AGENT SKILL

data-prep-and-generation

Prepare, validate, convert, and self-generate xTuring datasets for text, instruction, and preference workflows.

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

01 前言 Front Matter

name
data-prep-and-generation
license
Apache 2.0
metadata
{ "disco-role": "operating" }
description
Prepare, validate, convert, and self-generate xTuring datasets for text, instruction, and preference workflows.
disable-model-invocation
true

02 资源文件 5

03 SKILL.md 正文

data-prep-and-generation

Use this sub-skill when you need to build or validate xTuring data for text completion, instruction tuning, preference/DPO training, or API-backed self-instruct generation.

Route here for

  • TextDataset, InstructionDataset, PreferenceDataset, and ListPromptTemplate
  • JSON, JSONL, and saved-dataset layouts for the supported dataset classes
  • Alpaca-to-xTuring conversion and dataset persistence
  • Self-instruct generation through InstructionDataset.generate_dataset(...) and generate_dataset_from_dir(...)
  • OpenAI, Cohere, and Claude wrapper caveats used by generation helpers
  • Dataset schema troubleshooting before training or evaluation

Route away

  • Model loading, model registry choices, or generation from a saved model → models-and-inference
  • Fine-tuning, LoRA, INT8, or DPO training loops → training-and-alignment
  • CLI, API server, or UI serving → cli-api-ui
  • Evaluation harnesses or result persistence → evaluation
  • Text2ImageDataset is not implemented and should not be used as a data route

Read first

  • references/data-formats.md for supported schemas, constructors, JSONL shapes, and save behavior
  • references/self-instruct-and-apis.md for seed-task format, cache files, wrapper behavior, and text-extraction caveats
  • references/troubleshooting.md for schema assertions, malformed input, missing dependencies, and API retry failures

Skill-owned scripts

  • scripts/convert_alpaca_json.py — convert Alpaca JSON into a saved Hugging Face dataset compatible with InstructionDataset
  • scripts/validate_xturing_dataset.py — validate a candidate dataset against the text, instruction, or preference schema

Typical workflow

  1. Inspect the raw file and decide whether it is text, instruction, or preference data.
  2. Validate the schema and fix missing or extra fields before conversion.
  3. Convert or load the data into a Hugging Face dataset and save it to disk when needed.
  4. Use self-instruct generation only when you have a supported text-generation API wrapper and the required credentials.
  5. Revisit the troubleshooting guide if the constructor, JSONL loader, or API helper rejects the data.

Cross-links

  • If the data is valid but training fails later, switch to training-and-alignment.
  • If the request is actually about model selection or inference behavior, switch to models-and-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