tracking-and-registry
Use this MLflow sub-skill for experiment tracking, runs, params, metrics, tags, artifacts, datasets, search, tracking URI and store setup, fluent API versus MlflowClient decisions, and model registry workflows.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- tracking-and-registry
- license
- Apache 2.0
- metadata
- { "disco-role": "operating" }
- description
- Use this MLflow sub-skill for experiment tracking, runs, params, metrics, tags, artifacts, datasets, search, tracking URI and store setup, fluent API versus MlflowClient decisions, and model registry workflows.
- disable-model-invocation
- true
02 资源文件 4
- 参考文档references/api-reference.md8.6 KB
- 参考文档references/troubleshooting.md7.3 KB
- 参考文档references/workflows.md8.2 KB
- 脚本scripts/tracking_smoke.py5.8 KB
03 SKILL.md 正文
Tracking and Registry
Use This When
- Building or debugging MLflow experiment tracking with experiments, runs, params, metrics, tags, inputs, datasets, or artifacts.
- Choosing between the fluent
mlflow.*tracking API andmlflow.MlflowClient/mlflow.tracking.MlflowClientfor explicit CRUD workflows. - Configuring local SQLite, HTTP, Databricks, legacy file-store opt-out, or other supported tracking and registry URIs.
- Searching runs, experiments, logged models, registered models, or model versions.
- Registering a logged model, managing registered model versions, tags, descriptions, aliases, and stage-compatible lifecycle operations.
Route Elsewhere
- Model flavor logging, model signatures,
MLmodelcontents, packaging, loading, and prediction belong inmodels-and-flavors. - Starting or securing tracking servers, auth, deployment, projects, and serving belong in
serving-and-projects. - Traces, prompts, GenAI evaluation datasets, trace locations, and prompt registry work belong in
genai-observability.
Start Here
- Read
references/api-reference.mdfor exact entry points, URI behavior, search syntax, and registry capabilities. - Read
references/workflows.mdfor copy-ready tracking and registry patterns. - Read
references/troubleshooting.mdwhen repairing active-run, async logging, artifact URI, search, registry, workspace, or Databricks credential failures. - Run
scripts/tracking_smoke.py --help, thenpython scripts/tracking_smoke.pyto verify that the installedmlflowpackage can track locally without any source checkout.
Key Decision Rules
- Use the fluent API for in-process training code where an active run is natural; use
MlflowClientwhen operating on explicit experiment IDs, run IDs, registry names, pages, or lifecycle actions. - Set
mlflow.set_tracking_uri(...)before starting runs. It does not move an already-active run to a new store. - For local smoke and registry work, prefer a SQLite tracking URI such as
sqlite:///mlflow.db; MLflow 3 places filesystem tracking backends in maintenance mode unlessMLFLOW_ALLOW_FILE_STORE=trueis deliberately set. - Keep tracking URI and registry URI separate in debugging:
MlflowClient(registry_uri=...)andmlflow.set_registry_uri(...)can target a different registry than the tracking backend. - Treat Databricks tracking URIs specially: registry URI defaults toward Unity Catalog (
databricks-uc) unless explicitly overridden.
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).