text-models

Text Models

Use this sub-skill for DeepPavlov model families that operate on texts, tokens, spans, relations, labels, scores, or embeddings.

Route Here

• Text classification, sentence-pair classification, multiple-choice classification, few-shot classification, benchmark classification, and scalar text scoring.
• NER, sentence segmentation, morpho-syntax parsing, and other token-level tagging workflows.
• Entity detection, entity linking, and relation extraction workflows.
• Spelling correction workflows.
• Multitask workflows that share one backbone across several text heads.
• BERT-style and sentence-level embedding workflows, including the standalone embedding configs.

Reroute Elsewhere

• Core config syntax, nested configs, registry behavior, custom components, and train/evaluate plumbing -> pipelines.
• Document retrieval, ranking, FAQ, SQuAD, ODQA, and KBQA workflows -> retrieval-qa.
• REST and socket deployment, ports, payloads, probes, and response framing -> serving.
• Cross-cutting install/import/backend/cache issues -> the root troubleshooting reference once it is present in the generated skill tree.

Fast Workflow

1. Run scripts/list_config_categories.py to confirm the live config inventory in the installed package.
2. Match the task shape to references/model-catalog.md.
3. Use references/data-formats.md for chainer.in / chainer.out shapes and safe batching rules.
4. If the task needs config surgery, reader changes, iterator changes, or custom components, switch to pipelines.
5. If the task is about exposing an already-selected model over HTTP or sockets, switch to serving.

References

• Model catalog
• Data formats
• Troubleshooting
• Config inventory script