feat: add LocalEmbedder with GPU support, configurable batch size, and CodeRankEmbed opt-in#22

Open

mareurs wants to merge 23 commits intococoindex-io:mainfrom

mareurs:local-embedder-coderank

Commits on Feb 24, 2026

feat: add device auto-detection and trust_remote_code to Config

mareurs
and
claude
committed
fix: restore .cocoindex_code discovery priority and clean up test isolation

mareurs
and
claude
committed
feat: add LocalEmbedder with device and trust_remote_code support

mareurs
and
claude
committed
fix: clean up embedder imports and add memo key test coverage

mareurs
and
claude
committed
feat: use LocalEmbedder in shared.py with GPU device logging

mareurs
and
claude
committed
feat: increase chunk sizes to use jina-embeddings 8192-token context
mareurs
committed

Commits on Feb 25, 2026

fix: use module-level constant for patch target to satisfy ruff E501/N806
mareurs
committed
fix: remove jinaai trust_remote_code auto-detection (breaks transformers 5.x)
mareurs
committed
feat: switch to nomic-ai/CodeRankEmbed for code-specific embeddings

mareurs
and
claude
committed
fix: resolve pre-commit issues (mypy, ruff UP038)

mareurs
and
claude
committed
fix: log effective trust_remote_code and add embed_query test coverage

mareurs
and
claude
committed
style: add blank lines after module docstrings (ruff-format)

mareurs
and
claude
committed
docs: add design doc for configurable batch size and backwards-compatible defaults

mareurs
and
claude
committed
docs: add implementation plan for configurable batch size

mareurs
and
claude
committed
feat: add batch_size to Config and revert default model to all-MiniLM-L6-v2

mareurs
and
claude
committed
refactor: import config singleton from config.py instead of creating in shared.py

mareurs
and
claude
committed
feat: use config.batch_size in LocalEmbedder decorators (COCOINDEX_CODE_BATCH_SIZE)

mareurs
and
claude
committed
fix: import config directly from .config to satisfy mypy no-implicit-reexport

mareurs
and
claude
committed
style: move config import before TYPE_CHECKING block in embedder.py

mareurs
and
claude
committed
docs: add COCOINDEX_CODE_BATCH_SIZE to config table and CodeRankEmbed as GPU opt-in

mareurs
and
claude
committed
fix: validate COCOINDEX_CODE_BATCH_SIZE is a positive integer

mareurs
and
claude
committed
chore: remove internal planning docs from PR

mareurs
and
claude
committed
style: sort imports (ruff)

mareurs
and
claude
committed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

feat: add LocalEmbedder with GPU support, configurable batch size, and CodeRankEmbed opt-in#22

feat: add LocalEmbedder with GPU support, configurable batch size, and CodeRankEmbed opt-in#22
mareurs wants to merge 23 commits intococoindex-io:mainfrom
mareurs:local-embedder-coderank

Commits on Feb 24, 2026

feat: add device auto-detection and trust_remote_code to Config

fix: restore .cocoindex_code discovery priority and clean up test isolation

feat: add LocalEmbedder with device and trust_remote_code support

fix: clean up embedder imports and add memo key test coverage

feat: use LocalEmbedder in shared.py with GPU device logging

feat: increase chunk sizes to use jina-embeddings 8192-token context

Commits on Feb 25, 2026

fix: use module-level constant for patch target to satisfy ruff E501/N806

fix: remove jinaai trust_remote_code auto-detection (breaks transformers 5.x)

feat: switch to nomic-ai/CodeRankEmbed for code-specific embeddings

fix: resolve pre-commit issues (mypy, ruff UP038)

fix: log effective trust_remote_code and add embed_query test coverage

style: add blank lines after module docstrings (ruff-format)

docs: add design doc for configurable batch size and backwards-compatible defaults

docs: add implementation plan for configurable batch size

feat: add batch_size to Config and revert default model to all-MiniLM-L6-v2

refactor: import config singleton from config.py instead of creating in shared.py

feat: use config.batch_size in LocalEmbedder decorators (COCOINDEX_CODE_BATCH_SIZE)

fix: import config directly from .config to satisfy mypy no-implicit-reexport

style: move config import before TYPE_CHECKING block in embedder.py

docs: add COCOINDEX_CODE_BATCH_SIZE to config table and CodeRankEmbed as GPU opt-in

fix: validate COCOINDEX_CODE_BATCH_SIZE is a positive integer

chore: remove internal planning docs from PR

style: sort imports (ruff)