Nabu + Nisaba

""" Embedding generation for vector search. Provides multi-model embedding support: - UniXcoder: Code→code similarity (contrastive learning) - CodeBERT: Structural understanding (data flow graphs) Fusion: - Pythagorean³ consensus (P3): Proven UX×CB fusion approach """ from nabu.embeddings.base import ( EmbeddingGenerator, EmbeddingModel, compute_non_linear_consensus ) # Optional dependencies for embeddings (torch, transformers) # Only import if dependencies are available try: from nabu.embeddings.unixcoder_generator import UniXcoderGenerator from nabu.embeddings.codebert_generator import CodeBERTGenerator from nabu.embeddings.fusion_strategies import NonLinearConsensusFusion from nabu.embeddings.generator_cache import ( get_unixcoder_generator, get_codebert_generator, clear_generator_cache ) _EMBEDDINGS_AVAILABLE = True except ImportError: # Embeddings require torch/transformers, which are optional # Provide stub implementations that raise helpful errors UniXcoderGenerator = None # type: ignore CodeBERTGenerator = None # type: ignore NonLinearConsensusFusion = None # type: ignore get_unixcoder_generator = None # type: ignore get_codebert_generator = None # type: ignore clear_generator_cache = lambda: None # type: ignore _EMBEDDINGS_AVAILABLE = False __all__ = [ # Core generators 'EmbeddingGenerator', 'EmbeddingModel', 'UniXcoderGenerator', 'CodeBERTGenerator', # Fusion 'compute_non_linear_consensus', 'NonLinearConsensusFusion', # Cached getters (recommended for MCP tools) 'get_unixcoder_generator', 'get_codebert_generator', 'clear_generator_cache', ]