AGENT SKILL

bert-pytorch

Use BERT-pytorch for corpus preparation, vocabulary building, and tiny BERT pretraining workflows.

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

01 前言 Front Matter

name
bert-pytorch
license
Apache 2.0
metadata
{ "disco-role": "operating" }
description
Use BERT-pytorch for corpus preparation, vocabulary building, and tiny BERT pretraining workflows.
disable-model-invocation
true

02 资源文件 7

03 SKILL.md 正文

bert-pytorch

Use this skill when the user mentions BERT-pytorch, bert, bert-vocab, WordVocab, BERTDataset, BERT, BERTLM, or BERTTrainer.

Quick start

  1. Install the package with python -m pip install bert-pytorch if it is not already available in your Python environment.
  2. Run python scripts/check_install.py to confirm the package import, public exports, and console commands.
  3. If the task is about corpus layout, vocabulary files, or dataset rows, go to sub-skills/data-preparation/SKILL.md.
  4. If the task is about model construction, training, device choice, or checkpoints, go to sub-skills/training/SKILL.md.
  5. Use python scripts/make_tiny_corpus.py --output /tmp/bert-pytorch-corpus.txt when you need a stable two-line fixture.

Route map

  • sub-skills/data-preparation/SKILL.md: build or load vocabularies, validate corpus format, inspect BERTDataset, and diagnose pickle or row-format problems.
  • sub-skills/training/SKILL.md: instantiate BERT, BERTLM, and BERTTrainer; train on a tiny corpus/vocab pair; pick CPU or CUDA; and save checkpoints.

Shared facts

  • The corpus format is two sentences per line separated by a tab.
  • Tokenization happens before the package; BERT-pytorch does not tokenize raw text for you.
  • bert-vocab and bert are the documented console commands.
  • BERT hidden size must be divisible by the number of attention heads.
  • The bert CLI parses some boolean flags with type=bool; use the bundled smoke scripts or the Python API when you need an explicit CPU run or streaming mode.

Shared references

  • references/repo-provenance.md: source commit, package version, and evidence snapshot.
  • references/repo-routing-metadata.json: router metadata for managed import.
  • references/api-reference.md: public classes, functions, and object relationships.
  • references/cli-reference.md: command names, flags, defaults, and safety notes.
  • references/troubleshooting.md: cross-cutting install, import, CLI, corpus, and device issues.

Shared scripts

  • scripts/check_install.py: verify imports, API exports, and CLI help.
  • scripts/make_tiny_corpus.py: create a deterministic two-line corpus fixture.

When in doubt

  • Start with the sub-skill that matches the first concrete noun in the request: corpus/vocab/data rows -> data preparation; model/training/checkpoint/device -> training.
  • If the task mixes both, build the vocabulary first, then move to training.
  • Read references/troubleshooting.md before guessing about missing files, malformed rows, or device failures.

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