MI

MiniCPM

MiniCPMGGUF by vbuhoijymzoiUpdated 2024-02-10
Download on Hugging Face
Parameters2.72B
Context2K
Layers40
Architectureminicpm
Smallest quant1.5 GB
Quick answer

MiniCPM needs about 5.5 GB of memory at BF16 (a 2.0 GB file) with an 8K context window, so it fits on a 6 GB GPU. This repository has 6 quants, from 1.5 GB (BF16) to 5.5 GB (BF16).

MiniCPM is a 2.72B-parameter MiniCPM model in GGUF format, quantized by vbuhoijymzoi. Its trained context length is 2K tokens. The repository has 6 quant files, from 1.5 GB (BF16) to 5.5 GB (BF16).

Quants & memory

QuantFileEst. memoryvs VRAMResultDownload
BF161.5 GB 2.7 GB
Fits Get file →
BF161.7 GB 3.0 GB
Fits Get file →
BF162.0 GB 3.3 GB
Fits Get file →
BF162.4 GB 3.6 GB
Fits Get file →
BF162.9 GB 4.2 GB
Fits Get file →
BF165.5 GB 6.7 GB
Fits Get file →

What GPU do you need for MiniCPM?

These figures add the file size, an fp16 KV cache at an 8K context window (worked out from this model’s own header) and about 0.5 GB for the runtime. They are estimates, so leave a little headroom, and expect to use less context or a smaller quant if you are right at the limit.

  • 4 GB VRAM: no quant fits at an 8K context.
  • 6 GB VRAM: the best quant that fits is BF16 (2.4 GB file, about 5.9 GB in use).
  • 8 GB VRAM: the best quant that fits is BF16 (2.9 GB file, about 6.4 GB in use).
  • 12 GB VRAM: the best quant that fits is BF16 (5.5 GB file, about 9.0 GB in use).

Which MiniCPM quant should you choose?

BF16 (2.0 GB) is the middle option in this repository. The unquantized model. It gives original quality and needs the most memory, so it is mainly for conversion or reference.

If memory is tight, the smallest file is BF16 at 1.5 GB, with the quality trade-offs that come with fewer bits. The unquantized model. It gives original quality and needs the most memory, so it is mainly for conversion or reference.

New to quant names? Read GGUF quantization explained.

Specifications

Architecture from the GGUF header

Architectureminicpm
Parameters2.72B
Layers40
Trained context2,048 tokens
KV cache (fp16)360 KB per token

Source

Quantized byvbuhoijymzoi
FamilyMiniCPM
Downloads (HF)98
Last updated2024-02-10

Frequently asked questions about MiniCPM

How much VRAM does MiniCPM need?

About 5.5 GB at BF16 with an 8K context window. That is the 2.0 GB file, plus the KV cache and roughly 0.5 GB of runtime overhead. Smaller quants need less and longer context needs more.

Can I run MiniCPM on an 8 GB GPU?

Yes, with BF16: the 2.9 GB file needs about 6.4 GB at 8K context. Larger quants in this repository do not fit in 8 GB.

What is the file size of MiniCPM GGUF?

It depends on the quant: BF16 is 1.5 GB, BF16 is 1.7 GB, BF16 is 2.0 GB, BF16 is 2.4 GB, BF16 is 2.9 GB, BF16 is 5.5 GB.

What context length does MiniCPM support?

The GGUF header lists a trained context length of 2,048 tokens (2K). The KV cache grows with the context you use: about 3.0 GB at 8K and 0.8 GB at 2K.

Where can I download MiniCPM GGUF?

The files are on Hugging Face in the vbuhoijymzoi/MiniCPM-2B-sft-bf16-GGUF repository. The download buttons on this page open the file on Hugging Face. ggufmodels.com does not host any model files.

Similar models

Browse all →

More from vbuhoijymzoi

All by vbuhoijymzoi →

More MiniCPM models

All MiniCPM →

Other tiny models

All tiny models →