search-reranker-broad-v1

search-reranker-broad-v1

Published fp32 ONNX broad bilingual reranker for Irish government search.

This repo is the non-quantized sibling of temsa/search-reranker-broad-v1-qint8. It is the broad final-stage reranker line: stronger than the smaller fast K=10 lines on larger candidate pools, but much slower on CPU.

Recommended use

• Use this repo when you want the public fp32 ONNX path for the broad reranker family.
• For CPU deployment, the q8 sibling repo is usually the better serving choice.
• For strict K=10 latency targets, prefer the l5-k10 or l6-fast public lines instead of this broad model.

Trustworthy local deployment data

Artifacts

• Non-quantized ONNX artifact: model.onnx
• Tokenizer files and sentencepiece.bpe.model
• Eval metadata under eval/

Portfolio comparison

<!-- portfolio-comparison:start -->

Updated 2026-03-20 from local reranker reports only.

Use this section as the side-by-side public temsa reranker view. Cells are intentionally left out or summarized when the local report set does not contain a trustworthy non-quantized benchmark for the same public path.

General bilingual rerankers

Policy rerankers

Dynamic q8 CPU at K=20

These rows use the current local K=20 benchmark harness:

• proxy_k20: policy=none
• broad_policy_k20: policy=gov_broad_v1
• office_holder_k20: policy=office_holder_v1
• fixed threads=10, batch_size=32

Intentional gaps:

• search-reranker-broad-v1: the local reports include strong fp32 ONNX proxy and office data, but not a matching fp32 in-domain finephrase / hard-K10 / holdout-A03 set, so those are not claimed here.
• search-reranker-irishgov-l5-k10-v2: the local non-quantized reports are trustworthy for office / finephrase / hard-K10, but not for the same proxy runtime shape as the shipped q8 path.
• search-reranker-irishgov-l6-fast-v1: the local non-quantized reports cover proxy and office, but not the fresh holdout-A03 slice used for the q8 K=10 comparison.

<!-- portfolio-comparison:end -->