Model reference · open weights
reformer-crime-and-punishment is an open-weight language model from Google. reformer-crime-and-punishment (BF16) weighs 11 MB; the smallest configuration that runs it is RTX 3060 12 GB.
For self-hosting, compare this build's requirements below with the GPU you plan to rent.
Summary of the google/reformer-crime-and-punishment model card, 2026-10-01
What it is
| Released by | |
|---|---|
| Released | 2022-03-02 |
| VRAM | 11 MB for the weights |
What it runs on
| Card | Requests at once | Context max | Memory | |
|---|---|---|---|---|
| 8K each | 32K each | |||
| RTX 3060 12 GB | 221 | 55 | all 512K | 11.6 GB |
| RTX 4060 Ti 16 GB | 296 | 74 | all 512K | 15.4 GB |
| RTX 3090 24 GB | 455 | 113 | all 512K | 23.4 GB |
| RTX 4090 24 GB | 454 | 113 | all 512K | 23.4 GB |
| RTX 5090 32 GB | 606 | 151 | all 512K | 31.0 GB |
| L40S 48 GB | 864 | 216 | all 512K | 44.0 GB |
| A100 80 GB | 1000+ | 386 | all 512K | 78.2 GB |
| H100 80 GB | 1000+ | 374 | all 512K | 78.1 GB |
| RTX PRO 6000 Blackwell 96 GB | 1000+ | 452 | all 512K | 93.8 GB |
| DGX Spark (GB10) 128 GB unified | 1000+ | 519 | all 512K | 107 GB |
| H200 141 GB | 1000+ | 672 | all 512K | 138 GB |
| B200 180 GB | 1000+ | 861 | all 512K | 176 GB |
| Requests at once | 8K tokens each | 32K tokens each |
|---|---|---|
| 1 | 521 MB | 672 MB |
| 5 | 722 MB | 1.5 GB |
| 8 | 873 MB | 2.1 GB |
| 16 | 1.3 GB | 3.7 GB |
| 32 | 2.1 GB | 6.9 GB |
| 64 | 3.7 GB | 13.4 GB |
One card, with vLLM's small-card settings.
From the model card
Crime and Punishment is a novel written by Fyodor Dostoevsky and was translated into English.
Crime and Punishment training data was taken from gs://trax-ml/reformer/crime-and-punishment-2554.txt and contains
roughly 0.5M tokens.
The model is a language model that operates on small sub-word units. Text can be generated as follows:
model = ReformerModelWithLMHead.from_pretrained("google/reformer-crime-and-punishment")
tok = ReformerTokenizer.from_pretrained("google/reformer-crime-and-punishment")
tok.decode(model.generate(tok.encode("A few months later", return_tensors="pt"), do_sample=True,temperature=0.7, max_length=100)[0])
# gives:'A few months later on was more than anything in the flat.
# “I have already.” “That’s not my notion that he had forgotten him.
# What does that matter? And why do you mean? It’s only another fellow,” he said as he went out, as though he want'
Quoted from the model card on Hugging Face. The full card is behind the Hugging Face link above.