agentleFS
Sign inStart free
guojin-yanguojin-yan/DeploySharp/.agents/skills/deploysharp-backend-validation/SKILL.md

deploysharp-backend-validation

Skill179 starsChanged 12 days agoApache-2.0View on GitHub

Validate or diagnose an exact DeploySharp model and inference backend combination, including native runtime, dynamic shapes, parity, and support-matrix updates. Use for ONNX Runtime, OpenVINO, OpenCV DNN, TensorRT, or LlamaSharp backend work.

The file

We split this skill into the rules it gives an agent

7 rules. About 0 are common, 4 are distinctive, and 3 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈4

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 3

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Treat support as an exact tuple: model artifact, task, backend/provider, runtime version, device, input contract, and decoder.

    line 8
  2. Very specific:

    Read docs/articles/platform-support.md and the target row in docs/model-backend-verification-matrix.md.

    line 12
  3. Distinctive:

    Use ONNX Runtime CPU as the numerical baseline when applicable, then bind every target-backend input by name, type, rank, and concrete …

    line 14
  4. Very specific:

    Do not silently fall back from CUDA/TensorRT/OpenVINO GPU to CPU.

    line 22
  5. Very specific:

    Keep managed and native package versions, RID, architecture, CUDA/cuDNN/TensorRT, and driver details in the report.

    line 23
  6. Distinctive:

    When a third-party importer rejects a precise operator/shape, preserve the original diagnostic and mark that combination unsupported.

    line 25
  7. Distinctive:

    Keep raw logs, TRX, generated engines, and repeated JSON outside public docs.

    line 31

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

DeploySharp Backend Validation

Treat support as an exact tuple: model artifact, task, backend/provider, runtime version, device, input contract, and decoder.

Validation Flow

  1. Read docs/articles/platform-support.md and the target row in docs/model-backend-verification-matrix.md.
  2. Record the model and input SHA-256, runtime/native versions, OS, architecture, device, and source revision before testing.
  3. Use ONNX Runtime CPU as the numerical baseline when applicable, then bind every target-backend input by name, type, rank, and concrete shape.
  4. Separate model import, Session creation, input binding, execution, output copy, and decoder failures.
  5. Compare raw output shape and finite values before comparing decoded results. Use declared tolerances for scores and coordinates; compare discrete labels, tokens, text, and region order exactly when required.

Dynamic input support does not prove dynamic Batch. For flat NMS output, verify row separation through the count tensor. A Session pool must create independent native Sessions or execution contexts; never claim concurrency from queued calls on one stateful object.

Native Runtime Rules

  • Do not silently fall back from CUDA/TensorRT/OpenVINO GPU to CPU.
  • Keep managed and native package versions, RID, architecture, CUDA/cuDNN/TensorRT, and driver details in the report.
  • Build TensorRT engines on the target device with the tested profile.
  • When a third-party importer rejects a precise operator/shape, preserve the original diagnostic and mark that combination unsupported. Do not change decoder thresholds to hide import failures.

Evidence And Status

Contract/unit tests prove API behavior, not real model support. A smoke test proves execution, not accuracy or performance. Update the matrix only for the exact artifact tested: supported for real execution and result contract, unverified when not tested, unsupported when attempted and blocked, and not-applicable only when the combination has no valid meaning.

Keep raw logs, TRX, generated engines, and repeated JSON outside public docs. Public documentation should contain the current conclusion, environment, reproducible command, and limitation.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

Check your setup, free