Run ggml-org/gemma-4-12B-it-GGUF locally
ggml-org/gemma-4-12B-it-GGUF is a large instruction-tuned chat model with 11.91 billion parameters, built on the gemma4 architecture. It has been downloaded 373,264 times.
To run ggml-org/gemma-4-12B-it-GGUF locally at a 4,096-token context, its quantized versions need between 9.34 GB (Q4_K_M, lowest quality) and 24.82 GB (BF16, highest quality) of memory, weights plus KV cache and a system margin included.
For most users the best balance is Q8_0, needing about 14.41 GB. That means ggml-org/gemma-4-12B-it-GGUF fits entirely in the VRAM of a 10 GB GPU or larger, running fully on the GPU.
All quantizations
| Quant. | Bits | Quality | Weights | KV | Total | Speed~ | Verdict |
|---|---|---|---|---|---|---|---|
| Q4_K_M | 4.96 | Good | 6.87 GB | 1.67 GB | 9.34 GB | 7.3 t/s | Offload |
| Q8_0 | 8.62 | Excellent | 11.95 GB | 1.67 GB | 14.41 GB | 4.2 t/s | Offload |
| BF16 | 16.13 | Excellent | 22.36 GB | 1.67 GB | 24.82 GB | — | Insufficient |
KV cache estimated (architecture unavailable). Speed is a rough estimate bounded by memory bandwidth.
Frequently asked questions
How much VRAM do you need to run ggml-org/gemma-4-12B-it-GGUF?
You need about 9.34 GB of VRAM to run ggml-org/gemma-4-12B-it-GGUF entirely on the GPU using the Q4_K_M quantization (at a 4,096-token context). Smaller quantizations lower the requirement at the cost of quality.
Can I run ggml-org/gemma-4-12B-it-GGUF on an 8 GB GPU?
Partially. ggml-org/gemma-4-12B-it-GGUF only fits on an 8 GB GPU by offloading part of it to system RAM (with Q8_0), which runs but is slower.
Can I run ggml-org/gemma-4-12B-it-GGUF on a 16 GB GPU?
Yes. With 16 GB of VRAM you can run ggml-org/gemma-4-12B-it-GGUF fully on the GPU using Q8_0 (about 14.41 GB).
Can I run ggml-org/gemma-4-12B-it-GGUF on a 24 GB GPU?
Yes. With 24 GB of VRAM you can run ggml-org/gemma-4-12B-it-GGUF fully on the GPU using Q8_0 (about 14.41 GB).
What is the best quantization for ggml-org/gemma-4-12B-it-GGUF?
If memory allows, higher bits-per-weight means better quality. A common sweet spot is a Q4_K_M or Q5_K_M quantization, which keeps most of the quality while roughly halving the memory versus 8-bit. Pick the highest quantization that still fits in your VRAM.