{
  "schema_version": "5.1",
  "name": "qwen3.8-flash-next-p300x2",
  "tt_metal_version": "0.65.2.dev9355+g283ee088631",
  "arch": "blackhole",
  "device_count": 4,
  "producer": {
    "tt_kernel_version": "0.1.0",
    "created_at": "2026-09-17T16:44:18.193551+00:00",
    "hostname": "tt-quietbox"
  },
  "weights": {
    "repo_id": "Qwen/Qwen3.8-Flash-Next",
    "revision": "f5d08274bafd880402bd16f5e3e6c514136ec06c",
    "allow_patterns": null,
    "ignore_patterns": null,
    "repo_type": "model"
  },
  "mesh": null,
  "entrypoint": null,
  "resources": null,
  "capabilities": null,
  "env": {},
  "bundled": null,
  "deps": null,
  "container": {
    "image": {
      "registry": "hf",
      "repository": "qwen3.8-flash-next-p300x2",
      "tag": "tt-model/qwen3.8-flash-next-p300x2:ba49bf267e38",
      "digest": "sha256:ba49bf267e3846e8fa859e7156365a98bf1344bc2503ddc26c31a5e42a6d7a85"
    },
    "kind": "vllm-plugin",
    "runtime": {
      "vllm": {
        "version": "0.24.0"
      },
      "plugin": {
        "path": "/home/ttuser/dev/qwen3.8-flash-next/vllm-tt-plugin-a48857a",
        "sha": "a48857ac68b17c31303e4809f348caaebbf10f74",
        "dirty": true
      },
      "extra_models_dir": "models/autoports/qwen_qwen3_8_flash_next/vllm_bundle",
      "lock": "requirements.lock"
    },
    "serve": {
      "hardware": "p300x2",
      "mesh_device": "(4, 1)",
      "port": 8000,
      "max_model_len": null,
      "max_num_seqs": 1,
      "block_size": 64,
      "server_timeout": null,
      "capabilities": {
        "tool_parser": "qwen3_xml",
        "reasoning_parser": "qwen3"
      },
      "additional_config": {
        "tt": {
          "sample_on_device_mode": "all",
          "trace_region_size": 1073741824,
          "fabric_config": "FABRIC_1D",
          "fabric_packet_payload_bytes": 8192,
          "l1_small_size": 24576,
          "trace_mode": "decode_only"
        }
      },
      "args": [
        "--async-scheduling",
        "--enable-chunked-prefill",
        [
          "--max-num-batched-tokens",
          "1024"
        ],
        [
          "--served-model-name",
          "Qwen/Qwen3.8-Flash-Next"
        ]
      ],
      "env": {
        "TT_VISIBLE_DEVICES": "0,1,2,3",
        "TT_VLLM_BUILTIN_MODELS": "1",
        "QWEN38_VLLM_LOG_METRICS": "1",
        "QWEN38_PREFILL_SCHEDULE": "stack_major",
        "QWEN38_DECODE_TRACE_SCHEDULE": "resident_stack",
        "QWEN38_MOE_KERNEL": "sparse_bank",
        "QWEN38_EXPERT_WEIGHT_CACHE": "/weight-cache/resident-experts-f5d08274-p300-tp4-ep4-bfp4-v1",
        "QWEN38_PREFILL_CHUNK": "512",
        "QWEN38_PREFILL_CHUNK_ADAPTIVE": "1",
        "QWEN38_MOE_PREFILL_SLABS": "1"
      }
    },
    "serve_profiles": [
      {
        "hardware": null,
        "mesh_device": null,
        "port": null,
        "max_model_len": 262144,
        "max_num_seqs": null,
        "block_size": null,
        "server_timeout": null,
        "capabilities": null,
        "additional_config": {},
        "args": [],
        "env": {},
        "name": "long-context",
        "description": "Full 262,144-token capacity, physical batch one."
      }
    ],
    "default_profile": "long-context",
    "code_dir": "code",
    "verify": [
      "import pathlib; p = pathlib.Path('/opt/tt-metal/models/autoports/qwen_qwen3_8_flash_next/requirements.lock'); text = p.read_text(); assert len(text.splitlines()) >= 400 and 'torch==2.11.0+cpu' in text and 'transformers==5.16.0' in text",
      "import json, pathlib; p = pathlib.Path('/opt/tt-metal/models/autoports/qwen_qwen3_8_flash_next/doc/datatype_sweep/selected_precision_config.json'); c = json.loads(p.read_text()); assert c['parallelism'] == {'mesh_shape': [4, 1], 'dense_tp': 4, 'expert_parallel': 4, 'expert_mode': 'resident_ep4', 'kv_replication': 2, 'indexer_kv_replication': 4}, c['parallelism']",
      "from models.autoports.qwen_qwen3_8_flash_next.tt.generator_vllm import Qwen4ExpForConditionalGeneration; assert Qwen4ExpForConditionalGeneration",
      "from transformers.models.qwen4_exp.configuration_qwen4_exp import Qwen4ExpConfig; assert Qwen4ExpConfig",
      "import ttnn; assert hasattr(ttnn.experimental.kda, 'recurrent_gated_delta_rule'), 'native recurrent GDN op is missing'",
      "import vllm; import vllm_tt_plugin.platform as p; import inspect; assert 'Qwen4ExpForConditionalGeneration' in inspect.getsource(p), 'plugin lacks Qwen3.8 registration/capability support'",
      "import vllm.tool_parsers; from vllm.tool_parsers.abstract_tool_parser import ToolParserManager; assert ToolParserManager.get_tool_parser('qwen3_xml'), 'qwen3_xml tool-call parser is missing'; from vllm.reasoning import ReasoningParserManager; assert ReasoningParserManager.get_reasoning_parser('qwen3'), 'qwen3 reasoning parser is missing'"
    ],
    "built": {
      "image": "tt-model/qwen3.8-flash-next-p300x2:ba49bf267e38",
      "repo": "tt-hous/qwen3.8-flash-next-p300x2",
      "tt_model_version": "0.1.0",
      "created_at": "2026-09-17T16:37:48+00:00",
      "tt_metal": {
        "sha": "283ee08863112c2268805e8c9fa2c04bfec4d41d",
        "describe": "v0.77.0-dev20260818-415-g283ee08863-dirty",
        "dirty": true,
        "scm_version": "0.65.2.dev9355+g283ee088631",
        "mode": "local",
        "remote": "https://github.com/tenstorrent/tt-metal.git",
        "branch": "hous/qwen3.8-flash-next",
        "pushed": false
      },
      "code_sha256": "1b5c8d17b1caee5f8b7e159ffc11a803e07811dc4be4f7071a21fa2d0d979c32",
      "plugin": {
        "sha": "a48857ac68b17c31303e4809f348caaebbf10f74",
        "path": "/home/ttuser/dev/qwen3.8-flash-next/vllm-tt-plugin-a48857a",
        "dirty": true
      },
      "image_digest": "sha256:ba49bf267e3846e8fa859e7156365a98bf1344bc2503ddc26c31a5e42a6d7a85"
    }
  }
}