SAVRN Model Hub · Comparisons
distilbert-base-multilingual-cased vs roberta-base
Distilbert-base-multilingual-cased has 135M parameters and roberta-base has 125M parameters; distilbert-base-multilingual-cased is released under Apache License 2.0 and roberta-base under MIT License; at 16-bit, distilbert-base-multilingual-cased needs about 0.3 GB (1x MI300X from $1.85 an hour) and roberta-base about 0.3 GB (1x MI300X from $1.85 an hour).
| Field | distilbert-base-multilingual-cased distilbert/distilbert-base-multilingual-cased | roberta-base FacebookAI/roberta-base |
|---|---|---|
| Publisher | DistilBERT community | Facebook AI community |
| Task | Fill mask | Fill mask |
| Modality | Text | Text |
| Parameters, as reported | 135M parameters | 125M parameters |
| Architecture | DistilBertForMaskedLM | RobertaForMaskedLM |
| Library | transformers | transformers |
| Context length | 512 tokens | 514 tokens |
| Repository size | 2.9 GB | 2.8 GB |
| Artifact formats | safetensors, onnx, pytorch, tf | safetensors, pytorch, jax, rust, tf |
| License | apache-2.0 | mit |
| Access | Open weights, no gate | Open weights, no gate |
| Memory at 16-bit (weights and margin) | 0.3 GB | 0.3 GB |
| Cheapest GPUs at 16-bit, per hour | 1x MI300X, $1.85 | 1x MI300X, $1.85 |
| Memory at 4-bit (weights and margin) | 0.1 GB | 0.1 GB |
| Cheapest GPUs at 4-bit, per hour | 1x MI300X, $1.85 | 1x MI300X, $1.85 |
| Revision viewed | 45c032ab32cc | e2da8e2f811d |
| Downloads reported by the hub | 541.6k | 7.9M |
| Last observed | 2026-09-18 | 2026-10-16 |
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 distilbert-base-multilingual-cased
This is the distilled version of multilingual BERT: one encoder that covers the Wikipedia languages it was trained on, at 135M parameters and noticeably faster than the original. It is cased, so it treats capitalization as meaningful, which helps with names and acronyms.
It is a base model for fine-tuning, not a finished classifier. Apache 2.0 allows commercial use. For new multilingual work, compare it with XLM-RoBERTa and mDeBERTa-v3 base in our catalog, which usually fine-tune to higher accuracy; keep DistilBERT where speed per request matters most.
SAVRN's Notes on roberta-base
Case matters here: roberta-base, from the Facebook AI community, tells english from English. Pretrained on raw English text from bookcorpus and wikipedia by masked language modeling, its listed task is fill mask: token prediction and what you build on it, not chat. At 16-bit its 125M parameters take 0.2 GB of weights and 0.3 GB to run, against a cheapest listed setup of one 192 GB MI300X at $1.85 per hour on-demand; on our floor that shares a card, never owns one.
MIT is short: commercial use, modification and redistribution are allowed, the one duty being to carry the copyright and permission notices along. Check provenance and packaging. The releasing team wrote no model card, so bookcorpus and wikipedia are all the page discloses on data. The repository is 2.82 GB across 13 files in safetensors, pytorch, jax, rust and tf; pull only the format your stack loads.
Questions
Which is larger, distilbert-base-multilingual-cased or roberta-base?
distilbert-base-multilingual-cased (135M parameters) is larger than roberta-base (125M parameters), by the parameter counts their publishers report.
Which is cheaper to run, distilbert-base-multilingual-cased or roberta-base?
At 4-bit, distilbert-base-multilingual-cased fits on 1x MI300X from $1.85 an hour and roberta-base on 1x MI300X from $1.85 an hour, at the lowest on-demand prices the SAVRN Index lists.
Can I use distilbert-base-multilingual-cased commercially?
Yes. distilbert-base-multilingual-cased 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-base commercially?
Yes. roberta-base 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.