{
  "schema": 1,
  "parameters": {
    "werk": "target/debug/werk",
    "config": "/tmp/werk-mgpu-single.json",
    "out": "/tmp/werk-mgpu-single-final.json",
    "repetitions": 3,
    "tokens": 16,
    "prompt": "Return the integer result of 17 + 25, with no explanation.",
    "required_gpus": 1,
    "timeout": 600,
    "quality": true
  },
  "inventory_before": {
    "devices": [
      {
        "id": "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9",
        "physical_index": 0,
        "vendor": "nvidia",
        "name": "NVIDIA GeForce RTX 3090",
        "architecture": "sm_86",
        "runtime": "cuda",
        "pci_bus_id": "00000000:03:00.0",
        "total_bytes": 25769803776,
        "free_bytes": 19479396352,
        "numa_node": null,
        "pcie_link": null,
        "visible_index": 0
      }
    ],
    "host_available_bytes": 99917062144,
    "memory_topology": null,
    "topology": null,
    "warnings": []
  },
  "werk_revision": "8d8ee73a002e8473818d1e86099e0a21b3b51d23",
  "dirty_diff": " .gitignore                                |   3 +-\n docs/api.md                               |   7 ++\n docs/backends.md                          |   6 ++\n docs/reference/cli.md                     |   6 ++\n docs/reference/environment-variables.md   |   5 ++\n docs/reference/werk-protocol-v1.md        |   7 ++\n src/api.rs                                |   1 +\n src/api/anthropic/extended.rs             |   6 +-\n src/api/anthropic/mod.rs                  |   7 +-\n src/api/chat.rs                           |  44 ++++++++++--\n src/api/extended.rs                       |  22 ++++--\n src/api/generation.rs                     |  26 ++++++-\n src/api/observability.rs                  |   3 +-\n src/api/router.rs                         |   3 +\n src/api/state.rs                          |  57 +++++++++++++++\n src/api/werk/mod.rs                       |  40 +++++++++++\n src/backend/llama_process_lifecycle.rs    |  70 ++++++++++++++++--\n src/backend/llama_server.rs               | 106 +++++++++++++++++++++++++--\n src/backend/llama_server/fp4.rs           |   8 +++\n src/backend/llama_server/runtime_state.rs | 116 ++++++++++++++++--------------\n src/backend/vllm.rs                       |  76 +++++++++++++++++++-\n src/backend/vllm/runtime_control.rs       |   6 +-\n src/cli.rs                                |  68 +++++++++++++++++-\n src/inference_service.rs                  |   3 +-\n src/inference_service/resources.rs        |  92 ++++++++++++++++++++++++\n src/lib.rs                                |   1 +\n src/werk_protocol/client.rs               |  35 +++++++++\n 27 files changed, 736 insertions(+), 88 deletions(-)\n",
  "cold_definition": "fresh worker; OS page cache not flushed",
  "samples": [],
  "hardware_cases": {
    "single": "configured",
    "two_equal": "requires explicit config and >=2 GPUs",
    "two_unequal": "requires explicit config and >=2 unequal GPUs",
    "independent": "requires explicit config and >=2 GPUs",
    "mixed": "requires explicit config and >=3 GPUs",
    "replicas": "requires explicit config and >=2 GPUs"
  },
  "status": "rejected_plan",
  "reason": "Error: GPU GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9 cannot fit its declared budget",
  "installed_llama_revision": "ec928150501c2572fec05cb949061672bb424914"
}