graphormer
Routes Graphormer users to fairseq training, dataset customization, pretrained evaluation, model extension, and DiG workflows.
- License 许可
- MIT
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- graphormer
- license
- MIT
- metadata
- { "disco-role": "operating" }
- description
- Routes Graphormer users to fairseq training, dataset customization, pretrained evaluation, model extension, and DiG workflows.
- disable-model-invocation
- true
02 资源文件 6
- 参考文档references/installation-and-environment.md3.4 KB
- 参考文档references/repo-provenance.md1.7 KB
- 参考文档references/repo-routing-metadata.json501 B
- 参考文档references/source-script-map.md3.6 KB
- 参考文档references/troubleshooting.md4.6 KB
- 脚本scripts/check_graphormer_environment.py12.1 KB
03 SKILL.md 正文
Graphormer
Graphormer is a fairseq user-dir graph learning package for molecular modeling,
materials-style structure tasks, and a small set of research-only DiG
workflows. Use this skill when the task mentions Graphormer, Graphormer3D,
fairseq --user-dir, ZINC, PCQM4M, MolHIV, OC20, IS2RE, catalyst adsorption,
protein conformation sampling, protein-ligand diffusion, custom Graphormer
datasets, or Graphormer pretrained checkpoints.
This skill is a router, not a monolithic manual. Start with the most specific sub-skill that matches the workflow, then read the nearest bundled reference or helper script.
Start here
- Installation and environment
- Cross-cutting troubleshooting
- Source script map
- Repository provenance
- Environment check
Route map
- fairseq-training: Graphormer property-prediction and Graphormer3D training command templates for ZINC, PCQM4M, MolHIV FLAG, and OC20/IS2RE.
- datasets-and-customization: built-in dataset selection, custom dataset registration, preprocessing, and batch-schema validation.
- pretrained-and-evaluation: pretrained checkpoint selection, fine-tuning, and evaluation.
- model-extension: Graphormer fairseq registries, model/task/criterion wiring, and extension recipes.
- distributional-graphormer: optional DiG catalyst, property-guided, protein, and protein-ligand workflows.
Minimal environment check
If you only want to confirm that a Graphormer-compatible environment is ready, run the bundled checker from a shell that can see the target environment:
python scripts/check_graphormer_environment.py --user-dir <graphormer-package-dir> --format text
Add --require-complete when you want the script to confirm the expected
Graphormer model, task, criterion, and architecture registries. Add
--require-cuda on a CUDA host when you also want a tiny device smoke check.
What this skill assumes
- Graphormer is loaded through fairseq's
--user-dirmechanism. - The user-dir points at the directory that contains Graphormer's
models/,tasks/, andcriterions/packages. - Full native training, dataset download, checkpoint download, or DiG runs are optional and should be handled by a Researcher session only when the required data, hardware, and time budget are available.
How to choose a sub-skill
- Need a training command: start with
fairseq-training. - Need to select a dataset or validate a custom data module: start with
datasets-and-customization. - Need to load a checkpoint or compare evaluation metrics: start with
pretrained-and-evaluation. - Need to add or inspect a registered model/task/criterion/architecture:
start with
model-extension. - Need the DiG catalyst, protein, or protein-ligand research code: start with
distributional-graphormer.
When to read provenance
Read Repository provenance before treating this skill as current or before refreshing it against a newer Graphormer checkout.
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).