Skip to content

armgit / ODS

Public Imported from GitHub
0

ODS V3 Pre-Release: Public testing and refinement ahead of the official V3 launch. Turn your PC, Mac, or Linux box into a private AI server.

1.6 KB
Raw Blame
{
  "$schema": "https://osmantic.com/pixel/schemas/work-model-backend-config-v1.schema.json",
  "schemaVersion": 1,
  "environmentPath": "/etc/pixel-work/environment.json",
  "modelSource": {"kind": "directory", "path": "/srv/pixel/models/vllm-model"},
  "runtimeCacheSeed": {"path": "/srv/pixel/runtime-cache/vllm-seed", "artifactSha256": "aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa", "imageDigest": "sha256:aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa", "cacheFingerprint": "pinned-vllm-runtime-cache-v1", "subtrees": ["vllm", "triton", "sparkinfer", "tilelang"]},
  "backendNetwork": {"subnet": "172.31.255.0/29"},
  "containerUser": {"uid": 2000, "gid": 0},
  "publishLoopbackPort": null,
  "resources": {"memoryMiB": 196608, "cpuCores": 48, "pids": 4096, "nofile": 65536, "tmpfsMiB": 1024, "cacheMiB": 32768, "sharedMemoryMiB": 32768},
  "readiness": {"startupTimeoutSeconds": 900, "probeIntervalMilliseconds": 2000},
  "accelerator": {"class": "nvidia-cuda", "count": 2, "deviceIds": []},
  "providerOptions": {"provider": "vllm", "imageEntrypoint": "/usr/local/bin/pixel-dsv4-serve", "tensorParallelSize": 2, "maxSequences": 16, "gpuMemoryUtilizationPermille": 984, "dtype": "auto", "enforceEager": false, "reasoningParser": "deepseek_v4", "toolCallParser": "deepseek_v4"},
  "boundary": "Owner-private contained model backend configuration only. It selects exact local bytes and bounded runtime resources but grants no hashing result, policy mutation, model qualification, execution, container or network creation, image pull, device access, credential, external-effect, or completion authority."
}