fix: support quantization ranges for int8/uint8 sentence-transformers…

[DebadityaHait]

Related Issues

• fixes SentenceTransformersTextEmbedder zero embeddings for single text queries when using precision="int8" haystack#9100

Proposed Changes:

With precision="int8" or "uint8", sentence-transformers calibrates the scalar-quantization min/max range from the batch being encoded. For a single text (as in SentenceTransformersTextEmbedder.run) the range is degenerate (min == max), producing meaningless all-equal embeddings (e.g. all zeros).

• SentenceTransformersTextEmbedder and SentenceTransformersDocumentEmbedder accept a new optional quantization_ranges init parameter — explicit calibration ranges with shape (2, embedding_dim) (mins in first row, maxs in second).
• _SentenceTransformersEmbeddingBackend.embed encodes in float32 and forwards the ranges to sentence_transformers.quantize_embeddings when a quantized precision is used with explicit ranges; behavior is unchanged when quantization_ranges is None.
• A warning is logged when int8/uint8 is used without ranges, pointing users at the parameter.
• quantization_ranges is included in to_dict/from_dict serialization.

How did you test it?

• New unit tests: init/serialization, kwarg forwarding from both embedders, backend quantization path (with and without ranges), warning emission.
• New integration test: real tiny model (sentence-transformers-testing/stsb-bert-tiny-safetensors) with int8 + ranges produces a non-degenerate integer embedding.
• hatch run test:unit (162 passed), hatch run test:integration -k quantization (2 passed), hatch run test:types and hatch run fmt clean.

Notes for the reviewer

• This ports the fix originally prepared against haystack core (fix: support quantization ranges for int8 single-text embeddings haystack#11854); maintainers there indicated the sentence-transformers components now live in this repo.
• Ranges are converted with np.asarray before being passed to quantize_embeddings.

Checklist

• I have read the contributors guidelines and the code of conduct
• I have updated the related issue with new insights and changes
• I added unit tests and updated the docstrings
• I've used one of the conventional commit types for my PR title: fix:

[CLAassistant]

All committers have signed the CLA.

[github-actions]

Heads-up for maintainers

This PR is from a fork and touches integrations whose integration tests require API keys.
Those tests are skipped in CI because fork PRs don't have access to repo secrets for security reasons.

Affected integrations:

• sentence_transformers

Please run the integration tests locally (hatch run test:integration inside each folder) before approving.