indexing-and-querying
Build and query LlamaIndex indexes, persist and reload storage, configure retrievers/query engines/response synthesizers, and choose advanced retrieval patterns.
- License 许可
- MIT
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- indexing-and-querying
- license
- MIT
- metadata
- { "disco-role": "operating" }
- description
- Build and query LlamaIndex indexes, persist and reload storage, configure retrievers/query engines/response synthesizers, and choose advanced retrieval patterns.
- disable-model-invocation
- true
02 资源文件 4
- 参考文档references/api-reference.md8.5 KB
- 参考文档references/index-query-recipes.md9.8 KB
- 参考文档references/troubleshooting.md7.7 KB
- 脚本scripts/build_tiny_mock_index.py1.5 KB
03 SKILL.md 正文
Indexing and Querying
Use this sub-skill when a user needs practical LlamaIndex RAG mechanics after data has already been loaded or chunked into Document or BaseNode objects.
Owns
- Build
VectorStoreIndexandSummaryIndexfrom documents or nodes. - Persist and reload indexes with
StorageContextandload_index_from_storage. - Configure
as_retriever,as_query_engine,VectorIndexRetriever,RetrieverQueryEngine, and response synthesizers. - Compose advanced retrieval with
QueryFusionRetriever,RouterRetriever,RecursiveRetriever,RouterQueryEngine, andSubQuestionQueryEngine. - Diagnose indexing/query failures such as missing models, empty retrieval, persistence path mistakes, filters, tool metadata, and async/synthesis mode issues.
Route Elsewhere
- Use
../ingestion-and-loading/for file readers, loaders, chunking,SentenceSplitter,IngestionPipeline, metadata extraction, and node creation. - Use
../integrations-and-storage/for provider-specific vector stores, database/cloud installs, and integration-specific credentials. - Use
../customization-and-structured-outputs/for prompt templates, structured outputs, Pydantic output classes, and output formatting.
Quick Start
from llama_index.core import Document, Settings, VectorStoreIndex
from llama_index.core.embeddings.mock_embed_model import MockEmbedding
from llama_index.core.llms.mock import MockLLM
Settings.embed_model = MockEmbedding(embed_dim=8)
Settings.llm = MockLLM(max_tokens=64)
index = VectorStoreIndex.from_documents([
Document(text="LlamaIndex builds indexes over nodes."),
Document(text="Retrievers fetch relevant nodes for a query engine."),
])
query_engine = index.as_query_engine(similarity_top_k=2)
print(query_engine.query("What do retrievers do?"))
For a no-network smoke test, run the bundled script:
python sub-skills/indexing-and-querying/scripts/build_tiny_mock_index.py
Workflow
- Confirm the user already has
Documentor node objects; if not, route loading/chunking toingestion-and-loading. - Choose index type:
VectorStoreIndexfor semantic similarity retrieval;SummaryIndexfor small ordered corpora, summaries, or query-all behavior. - Set explicit
Settings.embed_modelandSettings.llmor pass models directly to constructors/query engines; avoid accidental remote defaults when the environment has no API keys. - Build with
VectorStoreIndex.from_documents(documents)orVectorStoreIndex(nodes=nodes); constructors accept nodes, not rawDocumentobjects. - Query simply with
index.as_query_engine(...), or split retrieval/synthesis withindex.as_retriever(...),get_response_synthesizer(...), andRetrieverQueryEngine(...). - Persist using
index.storage_context.persist(persist_dir=...); reload withStorageContext.from_defaults(persist_dir=...)andload_index_from_storage(...). - For multi-index systems, wrap engines/retrievers as tools with clear metadata before using routers or sub-question engines.
References
references/index-query-recipes.md: practical recipes for indexes, persistence, retrieval, routers, fusion, and synthesis.references/api-reference.md: verified constructor signatures and import paths for indexing/query APIs.references/troubleshooting.md: failure-mode diagnosis for embeddings, LLMs, storage, filters, routing, async, and optional integrations.
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).