Module `neuralops.runners`

NeuralOps Runners — pipeline orchestration, continuous sweeps, and monitoring.

Subpackages

module_sweeps Per-module sweep scripts (run_all_4_<module>.py) that run the full four-model benchmark suite for each attention / compression module.

Top-level modules

unified_orchestrator Central orchestration backend managing multi-model sweep scheduling. run_master_pipeline Master pipeline orchestrator with idempotent task queuing. run_autonomous_48h_pipeline 48-hour autonomous multi-stream GPU pipeline. run_continuous_orchestrator Continuous GPU orchestrator for sustained high-utilisation training. run_monitor_server Production telemetry + benchmark database web server.

Sub-modules

neuralops.runners.module_sweeps
neuralops.runners.monitor: NeuralOps Telemetry and Documentation Studio Server package.
neuralops.runners.run_autonomous_48h_pipeline: Autonomous 48-Hour Multi-Stream High-Performance GPU Pipeline. Executes all 6 benchmark stages CONCURRENTLY across 6 parallel worker streams: Stream …
neuralops.runners.run_continuous_orchestrator: Autonomous 48-Hour Continuous GPU Orchestrator. Supervises continuous multi-worker execution across all models and modules: - Ensures GPU compute …
neuralops.runners.run_master_pipeline: Master pipeline orchestrator. Runs 2 workers total across all task types. Priority order: zeroshot > rank_sweep > c4_ppl > latency > multiseed. …
neuralops.runners.run_monitor_server: Production-grade NeuralOps Telemetry, Benchmark Database & Documentation Studio Server …
neuralops.runners.run_sweep: Master runner for module sweeps. Replaces the 200+ duplicated run_all_4_*.py scripts.
neuralops.runners.unified_orchestrator: Unified Orchestrator Backend for NeuralOps.

Classes

class UnifiedOrchestrator (default_args: list | None = None)

Expand source code

class UnifiedOrchestrator:
    """Class wrapper for NeuralOps Unified Orchestrator."""

    def __init__(self, default_args: list | None = None) -> None:
        self.default_args = default_args or []

    def run(self, module_key: str, args: list | None = None) -> None:
        merged = list(self.default_args) + (args or [])
        run_orchestrator(module_key, merged)

Class wrapper for NeuralOps Unified Orchestrator.

Methods

def run(self, module_key: str, args: list | None = None) ‑> None

Expand source code

def run(self, module_key: str, args: list | None = None) -> None:
    merged = list(self.default_args) + (args or [])
    run_orchestrator(module_key, merged)