# esm2_t33_650M_UR50D by AI at Meta: Open-Weight Model Source: https://savrn.com/models/esm2-t33-650m-ur50d Markdown alternate of the page above; the site index is https://savrn.com/llms.txt --- ## Runs On What it takes to serve esm2_t33_650M_UR50D (652M parameters): the memory its weights need at each precision, and the cheapest way to rent enough data-center GPUs to hold them. | Precision | Weights | Memory needed | Cheapest setup | Per hour | Also fits | | --- | --- | --- | --- | --- | --- | | 16-bit | 1.3 GB | 1.6 GB | 1x [MI300X](https://savrn.com/ai-index/pricing/gpus/mi300x) (192 GB) Vultr | $1.85 | [1x H100](https://savrn.com/ai-index/pricing/gpus/h100) $1.99 · [1x MI325X](https://savrn.com/ai-index/pricing/gpus/mi325x) $2.00 | | 8-bit | 0.7 GB | 0.8 GB | 1x [MI300X](https://savrn.com/ai-index/pricing/gpus/mi300x) (192 GB) Vultr | $1.85 | [1x H100](https://savrn.com/ai-index/pricing/gpus/h100) $1.99 · [1x MI325X](https://savrn.com/ai-index/pricing/gpus/mi325x) $2.00 | | 4-bit | 0.3 GB | 0.4 GB | 1x [MI300X](https://savrn.com/ai-index/pricing/gpus/mi300x) (192 GB) Vultr | $1.85 | [1x H100](https://savrn.com/ai-index/pricing/gpus/h100) $1.99 · [1x MI325X](https://savrn.com/ai-index/pricing/gpus/mi325x) $2.00 | Memory is the weights at that precision plus 20% for the runtime and a short context; a long context needs more. Prices are the lowest on-demand hourly rates in the [SAVRN Index](https://savrn.com/ai-index/pricing/gpus), read Oct 24, 2026. [esm2_t33_650M_UR50D on every accelerator the SAVRN Index prices, at every precision](https://savrn.com/models/esm2-t33-650m-ur50d/gpus) ## SAVRN's Notes on esm2_t33_650M_UR50D A protein language model from AI at Meta, built for teams fine-tuning on sequence data: 652M parameters, a 1,026-token window, and a run cost of 1.6 GB at 16-bit, 0.8 GB at 8-bit, 0.4 GB at 4-bit — under one percent of the cheapest Index card at $1.85 an hour, so pack copies or fine-tuning jobs per GPU rather than dedicating one. The odd number is storage: 7.8 GB across nine files, because the repo ships float32 weights in safetensors, pytorch and tf formats. Pull one. MIT permits commercial use, modification and redistribution with the notice kept, so a fine-tuned variant can ship to customers. The 1,026-token cap is a real constraint on long proteins — plan chunking or a larger ESM2 rung before committing. Files date to March 2023 and have not moved since; no host prices it per token, and at this size none should. SAVRN Research ## Model Card By AI at Meta, published under mit, revision 08e4846e5371. ESM-2 is a state-of-the-art protein model trained on a masked language modelling objective. It is suitable for fine-tuning on a wide range of tasks that take protein sequences as input. For detailed information on the model architecture and training data, please refer to the accompanying paper. You may also be interested in some demo notebooks (PyTorch, TensorFlow) which demonstrate how to fine-tune ESM-2 models on your tasks of interest. Several ESM-2 checkpoints are available in the Hub with varying sizes. Larger sizes generally have somewhat better accuracy, but require much more memory and time to train Read AI at Meta's full model card ### ESM-2 ESM-2 is a state-of-the-art protein model trained on a masked language modelling objective. It is suitable for fine-tuning on a wide range of tasks that take protein sequences as input. For detailed information on the model architecture and training data, please refer to the [accompanying paper](https://www.biorxiv.org/content/10.1101/2022.07.20.500902v2). You may also be interested in some demo notebooks ([PyTorch](https://colab.research.google.com/github/huggingface/notebooks/blob/main/examples/protein_language_modeling.ipynb), [TensorFlow](https://colab.research.google.com/github/huggingface/notebooks/blob/main/examples/protein_language_modeling-tf.ipynb)) which demonstrate how to fine-tune ESM-2 models on your tasks of interest. Several ESM-2 checkpoints are available in the Hub with varying sizes. Larger sizes generally have somewhat better accuracy, but require much more memory and time to train: | Checkpoint name | Num layers | Num parameters | | --- | --- | --- | | [esm2_t48_15B_UR50D](https://huggingface.co/facebook/esm2_t48_15B_UR50D) | 48 | 15B | | [esm2_t36_3B_UR50D](https://savrn.com/models/esm2-t36-3b-ur50d) | 36 | 3B | | [esm2_t33_650M_UR50D](https://savrn.com/models/esm2-t33-650m-ur50d) | 33 | 650M | | [esm2_t30_150M_UR50D](https://huggingface.co/facebook/esm2_t30_150M_UR50D) | 30 | 150M | | [esm2_t12_35M_UR50D](https://huggingface.co/facebook/esm2_t12_35M_UR50D) | 12 | 35M | | [esm2_t6_8M_UR50D](https://savrn.com/models/esm2-t6-8m-ur50d) | 6 | 8M | ## Configuration Architecture EsmForMaskedLM Context length (tokens) 1,026 Layers 33 Hidden size 1,280 Feed-forward size 5,120 Attention heads 20 Vocabulary size 33 Stored precision float32 Model type esm ## Identity and Version Repository facebook/esm2_t33_650M_UR50D Publisher AI at Meta Task Fill mask Modality Text Library transformers Parameters 652M parameters Languages tf, esm Revision 08e4846e537177426273712802403f7ba8261b6c First published 2022-09-27 Last updated 2023-03-21 ## Files and Weights 9 files, 7.8 GB in total. The weights are 3 files totalling 7.8 GB in bin, h5, safetensors. Weights3 files · 7.8 GB Configuration2 files · 849 B Tokenizer2 files · 188 B Documentation1 file · 1.7 KB Repository1 file · 1.4 KB Every file | File | Type | Size | SHA-256 | | --- | --- | --- | --- | | model.safetensors | Weights | 2.6 GB | a08adabb949f | | pytorch_model.bin | Weights | 2.6 GB | c874668852c7 | | tf_model.h5 | Weights | 2.6 GB | 3110b0ee07a4 | | config.json | Configuration | 724 B | — | | special_tokens_map.json | Configuration | 125 B | — | | README.md | Documentation | 1.7 KB | — | | .gitattributes | Repository | 1.4 KB | — | | tokenizer_config.json | Tokenizer | 95 B | — | | vocab.txt | Tokenizer | 93 B | — | ## License and Download License mit Access Open weights, no gate Download size 7.8 GB [Download from AI at Meta](https://huggingface.co/facebook/esm2_t33_650M_UR50D) Released by AI at Meta through its official repository on Hugging Face. [Read the license](https://opensource.org/license/mit). ## Memory Requirements | Precision | Weights in memory | | --- | --- | | As published | 7.8 GB | | 16-bit | 1.3 GB | | 8-bit | 0.7 GB | | 4-bit | 0.3 GB | Weights only, from the published parameter count; the key-value cache and runtime add to this. ## Compare esm2_t33_650M_UR50D - [esm2_t33_650M_UR50D vs xlm-roberta-large](https://savrn.com/models/compare/esm2-t33-650m-ur50d-vs-xlm-roberta-large) ## Questions About esm2_t33_650M_UR50D ### How much GPU memory does esm2_t33_650M_UR50D need? About 1.6 GB at 16-bit and 0.4 GB at 4-bit: the weights (652M parameters) plus a working margin. A long context needs more. ### What is the cheapest GPU to run esm2_t33_650M_UR50D on? At 16-bit, 1x MI300X from $1.85 an hour; at 4-bit, 1x MI300X from $1.85 an hour, at the lowest on-demand prices the SAVRN Index lists. ### Can I use esm2_t33_650M_UR50D commercially? Yes. esm2_t33_650M_UR50D 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. ### What is esm2_t33_650M_UR50D's context length? 1,026 tokens, from the maximum position embeddings in its published configuration. ## Similar Models Model · Fill mask ### [minerva-mlm](https://savrn.com/models/minerva-mlm) [Garyk Brixi](https://savrn.com/model-publishers/gbrixi) Minerva-MLM is a genome language model for coevolutionary mining. The checkpoint includes the masked language model head and three interaction heads for base-pairing, repeat, and structure signals. Full documentation, finetuning scripts, notebooks, and setup are on the github. Try it in the browser here. Inference from this repo needs only Transformers and PyTorch: flash-attn is picked up automatically when present, otherwise Minerva falls back to PyTorch SDPA. The optional minerva-dna package adds plotting helpers, GenBank utilities and finetuning wrappers, and lets you load the model without trustremotecode: With minerva-dna installed, import the class directly and drop trustremotecode… Open weights apache-2.0 671M parameters transformers [View model](https://savrn.com/models/minerva-mlm) Model · Fill mask ### [minerva-mlm-8k](https://savrn.com/models/minerva-mlm-8k) [Garyk Brixi](https://savrn.com/model-publishers/gbrixi) Minerva-MLM is a genome language model for coevolutionary mining. The checkpoint includes the masked language model head and three interaction heads for base-pairing, repeat, and structure signals. Full documentation, finetuning scripts, notebooks, and setup are on the github. Try it in the browser here. Inference from this repo needs only Transformers and PyTorch: flash-attn is picked up automatically when present, otherwise Minerva falls back to PyTorch SDPA. The optional minerva-dna package adds plotting helpers, GenBank utilities and finetuning wrappers, and lets you load the model without trustremotecode: With minerva-dna installed, import the class directly and drop trustremotecode… Open weights apache-2.0 671M parameters transformers [View model](https://savrn.com/models/minerva-mlm-8k) Model · Fill mask ### [xlm-roberta-large](https://savrn.com/models/xlm-roberta-large) [Facebook AI community](https://savrn.com/model-publishers/facebookai) XLM-RoBERTa model pre-trained on 2.5TB of filtered CommonCrawl data containing 100 languages. It was introduced in the paper Unsupervised Cross-lingual Representation Learning at Scale by Conneau et al. and first released in this repository. Disclaimer: The team releasing XLM-RoBERTa did not write a model card for this model so this model card has been written by the Hugging Face team. XLM-RoBERTa is a multilingual version of RoBERTa. It is pre-trained on 2.5TB of filtered CommonCrawl data containing 100 languages. RoBERTa is a transformers model pretrained on a large corpus in a self-supervised fashion. This means it was pretrained on the raw texts only, with no humans labelling them in… Open weights mit 561M parameters 514 tokens transformers [View model](https://savrn.com/models/xlm-roberta-large) Model · Fill mask ### [ModernBERT-large](https://savrn.com/models/modernbert-large) [Answer.AI](https://savrn.com/model-publishers/answerdotai) ModernBERT is a modernized bidirectional encoder-only Transformer model (BERT-style) pre-trained on 2 trillion tokens of English and code data with a native context length of up to 8,192 tokens. ModernBERT leverages recent architectural improvements such as: - Rotary Positional Embeddings (RoPE) for long-context support. - Local-Global Alternating Attention for efficiency on long inputs. - Unpadding and Flash Attention for efficient inference. ModernBERT’s native long context length makes it ideal for tasks that require processing long documents, such as retrieval, classification, and semantic search within large corpora. The model was trained on a large corpus of text and code, making it… Open weights apache-2.0 396M parameters 8,192 tokens transformers [View model](https://savrn.com/models/modernbert-large) Model · Fill mask ### [ModernBERT-Large-Instruct-Logician-v0.5](https://savrn.com/models/modernbert-large-instruct-logician-v0-5) [John Gorriceta](https://savrn.com/model-publishers/johngorri) This model is a fine-tuned version of ModernBERT optimized for logical reasoning, deductive analysis, and structure-based token prediction. It relies on the ModernBertForMaskedLM architecture, making it highly effective at filling in missing contextual logic tokens (fill-mask). - As an encoder-based Masked Language Model, it is not designed for long-form generative text (like ChatGPT). It excels at predicting masked tokens inside structured prompts. Open weights apache-2.0 396M parameters 8,192 tokens [View model](https://savrn.com/models/modernbert-large-instruct-logician-v0-5) Model · Fill mask ### [roberta-large](https://savrn.com/models/roberta-large) [Facebook AI community](https://savrn.com/model-publishers/facebookai) Pretrained model on English language using a masked language modeling (MLM) objective. It was introduced in this paper and first released in this repository. This model is case-sensitive: it makes a difference between english and English. Disclaimer: The team releasing RoBERTa did not write a model card for this model so this model card has been written by the Hugging Face team. RoBERTa is a transformers model pretrained on a large corpus of English data in a self-supervised fashion. This means it was pretrained on the raw texts only, with no humans labelling them in any way (which is why it can use lots of publicly available data) with an automatic process to generate inputs and labels… Open weights mit 355M parameters 514 tokens transformers [View model](https://savrn.com/models/roberta-large) ## AI at Meta [All models and datasets](https://savrn.com/model-publishers/facebook) ## Versions - [08e4846e5371](https://savrn.com/models/esm2-t33-650m-ur50d/versions/08e4846e5371) · current 2026-10-24 ## Explore More - [All fill mask models](https://savrn.com/models/tasks/fill-mask) - [All models under mit](https://savrn.com/models/licenses/mit) - [Model comparisons](https://savrn.com/models/comparisons) - [The model directory](https://savrn.com/models) - [Open model prices by host](https://savrn.com/ai-index/pricing/open-models) ## Source - Repository metadata, read 2026-10-19. - [AI at Meta website](https://ai.meta.com/resources/models-and-libraries/) - [Hugging Face record](https://huggingface.co/facebook/esm2_t33_650M_UR50D) - [How the hub is built](https://savrn.com/model-hub/methodology)