MLRetrieval.load() Failure

[abhishekreddyallu]

MLRetrieval.load() ignores user supplied SentenceTransformer path

MLRetrieval.load(path=...) ignored the supplied path argument and passed self.path to SentenceTransformer instead.

Location:ontoaligner/aligner/retrieval/retrieval.py:268

Current code:

self.model = SentenceTransformer(self.path, device=self.kwargs["device"])

Since self.path is not assigned from the path argument, the model loader can receive an empty or stale path instead of the requested model.

SentenceTransformer received path: ''
SentenceTransformer received device: 'cpu'
AssertionError: BUG CONFIRMED: load(path='requested-model') ignored the path.

Fix

Use the supplied path argument directly like :

self.model = SentenceTransformer(path, device=self.kwargs["device"])

This matches the existing BiEncoderRetrieval.load() style:

self.model = SentenceTransformer(path, device=self.kwargs["device"], trust_remote_code=True)

After fix

PASS: MLRetrieval.load() forwarded the supplied path correctly.