parallel

Name: parallel
Author: ducktalez

Full diagnosis: `docs/PARALLEL_BENCHMARK_DIAGNOSIS.md`

by ducktalez(0) 0 installs

Free

Free account. Installing gives you the manifest plus copy-paste snippets.

See reviews

About

Instruction file imported from ducktalez/plagih (.github/instructions/parallel.instructions.md). Copyright stays with the author.

Parallel – Copilot Instructions

Key constraints

Windows pickle (P4): Strategies and error metrics must be top-level functions (no lambdas, closures). _ClipAutocast is a picklable class for this reason.
repair_all() after IPC (P1): Trees received from workers have stripped back-references. Already handled in _update_worker_state() and run_generation_parallel().
Worker count (P11): Use cpu_count_physical() — logical threads hurt CPU-bound work. Current sweet spot: 8 workers on 8-core machine.
Batch size: Sweet spot 32..128 tasks per batch (see diagnosis report).
Relational-on-Piecewise (P12): Workers use small runtime batches + timeout to surface pathological SymPy hangs.

Full diagnosis: docs/PARALLEL_BENCHMARK_DIAGNOSIS.md

Use it

Copy one of these into your project. Installing also returns the manifest and these snippets.

yaml

targets:
  - https://api.opensmartroute.ai/api/v1/registry/ducktalez-plagih-parallel-instructions/manifest   # or paste the manifest below

Manifest

An Open Capability Manifest: the router reads it to know what this does, what it costs and when to pick it.

ducktalez-plagih-parallel-instructions.ocm.jsonjson

{
  "ocm": "1",
  "id": "ducktalez-plagih-parallel-instructions",
  "kind": "skill",
  "name": "parallel",
  "description": "Full diagnosis: `docs/PARALLEL_BENCHMARK_DIAGNOSIS.md`",
  "publisher": "ducktalez",
  "version": "1.0.0",
  "capabilities": {
    "domains": [
      "medical"
    ],
    "tags": [
      "skill-md",
      "github-instructions"
    ],
    "languages": [
      "en"
    ]
  },
  "quality_prior": 0.6,
  "examples": [
    "Full diagnosis: `docs/PARALLEL_BENCHMARK_DIAGNOSIS.md`"
  ],
  "primary": false,
  "metadata": {
    "source": {
      "provider": "github-instructions",
      "repository": "https://github.com/ducktalez/plagih",
      "path": ".github/instructions/parallel.instructions.md",
      "ref": "931a98d8f29283e1b97aa06df6359c4d4f831feb",
      "url": "https://github.com/ducktalez/plagih/blob/931a98d8f29283e1b97aa06df6359c4d4f831feb/.github/instructions/parallel.instructions.md",
      "key": "ducktalez/plagih/.github/instructions/parallel.instructions.md"
    },
    "applies_to": "plagih/parallel.py"
  },
  "instructions": "# Parallel – Copilot Instructions\n\n## Key constraints\n\n- **Windows pickle** (P4): Strategies and error metrics must be top-level functions\n  (no lambdas, closures). `_ClipAutocast` is a picklable class for this reason.\n- **`repair_all()` after IPC** (P1): Trees received from workers have stripped\n  back-references. Already handled in `_update_worker_state()` and\n  `run_generation_parallel()`.\n- **Worker count** (P11): Use `cpu_count_physical()` — logical threads hurt\n  CPU-bound work. Current sweet spot: 8 workers on 8-core machine.\n- **Batch size**: Sweet spot **32..128 tasks per batch** (see",
  "cost": {
    "context_tokens": 198
  }
}

Fetch it by URL: GET /api/v1/registry/ducktalez-plagih-parallel-instructions/manifest?version=1.0.0

About

Parallel – Copilot Instructions

Key constraints

Use it

Manifest

Reviews