Model reference · open weights
cpm-ant is an open-weight language model from openbmb. cpm-ant-10b (BF16) weighs 38.1 GB; the smallest configuration that runs it is H100 80 GB.
For self-hosting, compare this build's requirements below with the GPU you plan to rent.
Summary of the openbmb/cpm-ant-10b model card, 2026-10-01
What it is
| Released by | openbmb |
|---|---|
| Released | 2023-01-15 |
| VRAM | 38.1 GB for the weights |
What it runs on
| Card | Requests at once | Context max | Memory | |
|---|---|---|---|---|
| 8K each | 32K each | |||
| RTX 3060 12 GB … RTX 5090 32 GB 5 smaller cards | ||||
| L40S 48 GB | 6K | 44.0 GB | ||
| A100 80 GB | 6 | 1 | 48K | 78.2 GB |
| H100 80 GB | 5 | 1 | 45K | 78.1 GB |
| RTX PRO 6000 Blackwell 96 GB | 8 | 2 | 64K | 93.8 GB |
| DGX Spark (GB10) 128 GB unified | 10 | 2 | 81K | 107 GB |
| H200 141 GB | 14 | 3 | 119K | 138 GB |
| B200 180 GB | 20 | 5 | 128K | 176 GB |
| 4× RTX 4090 24 GB tensor parallel | 8 | 2 | 65K | 23.4 GB a card |
| 4× RTX 3090 24 GB tensor parallel | 8 | 2 | 65K | 23.4 GB a card |
| 4× RTX 5090 32 GB tensor parallel | 12 | 3 | 103K | 31.0 GB a card |
| 2× H100 80 GB tensor parallel | 17 | 4 | 128K | 78.1 GB a card |
| 2× A100 80 GB tensor parallel | 18 | 4 | 128K | 78.2 GB a card |
| 4× L40S 48 GB tensor parallel | 20 | 5 | 128K | 44.0 GB a card |
| Requests at once | 8K tokens each | 32K tokens each |
|---|---|---|
| 1 | 45.2 GB | 64.5 GB |
| 5 | 70.9 GB | 168 GB |
| 8 | 90.3 GB | 245 GB |
| 16 | 142 GB | 451 GB |
| 32 | 245 GB | 863 GB |
| 64 | 451 GB | 1688 GB |
One card, with vLLM's small-card settings.
From the model card
pip install transformers
from transformers import CpmAntTokenizer, CpmAntForCausalLM
texts = "今天天气不错，"
model = CpmAntForCausalLM.from_pretrained("openbmb/cpm-ant-10b")
tokenizer = CpmAntTokenizer.from_pretrained("openbmb/cpm-ant-10b")
input_ids = tokenizer(texts, return_tensors="pt")
outputs = model.generate(**input_ids)
output_texts = tokenizer.batch_decode(outputs)
print(output_texts)
Quoted from the model card on Hugging Face. The full card is behind the Hugging Face link above.