AgentSkillsCN

hpc-context

Mycelia 基准测试与大规模分析的 SLURM 及 HPC 指导

SKILL.md
--- frontmatter
name: hpc-context
description: SLURM and HPC guidance for Mycelia benchmarks and large-scale analysis

Mycelia HPC Context

Guidance for running Mycelia on HPC systems with SLURM.

Benchmark Commands

Local small/medium/large benchmarks:

bash
julia --project=. benchmarking/benchmark_runner.jl small
julia --project=. benchmarking/benchmark_runner.jl medium
julia --project=. benchmarking/benchmark_runner.jl large

SLURM submission (resource-intensive):

bash
sbatch benchmarking/run_all_benchmarks.sh

Key Directories

  • •benchmarking/ - Heavy benchmark scripts
  • •results/ - Generated artifacts
  • •ci/hpc/ - HPC-specific CI configuration

Resource Considerations

  • •Use memory estimation utilities from utility-functions.jl for large-scale analyses
  • •Keep external tool calls isolated in helpers under src/
  • •Avoid committing generated outputs or large files
  • •Note external tool or dataset requirements explicitly

Portability

  • •Use joinpath for all file paths
  • •Test on multiple environments when possible
  • •Document any HPC-specific requirements
  • •Consider job scheduler compatibility (SLURM, PBS, etc.)

Tutorial/Benchmark Guidelines

  • •Do NOT add generally useful functions in tutorials/, test/, or benchmarking/
  • •If a helper is broadly useful, add it under src/ and call from tutorial
  • •Tutorials demonstrate core Mycelia functionality
  • •Avoid defining new helper functions except true one-offs
  • •Prefer Rhizomorph graph/assembly functionality over legacy base-graph APIs