Crate codemem_embeddings 

codemem-embeddings: Pluggable embedding providers for Codemem.

Supports multiple backends:

• Candle (default): Local BERT models via pure Rust ML (any HF BERT model)
• Ollama: Local Ollama server with any embedding model
• OpenAI: OpenAI API or any compatible endpoint (Together, Azure, etc.)

Modules

ollama

Ollama embedding provider for Codemem.

openai

OpenAI-compatible embedding provider for Codemem.

Structs

CachedProvider

Wraps any EmbeddingProvider with an LRU cache.

EmbeddingService

Embedding service with Candle inference (no internal cache — use CachedProvider wrapper).

Constants

CACHE_CAPACITY

Default LRU cache capacity.

DEFAULT_BATCH_SIZE

Default batch size for batched embedding forward passes. Configurable via EmbeddingConfig.batch_size or CODEMEM_EMBED_BATCH_SIZE.

DEFAULT_HF_REPO

Default HuggingFace model repo ID. Used internally and by commands_init for the default model download.

DEFAULT_REMOTE_DIMENSIONS

Default embedding dimensions for remote providers (Ollama/OpenAI). Candle reads hidden_size from the model’s config.json instead.

MODEL_NAME

Default model name (short form used for directory naming).

Traits

EmbeddingProvider

Trait for pluggable embedding providers.

Functions

from_env

Create an embedding provider from environment variables.

parse_dtype

Parse a dtype string into a Candle DType.