aizynthfinder
Use AiZynthFinder for retrosynthetic planning, configuration, route analysis, custom extensions, and focused development workflows.
- License 许可
- MIT
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- aizynthfinder
- license
- MIT
- metadata
- { "disco-role": "operating" }
- description
- Use AiZynthFinder for retrosynthetic planning, configuration, route analysis, custom extensions, and focused development workflows.
- disable-model-invocation
- true
02 资源文件 4
- 参考文档references/repo-provenance.md1.8 KB
- 参考文档references/repo-routing-metadata.json340 B
- 参考文档references/troubleshooting.md3.9 KB
- 脚本scripts/check_aizynthfinder_env.py5.5 KB
03 SKILL.md 正文
AiZynthFinder Repo Skill
Use this skill when a user is working with AiZynthFinder, a retrosynthetic route-finding package for planning molecule syntheses with policy-guided tree search.
First Steps
- Confirm the user has AiZynthFinder installed in a Python
>=3.10,<3.13environment. - If the task mentions a config, stock, policy model, template library, public-data bundle, or optional service, start with configuration-and-data.
- If the task is to run retrosynthesis or one-step expansion, use planning-workflows.
- If the task starts from an existing
output.json.gz, HDF5 output,trees.json, checkpoint, route collection, or reaction tree, use route-analysis. - If the task is about custom scorers, stocks, policies, search algorithms, plugin strategies, hooks, or source edits, use extension-and-development.
Install and Import Check
For end users, install the public package with base dependencies unless the task needs optional features:
python -m pip install aizynthfinder
python - <<'PY'
from aizynthfinder.aizynthfinder import AiZynthFinder, AiZynthExpander
print(AiZynthFinder, AiZynthExpander)
PY
Use python -m pip install "aizynthfinder[all]" only when the task needs optional MongoDB, molbloom, route-distance, timeout, or similar extras. Use TensorFlow/gRPC-related extras only for TensorFlow Serving or remote model workflows.
Run scripts/check_aizynthfinder_env.py when diagnosing import, version, CLI, or optional-dependency availability without launching a search.
Sub-Skill Routes
- planning-workflows: build and run
aizynthclicommands, useAiZynthFinderorAiZynthExpander, handle single vs batch SMILES, checkpoints, multiprocessing, clustering flags, and GUI/notebook launch decisions. - configuration-and-data: write and validate YAML configs, stock files, policy/filter model references, public data commands,
smiles2stock,download_public_data, and optional dependency choices. - route-analysis: inspect
output.json.gz, HDF5 tables,trees.json, checkpoints,ReactionTree,RouteCollection, route scores, clustering, images, and result summaries. - extension-and-development: implement or validate custom hooks, stocks, scorers, policies, plugin expansion strategies, search algorithms, and focused source-checkout tests.
Shared References
- Read references/troubleshooting.md for cross-cutting install, optional dependency, model/stock asset, CLI/API, and backend failures before drilling into sub-skill-specific troubleshooting.
- Read references/repo-provenance.md before deciding whether this skill is current for an AiZynthFinder checkout or should be refreshed.
references/repo-routing-metadata.jsonis structured metadata used by the managed repo-skills-router during import.
Routing Guardrails
- Do not run
download_public_data, MongoDB stock creation, long retrosynthesis batches, GUI launches, or plugin service calls unless the user explicitly asks for those side effects. - Do not assume user assets exist. Configs normally require a stock file, an expansion policy model, and a reaction template library; filter policies are optional.
- Prefer static validation and CLI/API help checks before running searches with user data.
- Keep optional plugins and service-backed models separate from core install troubleshooting; missing Chemformer, ModelZoo, MongoDB, molbloom, route-distance, or TensorFlow Serving dependencies are feature-specific, not proof that core AiZynthFinder is broken.
04 同仓库更多 Skills
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.
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.
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.
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.
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.
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).