_start_eta_ms": 0,
        "context_window_tokens": 0,
        "devices": {
          "ACCEL": {
            "in_flight": 0,
            "role": "primary",
            "status": "cold"
          }
        },
        "dispatchable": "DISPLACED",
        "embedding_dim": 0,
        "estimated_runtime_ms": 15000,
        "eta_ms": 0,
        "exclusive": true,
        "hot_state": "cold",
        "last_successful_inference_ts": "",
        "last_worker_error": "",
        "last_worker_error_reason": "",
        "last_worker_error_ts": "",
        "lifecycle": "COLD",
        "max_output_tokens": 0,
        "memory_watermark": {
          "kv_cache_per_1k_tokens_gb": 0.0,
          "per_worker_dram_gb": 0.6,
          "resident_model_gb": 0.6
        },
        "model_residency": "cold",
        "name": "ocr-rec",
        "ocr_contract": {
          "adaptive_render_dpi": [
            200,
            100,
            80
          ],
          "client_owned_pdf_render": true,
          "degradation_policy": "poll_scheduler_job_to_terminal_before_metadata_only",
          "input_media_types": [
            "image/png"
          ],
          "max_request_body_bytes": 16777216,
          "max_worker_timeout_ms": 600000,
          "model": "ocr-rec",
          "ocr_pipeline": {
            "det_model_path": "/root/models/spacemit-ai/vision/ppocr/PP-OCRv5_mobile_det.onnx",
            "dictionary_path": "/root/models/spacemit-ai/vision/ppocr/ppocrv5_dict.txt",
            "max_lines": 256,
            "max_side": 960,
            "mode": "recognize"
          },
          "one_rendered_page_per_call": true,
          "recommended_async": true,
          "recommended_document_timeout_ms": 180000,
          "recommended_endpoint": "/kb/tasks/kb.document.ocr_recognize:async",
          "recommended_poll_timeout_ms": 180000,
          "recommended_timeout_ms": 120000,
          "render_retry_policy": "retry_lower_dpi_before_scheduler_submit_failure",
          "scheduler_timeout_scope": "worker_call_timeout_ms_not_async_job_retention",
          "schema_version": "ocr_ingest_contract.v1",
          "task": "kb.document.ocr_recognize"
        },
        "quality_profile": {},
        "quantization": "unknown",
        "quarantine_scope": "none",
        "queue_capacity": 16,
        "queue_depth": 0,
        "replicas": 1,
        "resource_domain": "ACCEL.aux",
        "resource_key": "ACCEL-BATCH",
        "resource_lease_amount": 1,
        "resource_lease_type": "counting",
        "runtime_adapter": "ort_ep",
        "runtime_status": {
          "adapter_ref": "ort_ep",
          "last_successful_inference_ts": "",
          "last_worker_error_reason": "",
          "lifecycle": "COLD",
          "ready": false,
          "worker_kind": "ort_ep"
        },
        "service_class": "background_index",
        "state": "UNAVAILABLE",
        "state_revision": 0,
        "worker_kind": "ort_ep"
      },
      {
        "adapter_ref": "ort_ep",
        "artifact": {
          "availability": "local",
          "path": "/root/models/spacemit-ai/vision/ppocr/PP-OCRv5_mobile_det.onnx",
          "runtime": "ort_ep",
          "status": "present"
        },
        "backend": "NPU",
        "backend_profile": {
          "ocr_ingest": {
            "adaptive_render_dpi": [
              200,
              160,
              120
            ],
            "degradation_policy": "poll_scheduler_job_to_terminal_before_metadata_only",
            "recommended_document_timeout_ms": 180000,
            "recommended_poll_timeout_ms": 180000,
            "recommended_timeout_ms": 90000,
            "schema_version": "ocr_ingest_contract.v1"
          },
          "ocr_pipeline": {
            "max_lines": 256,
            "max_side": 960,
            "mode": "detect"
          }
        },
        "cold_start_eta_ms": 0,
        "context_window_tokens": 0,
        "devices": {
          "ACCEL": {
            "in_flight": 0,
            "role": "primary",
            "status": "cold"
          }
        },
        "dispatchable": "DISPLACED",
        "embedding_dim": 0,
        "estimated_runtime_ms": 5000,
        "eta_ms": 0,
        "exclusive": true,
        "hot_state": "cold",
        "last_successful_inference_ts": "",
        "last_worker_error": "",
        "last_worker_error_reason": "",
        "last_worker_error_ts": "",
        "lifecycle": "COLD",
        "max_output_tokens": 0,
        "memory_watermark": {
          "kv_cache_per_1k_tokens_gb": 0.0,
          "per_worker_dram_gb": 0.3,
          "resident_model_gb": 0.3
        },
        "model_residency": "cold",
        "name": "ocr-det",
        "ocr_contract": {
          "adaptive_render_dpi": [
            200,
            160,
            120
          ],
          "client_owned_pdf_render": true,
          "degradation_policy": "poll_scheduler_job_to_terminal_before_metadata_only",
          "input_media_types": [
            "image/png"
          ],
          "max_request_body_bytes": 16777216,
          "max_worker_timeout_ms": 600000,
          "model": "ocr-det",
          "ocr_pipeline": {
            "max_lines": 256,
            "max_side": 960,
            "mode": "detect"
          },
          "one_rendered_page_per_call": true,
          "recommended_async": true,
          "recommended_document_timeout_ms": 180000,
          "recommended_endpoint": "/kb/tasks/kb.document.ocr_detect:async",
          "recommended_poll_timeout_ms": 180000,
          "recommended_timeout_ms": 90000,
          "render_retry_policy": "retry_lower_dpi_before_scheduler_submit_failure",
          "scheduler_timeout_scope": "worker_call_timeout_ms_not_async_job_retention",
          "schema_version": "ocr_ingest_contract.v1",
          "task": "kb.document.ocr_detect"
        },
        "quality_profile": {},
        "quantization": "unknown",
        "quarantine_scope": "none",
        "queue_capacity": 16,
        "queue_depth": 0,
        "replicas": 1,
        "resource_domain": "ACCEL.aux",
        "resource_key": "ACCEL-BATCH",
        "resource_lease_amount": 1,
        "resource_lease_type": "counting",
        "runtime_adapter": "ort_ep",
        "runtime_status": {
          "adapter_ref": "ort_ep",
          "last_successful_inference_ts": "",
          "last_worker_error_reason": "",
          "lifecycle": "COLD",
          "ready": false,
          "worker_kind": "ort_ep"
        },
        "service_class": "background_index",
        "state": "UNAVAILABLE",
        "state_revision": 0,
        "worker_kind": "ort_ep"
      }
    ],
    "revision": 72,
    "schema_version": "models.v2"
  },
  "post_summary_chat": {
    "http": 200,
    "body": {
      "admission_trace": {
        "admission": "accepted",
        "api_surface": "scheduler",
        "deadline_ms": 30000,
        "estimated_runtime_ms": 3200,
        "eta_ms": 3200,
        "model": "llm-chat",
        "normalized_task": "kb.chat.complete",
        "priority": 96,
        "queue_capacity": 8,
        "queue_depth": 0,
        "reason": "sync_admitted",
        "resource_domain": "ACCEL.chat",
        "resource_key": "ACCEL-SVC",
        "resource_queue_capacity": 8,
        "resource_queue_depth": 0,
        "resource_queue_oldest_wait_ms": 0,
        "route": "POST /kb/tasks/kb.chat.complete",
        "runtime_estimate_samples": 1,
        "runtime_estimate_source": "static",
        "scheduled_as": "sync",
        "schema_version": "openai_scheduler_unified_admission.v1",
        "service_class": "realtime_answer"
      },
      "answer_profile": "normal_answer",
      "backend": "llama_http",
      "cache_hit": false,
      "cache_hit_source": "scheduler_prefix_hint",
      "cache_key": "fnv1a64:38330094b9dbe01d",
      "cache_revision": 3,
      "cached_tokens": 0,
      "cold_start_wait_ms": 0.0,
      "decode_ms": 929.195,
      "device_used": "ACCEL",
      "finish_reason": "stop",
      "latency_ms": 2023.729714,
      "max_output_tokens": 32,
      "model": "llm-chat",
      "model_residency": "hot_resident",
      "output_tokens": 31,
      "outputs": {
        "choices": [
          {
            "finish_reason": "stop",
            "index": 0,
            "message": {
              "content": "The current llm-chat model is MiniCPM5-1B Q4_K_M, and no additional citations are needed for this response.",
              "role": "assistant"
            }
          }
        ],
        "created": 1786497330,
        "id": "chatcmpl-Kvs5z2MdaGq80nTUKK200ZWKwj68qDxY",
        "model": "MiniCPM5-1B-Q4_K_M.gguf",
        "object": "chat.completion",
        "system_fingerprint": "b9488-21379f9d7",
        "timings": {
          "cache_n": 0,
          "predicted_ms": 929.195,
          "predicted_n": 31,
          "predicted_per_second": 33.362211376514075,
          "predicted_per_token_ms": 29.97403225806452,
          "prompt_ms": 1075.341,
          "prompt_n": 270,
          "prompt_per_second": 251.08314478849036,
          "prompt_per_token_ms": 3.982744444444444
        },
        "usage": {
          "completion_tokens": 31,
          "prompt_tokens": 270,
          "prompt_tokens_details": {
            "cached_tokens": 0
          },
          "total_tokens": 301
        }
      },
      "prefill_ms": 1075.341,
      "prompt_cache_key": "fnv1a64:38330094b9dbe01d",
      "prompt_tokens": 270,
      "queue_wait_ms": 200.350459,
      "scheduled_as": "sync",
      "schema_version": "kb_task.v1",
      "service_class": "realtime_answer",
      "startup_state": "hot_resident",
      "startup_wait_ms": 0.343107,
      "task": "kb.chat.complete",
      "telemetry": {
        "answer_profile": "normal_answer",
        "backend": "llama_http",
        "cache_hit": false,
        "cache_hit_source": "scheduler_prefix_hint",
        "cache_key": "fnv1a64:38330094b9dbe01d",
        "cache_revision": 3,
        "cached_tokens": 0,
        "cold_start_wait_ms": 0.0,
        "decode_ms": 929.195,
        "finish_reason": "stop",
        "latency_ms": 2023.729714,
        "max_output_tokens": 32,
        "model_residency": "hot_resident",
        "output_tokens": 31,
        "prefill_ms": 1075.341,
        "prompt_cache_key": "fnv1a64:38330094b9dbe01d",
        "prompt_tokens": 270,
        "queue_ms": 200.350459,
        "startup_state": "hot_resident",
        "startup_wait_ms": 0.343107,
        "trace_id": "k3-deb-e2e-post-summary-chat",
        "truncated": false,
        "worker_id": "llm-chat@ACCEL:3362296",
        "worker_pid": 3362296
      },
      "trace_id": "k3-deb-e2e-post-summary-chat",
      "truncated": false,
      "worker_id": "llm-chat@ACCEL:3362296",
      "worker_pid": 3362296
    }
  },
  "models": {
    "capability_revision": 77,
    "contract_revision": 77,
    "contract_version": "edge-scheduler-v2",
    "model_revision": 77,
    "models": [
      {
        "adapter_ref": "llama_http",
        "artifact": {
          "availability": "local",
          "path": "/root/models/spacemit-ai/llm/Qwen3-30B-A3B-Q4_0.gguf",
          "quantization": "Q4_0",
          "runtime": "llama_http",
          "status": "present"
        },
        "backend": "RVV/IME",
        "backend_profile": {
          "accelerator_path": "ime2",
          "api": "/v1/chat/completions",
          "artifact_source": "spacemit_model_zoo",
          "model_family": "Qwen3-30B-A3B",
          "quantization": "Q4_0",
          "requires_no_think": true,
          "role": "document_summary_and_long_context_organization",
          "runtime": "llama.cpp-tools-spacemit",
          "runtime_binary": "/usr/bin/llama-server"
        },
        "cold_start_eta_ms": 70000,
        "context_window_tokens": 8192,
        "devices": {
          "ACCEL": {
            "in_flight": 0,
            "role": "primary",
            "status": "cold"
          }
        },
        "dispatchable": "DISPLACED",
        "embedding_dim": 0,
        "estimated_runtime_ms": 15000,
        "eta_ms": 0,
        "exclusive": true,
        "hot_state": "cold",
        "last_latency_ms": 11149.048622,
        "last_success_ts": "2026-08-12T01:15:06Z",
        "last_successful_inference_ts": "2026-08-12T01:15:06Z",
        "last_worker_error": "",
        "last_worker_error_reason": "",
        "last_worker_error_ts": "",
        "lifecycle": "COLD",
        "max_output_tokens": 1024,
        "memory_watermark": {
          "kv_cache_per_1k_tokens_gb": 0.75,
          "per_worker_dram_gb": 18.0,
          "resident_model_gb": 18.0
        },
        "model_residency": "cold",
        "name": "llm-summary",
        "p50_latency_ms": 11149.048622,
        "quality_profile": {},
        "quantization": "Q4_0",
        "quarantine_scope": "none",
        "queue_capacity": 8,
        "queue_depth": 0,
        "replicas": 1,
        "resource_domain": "ACCEL.aux.session",
        "resource_key": "ACCEL-SVC",
        "resource_lease_amount": 1,
        "resource_lease_type": "exclusive_session",
        "runtime_adapter": "llama_http",
        "runtime_status": {
          "adapter_ref": "llama_http",
          "last_successful_inference_ts": "2026-08-12T01:15:06Z",
          "last_worker_error_reason": "",
          "lifecycle": "COLD",
          "ready": false,
          "worker_kind": "llama_http"
        },
        "service_class": "realtime_answer",
        "state": "UNAVAILABLE",
        "state_revision": 68,
        "worker_kind": "llama_http"
      },
      {
        "adapter_ref": "llama_http",
        "artifact": {
          "availability": "local",
          "path": "/root/models/spacemit-ai/llm/MiniCPM5-1B/MiniCPM5-1B-Q4_K_M.gguf",
          "quantization": "Q4_K_M",
          "runtime": "llama_http",
          "status": "present"
        },
        "backend": "RVV/IME",
        "backend_profile": {
          "accelerator_path": "ime2",
          "api": "/v1/chat/completions",
          "artifact_source": "vlm-llm-benchmark:minicpm-probe",
          "bench_evidence": "docs/minicpm-k3-rag-replacement.zh.md",
          "chat_template_kwargs": {
            "enable_thinking": false
          },
          "model_family": "MiniCPM5-1B",
          "quantization": "Q4_K_M",
          "recommended_stop": [
            "\n\n",
            "请注意",
            "注：",
            "停止。",
            "以上。"
          ],
          "requires_no_think": true,
          "role": "rag_answer_candidate",
          "runtime": "llama.cpp-tools-spacemit",
          "runtime_binary": "/usr/bin/llama-server"
        },
        "cold_start_eta_ms": 15000,
        "context_window_tokens": 8192,
        "devices": {
          "ACCEL": {
            "in_flight": 0,
            "role": "primary",
            "status": "ready"
          }
        },
        "dispatchable": "FREE",
        "embedding_dim": 0,
        "estimated_runtime_ms": 3200,
        "eta_ms": 0,
        "exclusive": true,
        "hot_state": "resident",
        "last_latency_ms": 2024.250811,
        "last_success_ts": "2026-08-12T01:15:30Z",
        "last_successful_inference_ts": "2026-08-12T01:15:30Z",
        "last_worker_error": "",
        "last_worker_error_reason": "",
        "last_worker_error_ts": "",
        "lifecycle": "READY",
        "max_output_tokens": 1024,
        "memory_watermark": {
          "kv_cache_per_1k_tokens_gb": 0.08,
          "per_worker_dram_gb": 1.2,
          "resident_model_gb": 1.2
        },
        "model_residency": "resident",
        "name": "llm-chat",
        "p50_latency_ms": 1864.3212961,
        "quality_profile": {},
        "quantization": "Q4_K_M",
        "quarantine_scope": "none",
        "queue_capacity": 8,
        "queue_depth": 0,
        "replicas": 1,
        "resource_domain": "ACCEL.chat",
        "resource_key": "ACCEL-SVC",
        "resource_lease_amount": 1,
        "resource_lease_type": "exclusive_session",
        "runtime_adapter": "llama_http",
        "runtime_status": {
          "adapter_ref": "llama_http",
          "last_successful_inference_ts": "2026-08-12T01:15:30Z",
          "last_worker_error_reason": "",
          "lifecycle": "READY",
          "ready": true,
          "worker_kind": "llama_http"
        },
        "service_class": "realtime_answer",
        "state": "READY_FAST",
        "state_revision": 77,
        "worker_kind": "llama_http"
      },
      {
        "adapter_ref": "llama_http",
        "artifact": {
          "availability": "local",
          "path": "/root/models/spacemit-ai/rerank/Bge-Reranker-V2-M3-Q4_0.gguf",
          "quantization": "Q4_0",
          "runtime": "llama_http",
          "status": "present"
        },
        "backend": "RVV/IME",
        "backend_profile": {
          "accelerator_path": "ime2",
          "api": "/v1/rerank",
          "artifact_source": "spacemit_model_zoo",
          "embedding_dim": 1024,
          "pooling": "rank",
          "quantization": "Q4_0",
          "runtime": "llama.cpp-tools-spacemit",
          "runtime_binary": "/usr/bin/llama-server"
        },
        "cold_start_eta_ms": 8000,
        "context_window_tokens": 0,
        "devices": {
          "ACCEL": {
            "in_flight": 0,
            "role": "primary",
            "status": "cold"
          }
        },
        "dispatchable": "DISPLACED",
        "embedding_dim": 1024,
        "estimated_runtime_ms": 145,
        "eta_ms": 0,
        "exclusive": false,
        "hot_state": "cold",
        "last_successful_inference_ts": "",
        "last_worker_error": "",
        "last_worker_error_reason": "",
        "last_worker_error_ts": "",
        "lifecycle": "COLD",
        "max_output_tokens": 0,
        "memory_watermark": {
          "kv_cache_per_1k_tokens_gb": 0.0,
          "per_worker_dram_gb": 0.8,
          "resident_model_gb": 0.8
        },
        "model_residency": "cold",
        "name": "reranker-int8",
        "quality_profile": {},
        "quantization": "Q4_0",
        "quarantine_scope": "none",
        "queue_capacity": 16,
        "queue_depth": 0,
        "replicas": 1,
        "resource_domain": "ACCEL.aux",
        "resource_key": "ACCEL-BATCH",
        "resource_lease_amount": 1,
        "resource_lease_type": "counting",
        "runtime_adapter": "llama_http",
        "runtime_status": {
          "adapter_ref": "llama_http",
          "last_successful_inference_ts": "",
          "last_worker_error_reason": "",
          "lifecycle": "COLD",
          "ready": false,
          "worker_kind": "llama_http"
        },
        "service_class": "realtime_retrieval",
        "state": "UNAVAILABLE",
        "state_revision": 0,
        "worker_kind": "llama_http"
      },
      {
        "adapter_ref": "llama_http",
        "artifact": {
          "availability": "local",
          "path": "/root/models/spacemit-ai/embed/Bge-Small-Zh-V1.5-Q4_K_M.gguf",
          "quantization": "Q4_K_M",
          "runtime": "llama_http",
          "status": "present"
        },
        "backend": "RVV/IME",
        "backend_profile": {
          "accelerator_path": "ime2",
          "api": "/v1/embeddings",
          "artifact_source": "spacemit_model_zoo",
          "embedding_dim": 1024,
          "pooling": "mean",
          "quantization": "Q4_K_M",
          "runtime": "llama.cpp-tools-spacemit",
          "runtime_binary": "/usr/bin/llama-server"
        },
        "cold_start_eta_ms": 8000,
        "context_window_tokens": 0,
        "devices": {
          "ACCEL": {
            "in_flight": 0,
            "role": "primary",
            "status": "ready"
          }
        },
        "dispatchable": "FREE",
        "embedding_dim": 1024,
        "estimated_runtime_ms": 25,
        "eta_ms": 0,
        "exclusive": false,
        "hot_state": "resident",
        "last_latency_ms": 9.527033,
        "last_success_ts": "2026-08-12T01:13:07Z",
        "last_successful_inference_ts": "2026-08-12T01:13:07Z",
        "last_worker_error": "",
        "last_worker_error_reason": "",
        "last_worker_error_ts": "",
        "lifecycle": "READY",
        "max_output_tokens": 0,
        "memory_watermark": {
          "kv_cache_per_1k_tokens_gb": 0.0,
          "per_worker_dram_gb": 0.7,
          "resident_model_gb": 2.8
        },
        "model_residency": "resident",
        "name": "embedding-int8",
        "p50_latency_ms": 9.527033,
        "quality_profile": {},
        "quantization": "Q4_K_M",
        "quarantine_scope": "none",
        "queue_capacity": 32,
        "queue_depth": 0,
        "replicas": 4,
        "resource_domain": "ACCEL.retrieval",
        "resource_key": "ACCEL-RETRIEVAL",
        "resource_lease_amount": 1,
        "resource_lease_type": "counting",
        "runtime_adapter": "llama_http",
        "runtime_status": {
          "adapter_ref": "llama_http",
          "last_successful_inference_ts": "2026-08-12T01:13:07Z",
          "last_worker_error_reason": "",
          "lifecycle": "READY",
          "ready": true,
          "worker_kind": "llama_http"
        },
        "service_class": "realtime_retrieval",
        "state": "READY_FAST",
        "state_revision": 22,
        "worker_kind": "llama_http"
      },
      {
        "adapter_ref": "openai_http",
        "artifact": {
          "availability": "external",
          "path": "sherpa-onnx-matcha-tts",
          "runtime": "openai_http",
          "status": "unknown"
        },
        "backend": "RVV/IME",
        "backend_profile": {
          "adapter": "openai_http",
          "api": "/v1/audio/speech",
          "artifact_source": "image_preinstalled",
          "legacy_contract_model": "/root/models/spacemit-ai/tts/synthetic-token-ids-v1.onnx",
          "model_family": "sherpa-onnx-matcha-tts",
          "not_ready_code": "native_tts_model_unavailable",
          "quality_scope": "native product TTS PCM16 WAV generation",
          "required_model_dir": "/opt/rv-sherpa-onnx/models/candidates/current/tts-matcha",
          "runtime": "rv-sherpa-onnx-runtime"
        },
        "cold_start_eta_ms": 7000,
        "context_window_tokens": 0,
        "devices": {
          "CPU": {
            "in_flight": 0,
            "role": "primary",
            "status": "cold"
          }
        },
        "dispatchable": "DISPLACED",
        "embedding_dim": 0,
        "estimated_runtime_ms": 800,
        "eta_ms": 0,
        "exclusive": true,
        "hot_state": "cold",
        "last_successful_inference_ts": "",
        "last_worker_error": "",
        "last_worker_error_reason": "",
        "last_worker_error_ts": "",
        "lifecycle": "COLD",
        "max_output_tokens": 0,
        "memory_watermark": {
          "kv_cache_per_1k_tokens_gb": 0.0,
          "per_worker_dram_gb": 0.5,
          "resident_model_gb": 0.5
        },
        "model_residency": "cold",
        "name": "tts-default",
        "quality_profile": {},
        "quantization": "unknown",
        "quarantine_scope": "none",
        "queue_capacity": 4,
        "queue_depth": 0,
        "replicas": 1,
        "resource_domain": "CPU",
        "resource_key": "CPU",
        "resource_lease_amount": 2,
        "resource_lease_type": "counting",
        "runtime_adapter": "openai_http",
        "runtime_status": {
          "adapter_ref": "openai_http",
          "last_successful_inference_ts": "",
          "last_worker_error_reason": "",
          "lifecycle": "COLD",
          "ready": false,
          "worker_kind": "openai_http"
        },
        "service_class": "realtime_cpu",
        "state": "UNAVAILABLE",
        "state_revision": 0,
        "worker_kind": "openai_http"
      },
      {
        "adapter_ref": "openai_http",
        "artifact": {
          "availability": "external",
          "path": "sherpa-onnx-streaming-zipformer",
          "runtime": "openai_http",
          "status": "unknown"
        },
        "backend": "RVV/IME",
        "backend_profile": {
          "accepts": [
            "audio_base64",
            "chunk_mel"
          ],
          "adapter": "openai_http",
          "api": "/v1/audio/transcriptions",
          "artifact_source": "image_preinstalled",
          "legacy_frontend_probe": "/root/models/spacemit-ai/vlm/qwen3-asr-0.6B/Qwen3-ASR-0.6B-encoder-frontend.dynq.onnx",
          "model_dir": "/opt/rv-sherpa-onnx/models/candidates/current/asr-streaming-zipformer",
          "model_family": "sherpa-onnx-streaming-zipformer-bilingual-zh-en",
          "runtime": "rv-sherpa-onnx-runtime",
          "stage": "native_audio_transcription"
        },
        "cold_start_eta_ms": 0,
        "context_window_tokens": 0,
        "devices": {
          "CPU": {
            "in_flight": 0,
            "role": "primary",
            "status": "cold"
          }
        },
        "dispatchable": "DISPLACED",
        "embedding_dim": 0,
        "estimated_runtime_ms": 500,
        "eta_ms": 0,
        "exclusive": true,
        "hot_state": "cold",
        "last_successful_inference_ts": "",
        "last_worker_error": "",
        "last_worker_error_reason": "",
        "last_worker_error_ts": "",
        "lifecycle": "COLD",
        "max_output_tokens": 0,
        "memory_watermark": {
          "kv_cache_per_1k_tokens_gb": 0.0,
          "per_worker_dram_gb": 0.4,
          "resident_model_gb": 0.4
        },
        "model_residency": "cold",
        "name": "asr-whisper",
        "quality_profile": {},
        "quantization": "unknown",
        "quarantine_scope": "none",
        "queue_capacity": 24,
        "queue_depth": 0,
        "replicas": 1,
        "resource_domain": "CPU",
        "resource_key": "CPU",
        "resource_lease_amount": 4,
        "resource_lease_type": "counting",
        "runtime_adapter": "openai_http",
        "runtime_status": {
          "adapter_ref": "openai_http",
          "last_successful_inference_ts": "",
          "last_worker_error_reason": "",
          "lifecycle": "COLD",
          "ready": false,
          "worker_kind": "openai_http"
        },
        "service_class": "realtime_cpu",
        "state": "UNAVAILABLE",
        "state_revision": 0,
        "worker_kind": "openai_http"
      },
      {
        "adapter_ref": "llama_http",
        "artifact": {
          "availability": "local",
          "path": "/root/models/spacemit-ai/vlm/Qwen3.5-2B/qwen3_5_2b-text-q41.gguf",
          "quantization": "Q4_1",
          "runtime": "llama_http",
          "status": "present"
        },
        "backend": "RVV/IME",
        "backend_profile": {
          "accelerator_path": "ime2+smt",
          "api": "/v1/chat/completions",
          "artifact_source": "spacemit_model_zoo",
          "media_backend": "smt",
          "model_family": "Qwen3.5-2B SMT",
          "quantization": "Q4_1",
          "requires_no_think": true,
          "runtime": "llama.cpp-tools-spacemit",
          "runtime_binary": "/usr/bin/llama-server",
          "smt_config_dir": "/root/models/spacemit-ai/vlm/Qwen3.5-2B"
        },
        "cold_start_eta_ms": 25000,
        "context_window_tokens": 4096,
        "devices": {
          "ACCEL": {
            "in_flight": 0,
            "role": "primary",
            "status": "cold"
          }
        },
        "dispatchable": "DISPLACED",
        "embedding_dim": 0,
        "estimated_runtime_ms": 11000,
        "eta_ms": 0,
        "exclusive": true,
        "hot_state": "cold",
        "last_successful_inference_ts": "",
        "last_worker_error": "",
        "last_worker_error_reason": "",
        "last_worker_error_ts": "",
        "lifecycle": "COLD",
        "max_output_tokens": 1024,
        "memory_watermark": {
          "kv_cache_per_1k_tokens_gb": 0.125,
          "per_worker_dram_gb": 5.5,
          "resident_model_gb": 5.5
        },
        "model_residency": "cold",
        "name": "vlm",
        "quality_profile": {
          "benchmark": "vlm_document_extraction",
          "field_accuracy": 1.0,
          "model_family": "Qwen3.5-2B SMT",
          "strict_json_rate": 0.967
        },
        "quantization": "Q4_1",
        "quarantine_scope": "none",
        "queue_capacity": 4,
        "queue_depth": 0,
        "replicas": 1,
        "resource_domain": "ACCEL.aux.session",
        "resource_key": "ACCEL-SVC",
        "resource_lease_amount": 1,
        "resource_lease_type": "exclusive_session",
        "runtime_adapter": "llama_http",
        "runtime_status": {
          "adapter_ref": "llama_http",
          "last_successful_inference_ts": "",
          "last_worker_error_reason": "",
          "lifecycle": "COLD",
          "ready": false,
          "worker_kind": "llama_http"
        },
        "service_class": "realtime_vlm_compact",
        "state": "UNAVAILABLE",
        "state_revision": 0,
        "worker_kind": "llama_http"
      },
      {
        "adapter_ref": "ort_ep",
        "artifact": {
          "availability": "local",
          "path": "/root/models/spacemit-ai/vision/ppocr/PP-OCRv5_mobile_rec.onnx",
          "runtime": "ort_ep",
          "status": "present"
        },
        "backend": "NPU",
        "backend_profile": {
          "ocr_ingest": {
            "adaptive_render_dpi": [
              200,
              100,
              80
            ],
            "degradation_policy": "poll_scheduler_job_to_terminal_before_metadata_only",
            "recommended_document_timeout_ms": 180000,
            "recommended_poll_timeout_ms": 180000,
            "recommended_timeout_ms": 120000,
            "schema_version": "ocr_ingest_contract.v1"
          },
          "ocr_pipeline": {
            "det_model_path": "/root/models/spacemit-ai/vision/ppocr/PP-OCRv5_mobile_det.onnx",
            "dictionary_path": "/root/models/spacemit-ai/vision/ppocr/ppocrv5_dict.txt",
            "max_lines": 256,
            "max_side": 960,
            "mode": "recognize"
          }
        },
        "cold_start_eta_ms": 0,
        "context_window_tokens": 0,
        "devices": {
          "ACCEL": {
            "in_flight": 0,
            "role": "primary",
            "status": "cold"
          }
        },
        "dispatchable": "DISPLACED",
        "embedding_dim": 0,
        "estimated_runtime_ms": 15000,
        "eta_ms": 0,
        "exclusive": true,
        "hot_state": "cold",
        "last_successful_inference_ts": "",
        "last_worker_error": "",
        "last_worker_error_reason": "",
        "last_worker_error_ts": "",
        "lifecycle": "COLD",
        "max_output_tokens": 0,
        "memory_watermark": {
          "kv_cache_per_1k_tokens_gb": 0.0,
          "per_worker_dram_gb": 0.6,
          "resident_model_gb": 0.6
        },
        "model_residency": "cold",
        "name": "ocr-rec",
        "ocr_contract": {
          "adaptive_render_dpi": [
            200,
            100,
            80
          ],
          "client_owned_pdf_render": true,
          "degradation_policy": "poll_scheduler_job_to_terminal_before_metadata_only",
          "input_media_types": [
            "image/png"
          ],
          "max_request_body_bytes": 16777216,
          "max_worker_timeout_ms": 600000,
          "model": "ocr-rec",
          "ocr_pipeline": {
            "det_model_path": "/root/models/spacemit-ai/vision/ppocr/PP-OCRv5_mobile_det.onnx",
            "dictionary_path": "/root/models/spacemit-ai/vision/ppocr/ppocrv5_dict.txt",
            "max_lines": 256,
            "max_side": 960,
            "mode": "recognize"
          },
          "one_rendered_page_per_call": true,
          "recommended_async": true,
          "recommended_document_timeout_ms": 180000,
          "recommended_endpoint": "/kb/tasks/kb.document.ocr_recognize:async",
          "recommended_poll_timeout_ms": 180000,
          "recommended_timeout_ms": 120000,
          "render_retry_policy": "retry_lower_dpi_before_scheduler_submit_failure",
          "scheduler_timeout_scope": "worker_call_timeout_ms_not_async_job_retention",
          "schema_version": "ocr_ingest_contract.v1",
          "task": "kb.document.ocr_recognize"
        },
        "quality_profile": {},
        "quantization": "unknown",
        "quarantine_scope": "none",
        "queue_capacity": 16,
        "queue_depth": 0,
        "replicas": 1,
        "resource_domain": "ACCEL.aux",
        "resource_key": "ACCEL-BATCH",
        "resource_lease_amount": 1,
        "resource_lease_type": "counting",
        "runtime_adapter": "ort_ep",
        "runtime_status": {
          "adapter_ref": "ort_ep",
          "last_successful_inference_ts": "",
          "last_worker_error_reason": "",
          "lifecycle": "COLD",
          "ready": false,
          "worker_kind": "ort_ep"
        },
        "service_class": "background_index",
        "state": "UNAVAILABLE",
        "state_revision": 0,
        "worker_kind": "ort_ep"
      },
      {
        "adapter_ref": "ort_ep",
        "artifact": {
          "availability": "local",
          "path": "/root/models/spacemit-ai/vision/ppocr/PP-OCRv5_mobile_det.onnx",
          "runtime": "ort_ep",
          "status": "present"
        },
        "backend": "NPU",
        "backend_profile": {
          "ocr_ingest": {
            "adaptive_render_dpi": [
              200,
              160,
              120
            ],
            "degradation_policy": "poll_scheduler_job_to_terminal_before_metadata_only",
            "recommended_document_timeout_ms": 180000,
            "recommended_poll_timeout_ms": 180000,
            "recommended_timeout_ms": 90000,
            "schema_version": "ocr_ingest_contract.v1"
          },
          "ocr_pipeline": {
            "max_lines": 256,
            "max_side": 960,
            "mode": "detect"
          }
        },
        "cold_start_eta_ms": 0,
        "context_window_tokens": 0,
        "devices": {
          "ACCEL": {
            "in_flight": 0,
            "role": "primary",
            "status": "cold"
          }
        },
        "dispatchable": "DISPLACED",
        "embedding_dim": 0,
        "estimated_runtime_ms": 5000,
        "eta_ms": 0,
        "exclusive": true,
        "hot_state": "cold",
        "last_successful_inference_ts": "",
        "last_worker_error": "",
        "last_worker_error_reason": "",
        "last_worker_error_ts": "",
        "lifecycle": "COLD",
        "max_output_tokens": 0,
        "memory_watermark": {
          "kv_cache_per_1k_tokens_gb": 0.0,
          "per_worker_dram_gb": 0.3,
          "resident_model_gb": 0.3
        },
        "model_residency": "cold",
        "name": "ocr-det",
        "ocr_contract": {
          "adaptive_render_dpi": [
            200,
            160,
            120
          ],
          "client_owned_pdf_render": true,
          "degradation_policy": "poll_scheduler_job_to_terminal_before_metadata_only",
          "input_media_types": [
            "image/png"
          ],
          "max_request_body_bytes": 16777216,
          "max_worker_timeout_ms": 600000,
          "model": "ocr-det",
          "ocr_pipeline": {
            "max_lines": 256,
            "max_side": 960,
            "mode": "detect"
          },
          "one_rendered_page_per_call": true,
          "recommended_async": true,
          "recommended_document_timeout_ms": 180000,
          "recommended_endpoint": "/kb/tasks/kb.document.ocr_detect:async",
          "recommended_poll_timeout_ms": 180000,
          "recommended_timeout_ms": 90000,
          "render_retry_policy": "retry_lower_dpi_before_scheduler_submit_failure",
          "scheduler_timeout_scope": "worker_call_timeout_ms_not_async_job_retention",
          "schema_version": "ocr_ingest_contract.v1",
          "task": "kb.document.ocr_detect"
        },
        "quality_profile": {},
        "quantization": "unknown",
        "quarantine_scope": "none",
        "queue_capacity": 16,
        "queue_depth": 0,
        "replicas": 1,
        "resource_domain": "ACCEL.aux",
        "resource_key": "ACCEL-BATCH",
        "resource_lease_amount": 1,
        "resource_lease_type": "counting",
        "runtime_adapter": "ort_ep",
        "runtime_status": {
          "adapter_ref": "ort_ep",
          "last_successful_inference_ts": "",
          "last_worker_error_reason": "",
          "lifecycle": "COLD",
          "ready": false,
          "worker_kind": "ort_ep"
        },
        "service_class": "background_index",
        "state": "UNAVAILABLE",
        "state_revision": 0,
        "worker_kind": "ort_ep"
      }
    ],
    "revision": 77,
    "schema_version": "models.v2"
  }
}