Downloads · 30 days
0
TahaRauf/granite-embedding-reranker-english-r2-onnx-fp16
granite-embedding-reranker-english-r2-onnx-fp16 is a text ranking model from TahaRauf. Use it for the text ranking task on the model card, and read the license before you ship it in a product. It is set up for onnx. The card lists the license as apache-2.0.
An fp16 ONNX conversion of ibm-granite/granite-embedding-reranker-english-r2, IBM's 149.6M-parameter ModernBERT cross-encoder reranker, released by IBM under the Apache License 2.0.
Downloads · 30 days
0
Access
Public
Updated Sep 23, 2026
Repo size
300 MB
Likes
0
Public
Click a slice to open those files.
.onnx300 MB · 99%
From the Hugging Face model README
An fp16 ONNX conversion of ibm-granite/granite-embedding-reranker-english-r2, IBM's 149.6M-parameter ModernBERT cross-encoder reranker, released by IBM under the Apache License 2.0.
This repository redistributes a MODIFIED copy of IBM's model. The changes are listed below. It is not affiliated with, or endorsed by, IBM. The original model, its card, its evaluation and its intended-use statement are IBM's and live in the repository linked above.
keep_io_types).No weight was retrained or fine-tuned. tokenizer.json is IBM's file, byte-for-byte unchanged.
| File | Bytes | sha256 |
|---|---|---|
model_fp16.onnx | 299,879,022 | b1a75200046b797987d16c96bb14c8959f4a4ec7c21ceb5e0d00227c84178695 |
tokenizer.json | 3,583,486 | 39bc3f741c62ba10b6f412f2bfaf2ee0e3241bd3801c3a056bb4e0dbdabcf416 |
LICENSE | 11,358 | the Apache License 2.0, as published at apache.org |
Verify a download before use: sha256sum model_fp16.onnx must print the hash above.
| Name | Direction | Type | Shape |
|---|---|---|---|
input_ids | input | int64 | [batch_size, sequence_length] |
attention_mask | input | int64 | [batch_size, sequence_length] |
logits | output | float32 | [batch_size, 1] |
Each row is one (query, passage) pair tokenised together as a cross-encoder input; logits is
the relevance score for that pair (apply a sigmoid for a value in 0–1). Tokenise with the
included tokenizer.json, the same tokenizer IBM ships.
Measured on 2026-09-22 against the fp32 reference over 200 query–passage pairs in 20 groups:
| Measure | Value |
|---|---|
| Maximum absolute logit difference | 0.011 |
| Spearman rank correlation | 0.99994 |
| Top-1 disagreements | 0 of 20 groups |
| Pairwise order inversions | 2 of 900 |
These figures describe agreement with the fp32 graph, not retrieval quality. For the model's own quality figures, see IBM's card.
LICENSE.