---

name: model-data-api description: "Inspect SPLADE model, data, index, and inference APIs plus validate small SPLADE-style data layouts without running full training." disable-model-invocation: true metadata: disco-role: operating license: NOASSERTION

SPLADE Model/Data API

Use this sub-skill when a future agent needs to reason about SPLADE Python APIs, data schemas, index objects, or small data validation without launching full training, indexing, retrieval, export, pruning, or evaluation jobs.

Start Here

• For model classes, get_model, bag-of-words helpers, HF Trainer model wrappers, DPR, and inverted index objects, read API reference.
• For raw.tsv, qrel JSON, score JSON, hard-negative pkl-gz, TREC run, and row-id/content-id expectations, read data formats.
• For safe API inspection and notebook-style bag-of-expanded-words inference concepts, read inference and API.
• For schema mismatches, row/content id bugs, special-token cleaning, offline model issues, and empty vectors, read troubleshooting.

Bundled Scripts

• Validate a SPLADE-style fixture root without importing SPLADE:

  python sub-skills/model-data-api/scripts/validate_splade_toy_data.py /path/to/dataset-root
  

• Inspect installed SPLADE imports and signatures without downloading model weights:

  python sub-skills/model-data-api/scripts/inspect_splade_api.py --json
  

Routing Boundaries

• Use this sub-skill for API signatures, object behavior, data schemas, toy validation, small smoke checks, and inference snippets that do not download weights by default.
• Route classic Hydra train/index/retrieve/evaluate module execution to ../hydra-pipelines/SKILL.md.
• Route HuggingFace Trainer training and reranking command construction to ../hf-training-reranking/SKILL.md.
• Route Anserini export, pruning, PISA, BEIR, and evaluation caveats to ../pruning-export-evaluation/SKILL.md.