SAVRN Model Hub · Comparisons
bart-large-mnli vs deberta-v3-large-zeroshot-v2.0
Bart-large-mnli has 407M parameters and deberta-v3-large-zeroshot-v2.0 has 435M parameters; both are released under MIT License; at 16-bit, bart-large-mnli needs about 1 GB (1x MI300X from $1.85 an hour) and deberta-v3-large-zeroshot-v2.0 about 1 GB (1x MI300X from $1.85 an hour).
| Field | bart-large-mnli facebook/bart-large-mnli | deberta-v3-large-zeroshot-v2.0 MoritzLaurer/deberta-v3-large-zeroshot-v2.0 |
|---|---|---|
| Publisher | AI at Meta | Moritz Borrett-Laurer (formerly Laurer) |
| Task | Zero-shot classification | Zero-shot classification |
| Modality | Text | Text |
| Parameters, as reported | 407M parameters | 435M parameters |
| Architecture | BartForSequenceClassification | DebertaV2ForSequenceClassification |
| Library | transformers | transformers |
| Context length | 1,024 tokens | 512 tokens |
| Repository size | 6.9 GB | 2.6 GB |
| Artifact formats | safetensors, pytorch, jax, rust | safetensors, onnx |
| License | mit | mit |
| Access | Open weights, no gate | Open weights, no gate |
| Memory at 16-bit (weights and margin) | 1 GB | 1 GB |
| Cheapest GPUs at 16-bit, per hour | 1x MI300X, $1.85 | 1x MI300X, $1.85 |
| Memory at 4-bit (weights and margin) | 0.2 GB | 0.3 GB |
| Cheapest GPUs at 4-bit, per hour | 1x MI300X, $1.85 | 1x MI300X, $1.85 |
| Revision viewed | d7645e127eaf | cf44676c28ba |
| Downloads reported by the hub | 3.1M | 165.3k |
| 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 bart-large-mnli
Zero-shot here means the labels arrive at inference time, not training time. The passage becomes the premise, each candidate label becomes a hypothesis, and the entailment score picks the winner, so a ten-label job is ten passes over the same text and throughput, not memory, is what to plan around. The cheapest listed setup is a single MI300X, 192 GB, at $1.85 per hour on-demand; at 16-bit the 407 million parameters need 1.0 GB of it, so share the card.
The license is MIT: keep the copyright and permission notice and use, modify and redistribute it commercially as you like. Two checks. The context is 1,024 tokens, so long documents get chunked. And the lineage: this is bart-large after training on MultiNLI; arXiv 1910.13461 covers BART and arXiv 1909.00161 the zero-shot method. No host on our Index prices it by the token, so you run it yourself.
Questions
Which is larger, bart-large-mnli or deberta-v3-large-zeroshot-v2.0?
deberta-v3-large-zeroshot-v2.0 (435M parameters) is larger than bart-large-mnli (407M parameters), by the parameter counts their publishers report.
Which is cheaper to run, bart-large-mnli or deberta-v3-large-zeroshot-v2.0?
At 4-bit, bart-large-mnli fits on 1x MI300X from $1.85 an hour and deberta-v3-large-zeroshot-v2.0 on 1x MI300X from $1.85 an hour, at the lowest on-demand prices the SAVRN Index lists.
Can I use bart-large-mnli commercially?
Yes. bart-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.
Can I use deberta-v3-large-zeroshot-v2.0 commercially?
Yes. deberta-v3-large-zeroshot-v2.0 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.