AGENT SKILL

jittor

Guide Jittor package workflows for tensor programming, training, data/model I/O, custom ops, runtime validation, and troubleshooting.

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

01 前言 Front Matter

name
jittor
license
Apache 2.0
metadata
{ "disco-role": "operating" }
description
Guide Jittor package workflows for tensor programming, training, data/model I/O, custom ops, runtime validation, and troubleshooting.
disable-model-invocation
true

02 资源文件 7

03 SKILL.md 正文

Jittor repo skill

Use this repo skill when a task involves the Jittor Python package or repository. It is a router, not a full manual.

Before relying on this skill for a checkout or installed package, read references/repo-provenance.md to compare the source commit, package version, and evidence paths. For setup and the first import check, read references/installation-and-runtime.md, then run scripts/check_jittor_env.py against the active Python environment.

Install baseline

For normal package use, either install the published package or install from a checkout:

python -m pip install jittor
# or, from a checkout
python -m pip install -e .

The public metadata and docs for this repo assume a working C++ compiler and the runtime dependencies listed in the package metadata.

Minimal import check

python scripts/check_jittor_env.py

That smoke should import jittor, run one tiny CPU operation, and report whether CUDA is actually available to Jittor.

Route map

Shared references

Operating rules

  1. Prefer the smallest route that matches the task family.
  2. Treat CPU importability as the baseline; do not claim CUDA, ROCm, MPI, or other accelerator support unless the matching smoke or native verification has passed.
  3. If a task spans multiple route families, start with the most infrastructure-sensitive sub-skill, usually runtime-and-installation.
  4. Use bundled skill files only; do not tell future agents to reopen docs, examples, tests, or scripts from the original repository checkout.
  5. If the current checkout no longer matches the provenance snapshot, refresh the skill before reusing it.

Quick routing hints

  • Tensor, gradient, or execution semantics: core-api-and-autograd.
  • Layers, optimizers, schedulers, or training loops: nn-training-workflows.
  • Datasets, transforms, model zoo, or checkpoint I/O: datasets-models-and-io.
  • Custom operators, console embedding, or utility CLI flags: custom-op-console-and-tools.
  • Install, compiler, cache, profiling, backend flags, or timing: runtime-and-installation.

Why this skill is organized this way

Jittor has a broad surface area, but most user questions cluster into a few repeated families: runtime setup, core tensor semantics, training, data/model I/O, and advanced extension tooling. The sub-skills keep those families separate so a later agent can answer a natural request without rereading the full repository.

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