retrieval-representation
Use Simple Transformers representation, dense retrieval, DPR, BEIR/MSMARCO, hard-negative, and retrieval-evaluation workflows.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- retrieval-representation
- license
- Apache 2.0
- metadata
- { "disco-role": "operating" }
- description
- Use Simple Transformers representation, dense retrieval, DPR, BEIR/MSMARCO, hard-negative, and retrieval-evaluation workflows.
- disable-model-invocation
- true
02 资源文件 5
- 参考文档references/api-reference.md2.5 KB
- 参考文档references/data-formats.md1.8 KB
- 参考文档references/troubleshooting.md1.4 KB
- 参考文档references/workflows.md2.0 KB
- 脚本scripts/validate_retrieval_data.py2.6 KB
03 SKILL.md 正文
Simple Transformers Retrieval and Representation Sub-skill
Use this sub-skill when a task asks for sentence/word embeddings,
RepresentationModel.encode_sentences, dense retrieval with query/context
encoders, DPR-style training, hard negatives, BEIR/MSMARCO/TREC data, FAISS
indexes, or retrieval metrics.
Owns
RepresentationModelfor hidden-state/token/word/sentence representations.RetrievalModel/RetrievalArgsfor dense retrieval train/eval/predict.PretrainRetrievalModelas an advanced pretraining route.- Retrieval data schemas:
query_text,gold_passage, optionaltitle, prediction queries, BEIR/MSMARCO/TREC conversion. - Optional
faiss,pytrec_eval, andbeirbranches.
Route elsewhere
- Cross-encoder sentence-pair classification reranking: classification.
- T5/monoT5 text-to-text reranking: generative-workflows.
- Extractive QA after passage retrieval: token-and-qa.
Read first
- API reference for constructors and method map.
- Data formats before preparing retrieval or representation inputs.
- Workflows for encode, train/eval/predict, and optional evaluation routes.
- Troubleshooting for optional dependencies, cache/index files, hard-negative shapes, and dependency compatibility.
Validation helper
python scripts/validate_retrieval_data.py --task retrieval-csv --input train.csv
python scripts/validate_retrieval_data.py --task query-lines --input queries.txt
python scripts/validate_retrieval_data.py --task beir-corpus-jsonl --input corpus.jsonl
python scripts/validate_retrieval_data.py --task beir-queries-jsonl --input queries.jsonl
python scripts/validate_retrieval_data.py --task tsv --input train.tsv
The helper checks file shape only. It does not import FAISS, build indexes, download datasets, train encoders, or run BEIR evaluation.
Key decisions
- Use
RepresentationModelwhen the output is embeddings or token vectors. - Use
RetrievalModelwhen the task includes query/passage training, hard negatives, or retrieval prediction. - Decide whether evaluation needs built-in metrics only,
pytrec_eval, or full BEIR evaluation. - Install FAISS only for workflows that actually build or load FAISS indexes.
- Keep CPU for small smoke checks; CUDA is performance acceleration, not required for schema validation.
Verification status
Constructor signatures were inspected with compatibility warnings. Native representation/retrieval examples require checkpoint/dataset downloads and optional dependencies, so default verification uses validators and static/API evidence unless the user approves larger checks.
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).