AGENT SKILL

adaptation-and-training

Route AdaptLLM, Instruction Pre-Training, PDS data selection, ResLoRA, and Learning Law tasks into safe planning and self-contained helpers.

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

01 前言 Front Matter

name
adaptation-and-training
license
MIT
metadata
{ "disco-role": "operating" }
description
Route AdaptLLM, Instruction Pre-Training, PDS data selection, ResLoRA, and Learning Law tasks into safe planning and self-contained helpers.
disable-model-invocation
true

02 资源文件 7

03 SKILL.md 正文

adaptation-and-training

Use this sub-skill for domain adaptation, instruction augmentation, data-selection training, ResLoRA wrapper planning, and Learning Law policy optimization.

Use when

  • A task asks to convert raw domain text into reading-comprehension style data.
  • A task asks to synthesize instruction-augmented corpora with a vLLM-based planner.
  • A task asks to plan or stage the optimal-control PDS pipeline.
  • A task asks to validate ResLoRA flag combinations or target-module mapping.
  • A task asks to understand Learning Law optimization or evaluation stages.

Route elsewhere

  • MiniLLM, DPKD, or Tuna distillation tasks go to ../distillation-and-post-training/SKILL.md.
  • VeRL-style RL or experiential-learning tasks go to ../rl-experiential-learning/SKILL.md.

What this sub-skill provides

  • references/domain-adaptation-workflows.md for AdaptLLM-style corpus conversion and domain inference planning.
  • references/instruction-and-data-selection.md for instruction augmentation and the PDS stage map.
  • references/reslora-and-learning-law.md for ResLoRA and Learning Law planning.
  • references/troubleshooting.md for cross-cutting failures and exclusion boundaries.
  • scripts/raw_to_reading_comprehension.py for a tiny, safe corpus transformer.
  • scripts/pds_pipeline_planner.py for a stage checklist and path-placeholder validator.
  • scripts/reslora_config_check.py for ResLoRA config validation.

Operating rule

Keep heavy training, large data downloads, remote inference, and cluster launches out of this skill. Use the bundled scripts to plan or validate inputs, then hand off full execution to a later researcher session only when the required environment is already available.

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