AxForge
Sign in Get started

Model reference · open weights

gemma-2-bnb

LLMs unsloth Text gen 2 builds Open, with conditions 41k dl/mo

gemma-2-bnb is an open-weight language model from unsloth. gemma-2-2b-it-bnb-4bit (INT4) weighs 2.2 GB; the smallest configuration that runs it is RTX 3060 12 GB.

Request Rent a GPU

For self-hosting, compare this build's requirements below with the GPU you plan to rent.

  • Gemma-2-bnb is a 2.7B parameter text-generation model created by unsloth and published as unsloth/gemma-2-2b-it-bnb-4bit.
  • It supports English with an 8192 token context length and is released under the gemma licence.
  • The model is directly quantized to 4bit using bitsandbytes.

Summary of the unsloth/gemma-2-2b-it-bnb-4bit model card, 2026-10-01

Hugging Face ↗

What it is

Released byunsloth
Released2024-07-31
Parameters2.7B
VRAM2.2 GB for the weights

What it runs on

Memory and cards for gemma-2-2b-it-bnb-4bit (INT4)

2.2 GBweights, file size
53 MBcache per 1K tokens
436 MBwindow cache per request
805 MBruntime overhead, at least
8,192 tokenscontext max
CardRequests at onceContext maxMemory
8K each
RTX 3060 12 GB9all 8K11.6 GB
RTX 4060 Ti 16 GB14all 8K15.4 GB
RTX 3090 24 GB23all 8K23.4 GB
RTX 4090 24 GB23all 8K23.4 GB
RTX 5090 32 GB32all 8K31.0 GB
L40S 48 GB46all 8K44.0 GB
A100 80 GB86all 8K78.2 GB
H100 80 GB79all 8K78.1 GB
RTX PRO 6000 Blackwell 96 GB97all 8K93.8 GB
DGX Spark (GB10) 128 GB unified113all 8K107 GB
H200 141 GB148all 8K138 GB
B200 180 GB191all 8K176 GB
Memory needed at each load
Requests at once8K tokens each
13.9 GB
57.4 GB
810.0 GB
1617.0 GB
3230.9 GB
6458.9 GB

One card, with vLLM's small-card settings.

Builds

Sizes, precisions & builds

BuildParamsPrecisionWeightsSmallest setup
gemma-2-2b-it-bnb-4bit (above) ↗ 2.7BINT4 2.2 GBRTX 3060 12 GB
gemma-2-27b-it-bnb-4bit ↗ 28.0BINT4 15.8 GB2× RTX 3060 12 GB

From the model card

What unsloth says about gemma-2-bnb

Read the model card

Reminder to use the dev version Transformers:

pip install git+https://github.com/huggingface/transformers.git

Finetune Gemma, Llama 3, Mistral 2-5x faster with 70% less memory via Unsloth!

Directly quantized 4bit model with bitsandbytes.

All notebooks are beginner friendly! Add your dataset, click "Run All", and you'll get a 2x faster finetuned model which can be exported to GGUF, vLLM or uploaded to Hugging Face.

  • * Kaggle has 2x T4s, but we use 1. Due to overhead, 1x T4 is 5x faster.

Quoted from the model card on Hugging Face. The full card is behind the Hugging Face link above.

© 2026 AxForge · EU-hosted AI infrastructure Pricing Docs Trust Privacy Terms