retrieval-and-materialization
Use for Feast offline/historical retrieval, online retrieval, point-in-time joins, materialization, push ingestion, saved datasets, local retrieval checks, and DQM/monitoring basics.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- retrieval-and-materialization
- license
- Apache 2.0
- metadata
- { "disco-role": "operating" }
- description
- Use for Feast offline/historical retrieval, online retrieval, point-in-time joins, materialization, push ingestion, saved datasets, local retrieval checks, and DQM/monitoring basics.
- disable-model-invocation
- true
02 资源文件 5
- 参考文档references/api-reference.md5.9 KB
- 参考文档references/data-formats.md4.9 KB
- 参考文档references/troubleshooting.md5.9 KB
- 参考文档references/workflows.md5.7 KB
- 脚本scripts/local_retrieval_smoke.py7.3 KB
03 SKILL.md 正文
Feast Retrieval and Materialization
Use this sub-skill when the user asks to retrieve training data, batch-score features, fetch online features for inference, materialize feature views into an online store, push streaming rows, save historical datasets, or diagnose null/stale retrieval results.
Route first
- Use this skill for
FeatureStore.get_historical_features,get_online_features,materialize,materialize_incremental,push, entity rows, feature refs,FeatureServiceretrieval, local file offline stores, and SQLite online stores. - Route feature object modeling,
Entity,FeatureView,FeatureService,PushSource, and field schema questions to../feature-definitions/SKILL.md. - Route feature repo initialization,
feature_store.yaml,feast apply,feast plan, registry paths, and CLI repo lifecycle to../feature-repos-and-cli/SKILL.md. - Route feature server, offline server, registry server, REST/gRPC, TLS, auth, and remote serving endpoints to
../servers-and-remote/SKILL.md. - Route vector document retrieval,
retrieve_online_documents, vector stores, and RAG retrievers to../rag-and-vector-search/SKILL.md.
Fast workflow
- Confirm the feature repo is applied and the registry is readable.
- For historical retrieval, build an entity dataframe with join keys and
event_timestamp, or use supported entity-lessstart_date/end_dateretrieval. - Select features with either a
FeatureServiceor feature refs such asdriver_hourly_stats:conv_rate. - For online retrieval, materialize the needed feature views first, then call
get_online_features(...).to_dict()or.to_df(). - If results are null, stale, or missing columns, check join keys, timestamp windows, TTL, materialization range, online-store config, and feature-view
online=Truesettings.
References
references/api-reference.mdfor SDK and CLI retrieval/materialization calls.references/workflows.mdfor historical, online, push, saved dataset, and DQM flows.references/data-formats.mdfor entity rows, entity dataframes, feature refs, outputs, and local stores.references/troubleshooting.mdfor common failures and validation signals.scripts/local_retrieval_smoke.pyfor a safe local repo checker and command planner.
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).