Run unsloth/gemma-4-31B-it-GGUF locally

License: apache-2.0 ⬇ 591,899 ❤ 505
Parameters30.7B
Context262,144

unsloth/gemma-4-31B-it-GGUF is a very large instruction-tuned chat model with 30.7 billion parameters, built on the gemma4 architecture. It is released under the apache-2.0 license and has been downloaded 591,899 times.

To run unsloth/gemma-4-31B-it-GGUF locally at a 4,096-token context, its quantized versions need between 3.95 GB (GGUF, lowest quality) and 62.68 GB (BF16, highest quality) of memory, weights plus KV cache and a system margin included.

For most users the best balance is F32, needing about 5.62 GB. That means unsloth/gemma-4-31B-it-GGUF fits entirely in the VRAM of a 6 GB GPU or larger, running fully on the GPU.

→ Guide: How much VRAM do you need?

All quantizations

Quant.Bits QualityWeights KVTotal Speed~Verdict
GGUF 0.13 Very low 0.48 GB 2.67 GB 3.95 GB 834.5 t/s Fits in VRAM
F16 0.56 Very low 2.01 GB 2.67 GB 5.48 GB 199.4 t/s Fits in VRAM
F32 0.6 Very low 2.14 GB 2.67 GB 5.62 GB 186.5 t/s Fits in VRAM
IQ2_XXS 2.22 Very low 7.95 GB 2.67 GB 11.42 GB 6.3 t/s Offload
IQ2_M 2.8 Low 10.01 GB 2.67 GB 13.49 GB 5.0 t/s Offload
Q2_K_XL 3.07 Low 10.97 GB 2.67 GB 14.44 GB 4.6 t/s Offload
IQ3_XXS 3.09 Low 11.02 GB 2.67 GB 14.5 GB 4.5 t/s Offload
Q3_K_S 3.44 Fair 12.3 GB 2.67 GB 15.78 GB 4.1 t/s Offload
Q3_K_M 3.84 Fair 13.72 GB 2.67 GB 17.2 GB 3.6 t/s Offload
Q3_K_XL 4.01 Fair 14.32 GB 2.67 GB 17.8 GB 3.5 t/s Offload
IQ4_XS 4.27 Good 15.25 GB 2.67 GB 18.72 GB 3.3 t/s Offload
IQ4_NL 4.51 Good 16.1 GB 2.67 GB 19.57 GB 3.1 t/s Offload
Q4_0 4.52 Good 16.15 GB 2.67 GB 19.62 GB 3.1 t/s Offload
Q4_K_S 4.53 Good 16.2 GB 2.67 GB 19.68 GB 3.1 t/s Offload
Q4_K_M 4.78 Good 17.07 GB 2.67 GB 20.54 GB 2.9 t/s Offload
Q4_K_XL 4.91 Good 17.53 GB 2.67 GB 21.0 GB 2.9 t/s Offload
Q4_1 4.98 Good 17.81 GB 2.67 GB 21.28 GB 2.8 t/s Offload
Q5_K_S 5.51 Very good 19.67 GB 2.67 GB 23.15 GB 2.5 t/s Offload
Q5_K_M 5.64 Very good 20.17 GB 2.67 GB 23.64 GB 2.5 t/s Offload
Q5_K_XL 5.7 Very good 20.39 GB 2.67 GB 23.86 GB 2.5 t/s Offload
Q6_K 6.57 Excellent 23.47 GB 2.67 GB 26.94 GB Insufficient
Q6_K_XL 7.17 Excellent 25.63 GB 2.67 GB 29.1 GB Insufficient
Q8_0 8.64 Excellent 30.87 GB 2.67 GB 34.35 GB Insufficient
Q8_K_XL 9.13 Excellent 32.61 GB 2.67 GB 36.09 GB Insufficient
BF16 16.57 Excellent 59.2 GB 2.67 GB 62.68 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 unsloth/gemma-4-31B-it-GGUF?

You need about 5.62 GB of VRAM to run unsloth/gemma-4-31B-it-GGUF entirely on the GPU using the F32 quantization (at a 4,096-token context). Smaller quantizations lower the requirement at the cost of quality.

Can I run unsloth/gemma-4-31B-it-GGUF on an 8 GB GPU?

Yes. With 8 GB of VRAM you can run unsloth/gemma-4-31B-it-GGUF fully on the GPU using F32 (about 5.62 GB).

Can I run unsloth/gemma-4-31B-it-GGUF on a 16 GB GPU?

Yes. With 16 GB of VRAM you can run unsloth/gemma-4-31B-it-GGUF fully on the GPU using Q3_K_S (about 15.78 GB).

Can I run unsloth/gemma-4-31B-it-GGUF on a 24 GB GPU?

Yes. With 24 GB of VRAM you can run unsloth/gemma-4-31B-it-GGUF fully on the GPU using Q5_K_XL (about 23.86 GB).

What is the best quantization for unsloth/gemma-4-31B-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.