rag-document-processing
Guides LazyLLM RAG, document ingestion, readers, transforms, retrievers, rerankers, stores, BM25, and parser-service workflows.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- rag-document-processing
- license
- Apache 2.0
- metadata
- { "disco-role": "operating" }
- description
- Guides LazyLLM RAG, document ingestion, readers, transforms, retrievers, rerankers, stores, BM25, and parser-service workflows.
- disable-model-invocation
- true
02 资源文件 4
- 参考文档references/data-and-service-formats.md2.6 KB
- 参考文档references/rag-workflows.md3.1 KB
- 参考文档references/troubleshooting.md2.3 KB
- 脚本scripts/rag_bm25_smoke.py1.8 KB
03 SKILL.md 正文
LazyLLM RAG and Document Processing
Use this sub-skill for LazyLLM tasks involving Document, DocNode, document readers, transforms, node groups, retrievers, rerankers, BM25, stores/indexes, document service metadata, parser service, Milvus/OpenSearch, or RAG examples.
Start here when
- The task mentions
lazyllm.tools.rag,Document,Retriever,Reranker,DocNode, node groups, chunking, embedding, reranking, or BM25. - An import error says to run
lazyllm install rag. - The user wants a local RAG app, parser-service architecture, URL/file retriever, map store, Milvus/OpenSearch index, or RAG with online/local model modules.
- You need to separate local text retrieval from optional embedding/vector database/model backends.
Files to read
- rag-workflows.md for local RAG recipes, object relationships, and backend escalation.
- data-and-service-formats.md for document/node/service request shapes and parser-service boundaries.
- troubleshooting.md for missing dependencies, parser service, vector DB, and retrieval issues.
- scripts/rag_bm25_smoke.py for a no-network BM25 retrieval check.
Safe RAG workflow
- Install/check the RAG extra when importing RAG objects:
lazyllm install rag python ../../scripts/check_lazyllm_env.py --require-rag - Start local and text-only. Use
DocNodeand BM25 or local readers before adding embeddings, rerankers, vector DBs, or model modules. - Decide ownership of model nodes. Model modules used for embeddings/reranking/chat belong to model-deployment for backend checks.
- Decide service boundaries. Parser service, document service, Milvus/OpenSearch/Redis, object storage, and external URLs require service credentials or local fixtures.
- Run local smoke.
python scripts/rag_bm25_smoke.py
Key public surfaces
Document(*args, **kw)is the main RAG container and index manager.Retriever(doc, group_name, similarity=None, similarity_cut_off=-inf, index='default', topk=6, embed_keys=None, target=None, output_format=None, join=False, weight=None, priority=None, **kwargs)pulls contexts from a document/index.Reranker(name='ModuleReranker', *args, **kwargs)reranks retrieved contexts and may depend on a model/backend.DocNoderepresents text chunks with metadata and relationships.- BM25 supports local English/Chinese retrieval with no embedding server.
Backend posture
- Required for local guidance: base LazyLLM plus
ragextra. - Safe CPU candidates: BM25, document/node transforms, reader utilities, retriever logic, local doc-service manager with SQLite/mocked parser client.
- Optional external services: Milvus, OpenSearch/Elasticsearch, Redis/RedisVL, parser service workers, object storage, remote URLs.
- Optional model/GPU: embedding models, rerankers, local LLM answer generation, multimodal OCR/audio ingestion.
Handoff checklist
When you complete a RAG task, report:
- document source type and data format,
- node/chunking/grouping plan,
- retrieval method and index/store backend,
- model/embedding/reranker backend status,
- local smoke command or reason it is optional/skipped,
- clear separation between local artifacts and external services.
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).