Pass 3 of 3 · Deep dive of PHAROSPass 1 · OverviewPass 2 · How it works
Code Observatory / Source
Embed the question
ONNX inference on the CPU, with the model's pooling and L2 normalisation.
PINNED SOURCE Copied from the repository at this commit. Source, not an execution.
PHAROS at 975a38dMIT license
OnnxEmbedder.encode_rawbackend/app/codeintel/embeddings.py, lines 341–364Open source
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 np self._load() out: list[list[float]] = [] for i in range(0, len(texts), self.batch_size): batch = self._tokenizer.encode_batch(texts[i : i + self.batch_size]) ids = np.array([e.ids for e in batch], dtype=np.int64) mask = np.array([e.attention_mask for e in batch], dtype=np.int64) feeds = {"input_ids": ids, "attention_mask": mask} if "token_type_ids" in self._input_names: feeds["token_type_ids"] = np.zeros_like(ids) hidden = self._session.run(None, feeds)[0] # (batch, seq, dim) if self.pooling == "cls": pooled = hidden[:, 0, :].astype(np.float32) else: m = mask[..., None].astype(np.float32) pooled = (hidden * m).sum(axis=1) / np.clip(m.sum(axis=1), 1e-9, None) pooled /= np.clip(np.linalg.norm(pooled, axis=1, keepdims=True), 1e-12, None) out.extend(_pad(v.tolist(), self.dim) for v in pooled) return outWhere this comes from
- Commit
975a38d5d17839c26ee219295a5be6ac66b6d910- File
backend/app/codeintel/embeddings.py- Lines
- 341–364
- Git blob
df7f8202f23ca6ad2c57aca5f09b435eb6486d79- License
- MIT, © 2024–2026 Pharos Contributors. Notice
Read it on GitHub at this commit
Check it yourself
From a clone of the repository, the first command prints the blob above and the second prints these lines.
git rev-parse 975a38d:backend/app/codeintel/embeddings.py
git show 975a38d:backend/app/codeintel/embeddings.py | sed -n '341,364p'