Code Search Model Export Workflow

Use this workflow to download and prepare Hugging Face embedding models for MemorySmith into Data/Models with a repo-local Python virtual environment.

One-command install and export

From the repo root:

./Scripts/Install-CodeSearchModel.ps1

Default behavior: - Creates or reuses .venv-model-export - Installs Scripts/model-tools/requirements-model-export.txt - Downloads nomic-ai/nomic-embed-code - Exports or copies ONNX artifacts into Data/Models - Copies tokenizer assets and writes a manifest json

Common examples

Use a model that already publishes ONNX files:

./Scripts/Install-CodeSearchModel.ps1 -ModelId nomic-ai/nomic-embed-text-v1.5 -OutputName nomic-embed-text-v1.5.onnx

Force a clean environment and redownload:

./Scripts/Install-CodeSearchModel.ps1 -RecreateVenv -ForceRedownload

Dry run:

./Scripts/Install-CodeSearchModel.ps1 -WhatIf

Notes on runtime compatibility

• The export workflow can convert models to ONNX, but runtime compatibility still depends on MemorySmith tokenizer support.
• Current defaults are WordPiece-oriented (vocab.txt expected).
• If a model snapshot does not include vocab.txt (for example some Nomic code-model snapshots), the generated manifest includes a compatibility note.

Output files

Artifacts are written under Data/Models, including: - <model-slug>.onnx - <model-slug>.manifest.json - <model-slug>-tokenizer/ tokenizer/config assets

Related tools

• Scripts/Install-CodeSearchModel.ps1
• Scripts/model-tools/export_hf_embedding_model.py
• Scripts/model-tools/requirements-model-export.txt