{
  "schema_version": "5.1",
  "name": "qwen3.8-27b-dflash2-vision-p300x2-q4kv",
  "tt_metal_version": "0.0.0.dev0",
  "arch": "blackhole",
  "device_count": 4,
  "producer": {
    "tt_kernel_version": "0.1.0",
    "created_at": "2026-09-18T15:45:26.832220+00:00",
    "hostname": "tt-quietbox"
  },
  "weights": {
    "repo_id": "Qwen/Qwen3.8-27B",
    "revision": "1d4bf0f2ff6012fd82039f2fa52739d0dd7c60c0",
    "allow_patterns": null,
    "ignore_patterns": null,
    "repo_type": "model"
  },
  "auxiliary_weights": [
    {
      "repo_id": "incoai/Qwen3.8-27B-DFlash2",
      "revision": "dedf8df68adfb1afeaf7b7480c0a0243108177b4",
      "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": "qwen3.8-27b-dflash2-vision-p300x2-q4kv",
      "tag": "tt-model/qwen3.8-27b-dflash2-vision-p300x2-q4kv:edc764d20366",
      "digest": "sha256:edc764d203665e44caf1b48385e243400c52262ccff91c737aefe26e31afe0a0"
    },
    "kind": "vllm-plugin",
    "runtime": {
      "vllm": {
        "version": "0.26.0"
      },
      "plugin": {
        "path": "/home/ttuser/qb2-unified-work/vllm-tt-plugin",
        "sha": "67c56f09abf0ad776955a34da20d36d17d7841e7",
        "dirty": false
      },
      "extra_models_dir": "models/demos/blackhole/qwen36/vllm_bundle",
      "lock": "requirements.lock"
    },
    "serve": {
      "hardware": "p300x2",
      "mesh_device": "P150x4",
      "port": 8000,
      "max_model_len": 262144,
      "max_num_seqs": 8,
      "block_size": 64,
      "server_timeout": null,
      "capabilities": {
        "tool_parser": "qwen3_coder",
        "reasoning_parser": "qwen3"
      },
      "additional_config": {
        "tt": {
          "fabric_config": "FABRIC_1D",
          "trace_region_size": 1073741824,
          "l1_small_size": 24576,
          "sample_on_device_mode": "decode_only"
        }
      },
      "args": [
        [
          "--max-num-batched-tokens",
          "262144"
        ]
      ],
      "env": {
        "ARCH_NAME": "blackhole",
        "TT_QWEN35_TEXT_VER": "qwen36_blackhole",
        "TT_MESH_GRAPH_DESC_PATH": "/opt/tt-metal/tt_metal/fabric/mesh_graph_descriptors/p300_x2_mesh_graph_descriptor.textproto",
        "QWEN36_MAX_TOKENS_ALL_USERS": "1050624",
        "QWEN_SDPA_KV_DTYPE": "bf4",
        "VLLM_RPC_TIMEOUT": "900000",
        "VLLM_CONFIGURE_LOGGING": "1",
        "TORCHDYNAMO_DISABLE": "1",
        "HF_HUB_OFFLINE": "1",
        "DFLASH_WEIGHTS": "incoai/Qwen3.8-27B-DFlash2@dedf8df68adfb1afeaf7b7480c0a0243108177b4",
        "QWEN36_DFLASH_BLOCK": "8",
        "QWEN36_DFLASH_MAX_PROMPT": "2048",
        "QWEN36_DFLASH_SERVE_BLOCK": "32",
        "QWEN36_ENABLE_VISION": "1",
        "QWEN36_MAX_IMAGES": "4"
      }
    },
    "serve_profiles": [
      {
        "hardware": null,
        "mesh_device": null,
        "port": null,
        "max_model_len": null,
        "max_num_seqs": 8,
        "block_size": null,
        "server_timeout": null,
        "capabilities": null,
        "additional_config": {},
        "args": [
          [
            "--max-num-batched-tokens",
            "262144"
          ],
          "--no-async-scheduling"
        ],
        "env": {
          "QWEN36_DRAFTER": "dflash2",
          "QWEN36_GDN_SPEC_FUSED": "1",
          "QWEN36_SPEC_EXACT_SDPA": "1",
          "QWEN36_DFLASH_BLOCK": "8",
          "QWEN36_DFLASH_BUCKETS": "1x8,4x8,8x4",
          "QWEN36_DFLASH_RAGGED": "1",
          "QWEN36_DFLASH_FOLD_SEED": "1",
          "QWEN36_ENABLE_VISION": "1",
          "QWEN36_MAX_IMAGES": "4"
        },
        "name": "unified",
        "description": "Up to eight mixed text/image requests with Q4 KV, automatic lossless DFlash routing, and exact plain-decode fallback for supported sampling and output controls."
      }
    ],
    "default_profile": "unified",
    "code_dir": "code",
    "verify": [
      "from models.demos.blackhole.qwen36.tt.qwen36_vllm import Qwen36ForCausalLM; assert Qwen36ForCausalLM",
      "import os; os.environ['QWEN36_DRAFTER']='dflash2'; os.environ['QWEN36_DFLASH_BUCKETS']='1x8,4x8,8x4'; from models.demos.blackhole.qwen36.tt.qwen36_vllm_dflash import Qwen36DFlashForCausalLM as C; c=C.get_model_capabilities(); assert c['tt_adaptive_block_output'] and c['tt_adaptive_block_batched'] and c['tt_plain_fallback'] and c['output_tokens_per_step'] == 32, c",
      "from models.demos.blackhole.qwen36.tt.dflash2_serving import DFlash2ServingDecoder; assert DFlash2ServingDecoder",
      "from models.demos.blackhole.qwen36.tt.dflash2_serving import DFlash2DualBucketDecoder, BucketPlanner; assert DFlash2DualBucketDecoder and BucketPlanner",
      "import os; os.environ['QWEN36_DRAFTER']='dflash2'; from models.demos.blackhole.qwen36.tt.qwen36_vllm_dflash import parse_buckets, check_buckets; assert check_buckets(parse_buckets('1x8,4x8,8x4'), 8, 7, True) == ((1, 8), (4, 8), (8, 4))",
      "import json; m=json.load(open('/opt/tt-metal/models/demos/blackhole/qwen36/vllm_bundle/qwen36/vllm_metadata.json')); assert m['main_class'].endswith(':Qwen36DFlashForCausalLM'), m",
      "from vllm_tt_plugin.config import get_tt_block_output_kv_lookahead_tokens, is_tt_adaptive_block_output_model, is_tt_adaptive_block_batched, is_tt_adaptive_block_ragged",
      "import os; os.environ['QWEN36_DRAFTER']='dflash2'; os.environ['QWEN36_DFLASH_RAGGED']='1'; from models.demos.blackhole.qwen36.tt.qwen36_vllm_dflash import Qwen36DFlashForCausalLM as C; c=C.get_model_capabilities(); assert c['tt_adaptive_block_batched'] and c['tt_adaptive_block_ragged']",
      "from vllm_tt_plugin.config import classify_tt_execution_lane; assert classify_tt_execution_lane(type('P', (), {'temperature': 0, 'n': 1, 'logprobs': None})())[0] == 'dflash'",
      "from models.demos.blackhole.qwen36.tt.dflash2_decode import DFlash2Decoder; from models.demos.blackhole.qwen36.tt.dflash2_tp import DFlash2DrafterTP; assert DFlash2Decoder and DFlash2DrafterTP",
      "import vllm.model_executor.models.qwen3_5",
      "import transformers.models.qwen3_5",
      "from models.demos.blackhole.qwen36.tt.vision.model import interpolate_pos_embeddings; assert interpolate_pos_embeddings",
      "import os; from models.demos.blackhole.qwen36.tt.qwen36_vllm import TT_Qwen3_5ProcessingInfo as I; os.environ['QWEN36_ENABLE_VISION']='1'; os.environ['QWEN36_MAX_IMAGES']='4'; assert I.get_supported_mm_limits(None) == {'image': 4}",
      "from pathlib import Path; assert Path('/opt/tt-metal/tt_metal/fabric/mesh_graph_descriptors/p300_x2_mesh_graph_descriptor.textproto').is_file()",
      "from pathlib import Path; assert Path('/opt/tt-metal/models/model_trace_region_sizes.yaml').is_file()",
      "from pathlib import Path; assert Path('/opt/tt-metal/models/demos/blackhole/qwen36/vllm_bundle/qwen36/vllm_metadata.json').is_file()"
    ],
    "built": {
      "image": "tt-model/qwen3.8-27b-dflash2-vision-p300x2-q4kv:edc764d20366",
      "repo": "mando2222/qwen3.8-27b-dflash2-vision-p300x2-q4kv",
      "tt_model_version": "0.1.0",
      "created_at": "2026-09-18T15:40:01+00:00",
      "tt_metal": {
        "sha": "547841b0109bf57f020355287841cd4f38eec98b",
        "describe": null,
        "dirty": false,
        "scm_version": "0.0.0.dev0",
        "mode": "local",
        "remote": "https://github.com/tenstorrent/tt-metal.git",
        "branch": "qb2-unified-vision",
        "pushed": false
      },
      "code_sha256": "ce4a1982fe5943272979f6ef19e229db27320346ed4221de1788224288c69af2",
      "plugin": {
        "sha": "67c56f09abf0ad776955a34da20d36d17d7841e7",
        "path": "/home/ttuser/qb2-unified-work/vllm-tt-plugin",
        "dirty": false
      },
      "image_digest": "sha256:edc764d203665e44caf1b48385e243400c52262ccff91c737aefe26e31afe0a0"
    }
  }
}