embeddings-and-similarity
Dense SentenceTransformer embedding and similarity workflows for text or multimodal inputs, including construction, encode/encode_query/encode_document options, prompt and truncation decisions, precision, normalization, batching/devices, local cache cautions, and output validation.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- embeddings-and-similarity
- license
- Apache 2.0
- metadata
- { "disco-role": "operating" }
- description
- Dense SentenceTransformer embedding and similarity workflows for text or multimodal inputs, including construction, encode/encode_query/encode_document options, prompt and truncation decisions, precision, normalization, batching/devices, local cache cautions, and output validation.
- disable-model-invocation
- true
02 资源文件 4
- 参考文档references/api-reference.md5.5 KB
- 参考文档references/troubleshooting.md4.2 KB
- 参考文档references/workflows.md3.9 KB
- 脚本scripts/dense_embedding_smoke.py4.9 KB
03 SKILL.md 正文
Embeddings and Similarity
Use this sub-skill when the user needs dense SentenceTransformer embeddings or direct embedding similarity scores.
Route Here
- Construct or load a
SentenceTransformerfor embedding inference. - Encode text, images, audio, video, multimodal dicts, or chat-style message inputs with dense embedding models.
- Choose between
encode,encode_query, andencode_document, including prompt-aware retrieval embedding. - Tune
batch_size,device,precision,truncate_dim,normalize_embeddings, output conversion, and similarity metric choices. - Validate embedding shapes, dtypes, normalization, and similarity matrix or pairwise similarity results.
Route Elsewhere
- Use
../retrieval-and-utilities/forutil.semantic_search, hard-negative mining, paraphrase mining, quantized retrieval indexes, or vector database workflows. - Use
../evaluation-and-training/for fine-tuning, losses, trainers, evaluators, and dataset formatting. - Use
../backend-export-optimization/for ONNX/OpenVINO backend selection, export, optimization, or deployment packaging. - Use
../reranking-cross-encoder/forCrossEncoder.predict,CrossEncoder.rank, or reranker scoring. - Use
../sparse-encoder-search/forSparseEncoder, SPLADE-style sparse vectors, sparse search, or sparsity diagnostics.
Fast Workflow
- Confirm the task is dense embedding inference, not retrieval orchestration or model training.
- Load a model with explicit network/cache policy, e.g.
SentenceTransformer(model_id_or_path, local_files_only=True)for offline use. - Prefer
encode_queryandencode_documentfor retrieval models with query/document prompts; useencodefor symmetric STS, clustering, or classification features. - Set output options deliberately:
convert_to_numpy=Truefor NumPy workflows,convert_to_tensor=Truefor Torch workflows, and avoid both being ambiguous in reusable code. - Validate the returned row count, embedding dimension, dtype, finite values, and similarity output shape before using scores downstream.
References
- API details:
references/api-reference.md - Workflow recipes:
references/workflows.md - Failure diagnosis:
references/troubleshooting.md - Safe smoke helper:
scripts/dense_embedding_smoke.py
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).