AGENT SKILL

model-config-and-api

Inspect and modify AlphaFold model presets, RunModel configuration, feature processing, parameter loading, and backend dependency issues.

License 许可
Apache-2.0
最近更新
2026-09-03
Stars
★ 351
Forks
27

01 前言 Front Matter

name
model-config-and-api
license
Apache 2.0
metadata
{ "disco-role": "operating" }
description
Inspect and modify AlphaFold model presets, RunModel configuration, feature processing, parameter loading, and backend dependency issues.
disable-model-invocation
true

02 资源文件 4

03 SKILL.md 正文

AlphaFold Model Config and API

Use this sub-skill when a task is about AlphaFold's programmatic model layer: choosing model presets, inspecting or editing ml_collections configs, constructing RunModel objects, processing model features, loading Haiku parameters, diagnosing JAX/Haiku/TensorFlow imports, or using model-side geometry and lDDT helper APIs.

Start Here

  • Run scripts/inspect_model_presets.py to list installed model presets and selected configuration fields without loading weights or running inference.
  • Read references/model-overview.md for preset selection, config-editing patterns, RunModel lifecycle, dependency roles, and weight compatibility rules.
  • Read references/api-reference.md for model config APIs, RunModel, parameter loading, feature processing, and numerical helper contracts.
  • Read references/troubleshooting.md for import failures, JAX/JAXLIB mismatch, parameter path mismatches, TensorFlow/NumPy pin conflicts, GPU memory caveats, and preset/weight confusion.

Routing Boundaries

  • Use this sub-skill for alphafold.model.config, alphafold.model.model.RunModel, alphafold.model.data.get_model_haiku_params, alphafold.model.features, model geometry helpers, lDDT/FAPE helpers, and dependency/backend diagnostics.
  • Use ../prediction-cli/ for building run_alphafold commands, CLI flags, database path flags, MSA reuse, random seeds, and output-directory orchestration.
  • Use ../outputs-and-confidence/ for interpreting ranking_debug.json, confidence_*.json, pae_*.json, pLDDT, PAE, pTM, ipTM, PDB, mmCIF, AFDB, or AlphaFold Server JSON outputs.
  • Use ../relaxation/ for Amber/OpenMM relaxation configuration, stereochemical cleanup, GPU relaxation, and relaxation failure modes.
  • Use ../input-data-and-formats/ for FASTA/MSA/template parsing and data-pipeline input formats before model feature processing.

Safe Operating Rules

  • Do not load model weights, instantiate expensive model runners, call RunModel.predict, or run JAX compilation as a routine validation step.
  • Treat model_config() inspection as safe; treat RunModel construction and feature processing as dependency-sensitive because JAX, Haiku, and TensorFlow may import or initialize runtime backends.
  • Keep model names and parameter files aligned: model_1_ptm expects params_model_1_ptm.npz, while multimer v3 preset names expect params_model_1_multimer_v3.npz through params_model_5_multimer_v3.npz.
  • Prefer explicit preset selection over manual ad hoc config edits, then make small config changes with with cfg.unlocked() or with dataclass_cfg.unfreeze() only when the task truly needs them.

Bundled Helper

python sub-skills/model-config-and-api/scripts/inspect_model_presets.py --json

The helper imports only alphafold.model.config, reads installed package metadata when available, prints preset/config summaries, and never opens parameter files or runs inference.

04 同仓库更多 Skills

VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-triton-opt

Make a single PassNet Triton kernel fast AND numerically faithful: performance model (when a replacement can win at all), block/grid/warp tuning, autotune policy, launch overhead, and per-op numeric recipes to pass the dtype baseline tolerances. Use when a pass matches and is correct but speedup below expected, or when correctness fails by small numeric margins.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-solve

END-TO-END playbook for solving one PassNet sample: analyze the computation graphs, decide the optimization strategy (what to fuse, what to replace, what to leave alone), drive the iteration loop, and maximize the sample score. This is the ENTRY skill — invoke it first for any PassNet optimization task; it tells you when to use passnet-pattern-fusion, passnet-triton-opt and passnet-feedback.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-skill

PassNet GPU kernel optimization via compiler passes. Design and implement Triton-based optimization passes, create pass files under ./pass_dir/, self-evaluate with pass_evaluator, and iterate to maximize GPU speedup.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-pattern-fusion

Author PassNet pass files: write patterns that actually MATCH the FX graph, pick fusion regions (which ops to absorb into one Triton kernel), and structure multi-pass submissions with the mandatory shared-dispatch architecture. Use when creating or fixing pattern()/replacement_args()/replacement_func() files, when a pass "failed to match", or when deciding how to fuse multiple kernels/ops.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-orchestrate

Round/batch planning and decision support for PassNet graph optimization. Use for multi-sample triage, worker allocation, eval-budget policy, and post-eval keep/revert decisions. For solving one concrete sample end to end, use passnet-solve as the entry skill; consult this skill only when you need broader planning or a specific decision gate.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-feedback

Fast, deterministic feedback for PassNet work: pre-flight pattern/match verification WITHOUT burning a GPU evaluation, per-node bottleneck analysis of a sample's graphs, and eval-log parsing into per-variant status + estimated score + failure classification. Use BEFORE every GPU evaluation (check_pattern), at the START of a sample (analyze_graph), and AFTER every evaluation (parse_eval_log).

★ 351