Pass 3 of 3 · Deep dive of PHAROSPass 1 · OverviewPass 2 · How it works
Code Observatory / class
OnnxEmbedder
codeintel.embeddings.OnnxEmbedder
PINNED SOURCE Source and edges from the index at this commit. An execution appears only where one was recorded.
PHAROS at 975a38dMIT license
Around this symbol
Search all symbolsLoading
| Symbol | File | Resolved by |
|---|---|---|
| main | edge_server.py | through an import |
| get_embedder | embeddings.py | same file |
| Symbol | File | Resolved by |
|---|---|---|
| __init__ | embeddings.py | same file |
| _load | embeddings.py | same file |
| embed_documents | embeddings.py | same file |
| embed_query | embeddings.py | same file |
| encode_raw | embeddings.py | same file |
| Symbol | File | Resolved by |
|---|---|---|
| embeddings | embeddings.py | same file |
codeintel.embeddings.OnnxEmbedderbackend/app/codeintel/embeddings.py, lines 286–345Open source
class OnnxEmbedder(Embedder): """A sentence-embedding model exported to ONNX, run on CPU. ``pharos_model.json`` in ``model_dir`` (written by ``scripts/fetch_onnx_model.py``) gives the model's pooling (``mean`` or ``cls``), query / document prefixes, maximum length, ONNX file name and model id; without it the defaults suit all-MiniLM-L6-v2. """ def __init__(self, model_dir: str, model_id: str | None = None, max_length: int | None = None, batch_size: int = 32): self.model_dir = model_dir profile = _read_profile(model_dir) self.model_id = model_id or profile.get("model_id") or f"onnx:{os.path.basename(os.path.normpath(model_dir))}" self.max_length = max_length or int(profile.get("max_length", 256)) self.pooling = profile.get("pooling", "mean") if self.pooling not in ("mean", "cls"): raise ValueError(f"unsupported pooling {self.pooling!r} in {model_dir}/pharos_model.json") self.query_prefix = profile.get("query_prefix", "") self.document_prefix = profile.get("document_prefix", "") self.onnx_file = profile.get("onnx_file", "model.onnx") self.batch_size = batch_size self._session = None self._tokenizer = None self._lock = threading.Lock() def _load(self): if self._session is not None: return with self._lock: if self._session is not None: return try: import onnxruntime as ort from tokenizers import Tokenizer except ImportError as exc: raise EmbeddingUnavailable( "onnx backend needs onnxruntime and tokenizers (pip install -r requirements-onnx.txt)" ) from exc model_path = os.path.join(self.model_dir, self.onnx_file) tok_path = os.path.join(self.model_dir, "tokenizer.json") if not (os.path.isfile(model_path) and os.path.isfile(tok_path)): raise EmbeddingUnavailable(f"{self.model_dir} must contain {self.onnx_file} and tokenizer.json") tokenizer = Tokenizer.from_file(tok_path) tokenizer.enable_truncation(max_length=self.max_length) tokenizer.enable_padding() opts = ort.SessionOptions() opts.intra_op_num_threads = int(os.getenv("EMBEDDING_ONNX_THREADS", "0")) or os.cpu_count() or 1 self._session = ort.InferenceSession(model_path, opts, providers=["CPUExecutionProvider"]) self._input_names = {i.name for i in self._session.get_inputs()} self._tokenizer = tokenizer def embed_documents(self, texts: list[str]) -> list[list[float]]: return self.encode_raw([self.document_prefix + t for t in texts]) def embed_query(self, text: str) -> list[float]: return self.encode_raw([self.query_prefix + text])[0] def encode_raw(self, texts: list[str]) -> list[list[float]]: import numpy as npShows the first 60 of 79 lines. The rest is on GitHub at this commit.
What the index recorded
- Signature
class OnnxEmbedder(Embedder):- File
backend/app/codeintel/embeddings.py- Lines
- 286–364
- Parser
- python-ast
- Git blob
df7f8202f23ca6ad2c57aca5f09b435eb6486d79- Outside this subset
- 1 more edges reach code outside the graph shown here
