SAVRN Model Hub · Comparisons
bge-reranker-v2-m3 vs roberta-large-mnli
Bge-reranker-v2-m3 has 568M parameters and roberta-large-mnli has 356M parameters; bge-reranker-v2-m3 is released under Apache License 2.0 and roberta-large-mnli under MIT License; at 16-bit, bge-reranker-v2-m3 needs about 1.4 GB (1x MI300X from $1.85 an hour) and roberta-large-mnli about 0.9 GB (1x MI300X from $1.85 an hour).
| Field | bge-reranker-v2-m3 BAAI/bge-reranker-v2-m3 | roberta-large-mnli FacebookAI/roberta-large-mnli |
|---|---|---|
| Publisher | Beijing Academy of Artificial Intelligence | Facebook AI community |
| Task | Text classification | Text classification |
| Modality | Text | Text |
| Parameters, as reported | 568M parameters | 356M parameters |
| Architecture | XLMRobertaForSequenceClassification | RobertaForSequenceClassification |
| Library | sentence-transformers | transformers |
| Context length | 8,194 tokens | 514 tokens |
| Repository size | 2.3 GB | 5.7 GB |
| Artifact formats | safetensors | safetensors, pytorch, jax, tf |
| License | apache-2.0 | mit |
| Access | Open weights, no gate | Open weights, no gate |
| Memory at 16-bit (weights and margin) | 1.4 GB | 0.9 GB |
| Cheapest GPUs at 16-bit, per hour | 1x MI300X, $1.85 | 1x MI300X, $1.85 |
| Memory at 4-bit (weights and margin) | 0.3 GB | 0.2 GB |
| Cheapest GPUs at 4-bit, per hour | 1x MI300X, $1.85 | 1x MI300X, $1.85 |
| Revision viewed | 953dc6f6f85a | 2a8f12d27941 |
| Downloads reported by the hub | 17.7M | 238.7k |
| Last observed | 2026-09-18 | 2026-09-18 |
An evaluation row appears only where at least two of these models report the same benchmark with the same stated configuration, metric, unit and setup. Different evaluators stay named in each cell. Values are shown as reported: no unit conversion, no ranking.
SAVRN's Notes on bge-reranker-v2-m3
Feed it a query and a passage and it returns one relevance score, mapped by a sigmoid into the 0 to 1 range. That is the job of this 568M-parameter reranker from the Beijing Academy of Artificial Intelligence: it scores pairs instead of producing embeddings, so it belongs behind whatever retrieval step supplies the candidates. At 16-bit the weights take 1.1 GB and the run needs 1.4 GB, against a cheapest listed setup of one MI300X with 192 GB at $1.85 an hour. Nobody dedicates that card to 1.4 GB; the figure says it rides on whatever accelerator hosts your generator.
Apache 2.0 permits commercial use, modification and redistribution provided the notices and NOTICE file travel with it and significant changes are stated. Check that 8,194 tokens covers query plus passage together, and that no host on the SAVRN Index prices it per token, so you run it yourself.
Questions
Which is larger, bge-reranker-v2-m3 or roberta-large-mnli?
bge-reranker-v2-m3 (568M parameters) is larger than roberta-large-mnli (356M parameters), by the parameter counts their publishers report.
Which is cheaper to run, bge-reranker-v2-m3 or roberta-large-mnli?
At 4-bit, bge-reranker-v2-m3 fits on 1x MI300X from $1.85 an hour and roberta-large-mnli on 1x MI300X from $1.85 an hour, at the lowest on-demand prices the SAVRN Index lists.
Can I use bge-reranker-v2-m3 commercially?
Yes. bge-reranker-v2-m3 is released under Apache License 2.0. The Apache License 2.0 is a permissive open-source license. It permits commercial use, modification and redistribution. It requires keeping the license and copyright notices and any NOTICE file, stating significant changes, and it includes an express patent grant from contributors.
Can I use roberta-large-mnli commercially?
Yes. roberta-large-mnli is released under MIT License. The MIT License is a short permissive license. It permits commercial use, modification and redistribution, provided the copyright notice and permission notice are included.