Skip to content

debt(embedding): infra scripts and the NPU worker still hard-code nomic-embed-text outside the literal guard's roots #18382

Open

Description

@mrveiss

Parent: #18143. Follow-up to #18328.

Problem

#18328 widened the embedding-model literal guard (repo_tests/embedding_model_literal_guard_18143_test.py) to namespaced ids, the e5/gte/jina families and the two infra collection builders (autobot-infrastructure/shared/scripts/populate_kb_chromadb.py and autobot-infrastructure/shared/tools/index_documentation.py, now on active_embedding_model()). These infrastructure sites still hard-code nomic-embed-text outside the guard's roots:

File Kind
autobot-infrastructure/shared/scripts/utilities/migrate_redis_to_chromadb.py:139 builds a collection
autobot-infrastructure/shared/scripts/verify_knowledge_consistency.py:196, :243, :312, :336 queries collections
autobot-infrastructure/shared/scripts/utilities/npu_worker.py NPU worker script
autobot-infrastructure/shared/scripts/analysis/npu_performance_measurement.py, redis_final_analysis.py, redis_vector_analysis.py analysis scripts
autobot-infrastructure/shared/scripts/ai-ml/optimize_llm_models.py model tooling
autobot-npu-worker/ (the Windows app, about 8 files) the worker's own default

The first two matter most. A builder or verifier on a hard-coded model disagrees with the index as soon as the configured embedding model changes (#18143). Locations are from an agent sweep and must be re-verified before acting.

Acceptance criteria

Metadata

Metadata

Assignees

No one assigned

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions