configuration-and-data
Build, inspect, validate, and round-trip experiment configs and datasets for all supported H2O LLM Studio problem types.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- configuration-and-data
- license
- Apache 2.0
- metadata
- { "disco-role": "operating" }
- description
- Build, inspect, validate, and round-trip experiment configs and datasets for all supported H2O LLM Studio problem types.
- disable-model-invocation
- true
02 资源文件 6
- 参考文档references/configuration-reference.md6.9 KB
- 参考文档references/data-formats.md6.4 KB
- 参考文档references/problem-types.md5.7 KB
- 参考文档references/troubleshooting.md5.0 KB
- 脚本scripts/inspect_config.py8.1 KB
- 脚本scripts/validate_dataset.py9.0 KB
03 SKILL.md 正文
configuration-and-data
Use this sub-skill when you need to load, inspect, validate, or round-trip H2O LLM Studio experiment configs and dataset files.
Owns
- YAML experiment configs and nested dataclass sections.
- Problem type selection and per-problem defaults.
- Dataset connectors, local CSV / Parquet file rules, and import-time column mapping.
- Prompt, answer, system, parent-id, and DPO rejection columns.
- Conversation chains and default seeded datasets.
- Config checks that happen before training.
Does not own
- Training execution, CLI launches, or GPU setup. Use
training-and-experiments. - Model, loss, metric, and plot internals. Use
modeling-and-evaluation. - GUI navigation or app startup. Use
app-and-ui.
Supported problem types
- Causal language modeling
- Causal classification modeling
- Causal regression modeling
- Sequence-to-sequence modeling
- DPO modeling
Read first
references/configuration-reference.mdreferences/data-formats.mdreferences/problem-types.mdreferences/troubleshooting.md
Skill-owned scripts
scripts/inspect_config.py— load a YAML config, report the resolved problem type, and verify a canonical round-trip.scripts/validate_dataset.py— load the config, validate the referenced dataset files, and inspect chain and label rules.
Typical workflow
- Confirm the intended problem type.
- Load the YAML config and compare it with the expected schema.
- Validate the train and validation datasets, including required columns.
- Check the config round-trip before handing the experiment off to training.
Cross-links
- If the task needs training or checkpoint debugging, switch to
training-and-experiments. - If the task is about loss curves, metrics, or predictions, switch to
modeling-and-evaluation. - If the task is about import dialogs, dataset pages, or app screens, switch to
app-and-ui.
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).