Laguna-S-2.1 GGUF size and VRAM requirements

License: openmdw-1.1 ⬇ 129,601 ❤ 236
Parameters117.56B
Context262,144

unsloth/Laguna-S-2.1-GGUF is a very large language model with 117.56 billion parameters, built on the laguna architecture. It is released under the openmdw-1.1 license and has been downloaded 129,601 times.

To run unsloth/Laguna-S-2.1-GGUF locally at a 4,096-token context, its quantized versions need between 32.44 GB (IQ1_S, lowest quality) and 220.04 GB (BF16, highest quality) of memory, weights plus KV cache and a system margin included.

Available GGUF quantizations for unsloth/Laguna-S-2.1-GGUF include IQ1_S, IQ1_M, IQ2_XXS, IQ2_M, Q2_K_XL, IQ3_XXS, IQ3_S, Q3_K_M, Q3_K_XL, IQ4_XS, IQ4_NL, Q4_K_S, GGUF, Q4_K_M, Q4_K_XL, Q5_K_S, Q5_K_M, Q5_K_XL, Q6_K, Q6_K_XL, Q8_0, Q8_K_XL, BF16. The model supports a native context length of up to 262,144 tokens; a longer context grows the KV cache and the memory needed.

→ Guide: How much VRAM do you need?

GGUF file size and memory by quantization

Compare real GGUF weight sizes, estimated KV cache and total memory for Q4, Q5, Q8 and every quantization published in this repository.

Quant.Bits QualityWeights KVTotal Speed~Verdict
IQ1_S 2.3 Very low 31.45 GB 0.19 GB 32.44 GB Insufficient
IQ1_M 2.43 Very low 33.19 GB 0.19 GB 34.18 GB Insufficient
IQ2_XXS 2.53 Very low 34.64 GB 0.19 GB 35.62 GB Insufficient
IQ2_M 2.54 Very low 34.71 GB 0.19 GB 35.7 GB Insufficient
Q2_K_XL 2.7 Low 36.96 GB 0.19 GB 37.95 GB Insufficient
IQ3_XXS 3.01 Low 41.24 GB 0.19 GB 42.23 GB Insufficient
IQ3_S 3.3 Low 45.1 GB 0.19 GB 46.09 GB Insufficient
Q3_K_M 3.68 Fair 50.31 GB 0.19 GB 51.3 GB Insufficient
Q3_K_XL 3.68 Fair 50.38 GB 0.19 GB 51.37 GB Insufficient
IQ4_XS 3.92 Fair 53.61 GB 0.19 GB 54.6 GB Insufficient
IQ4_NL 4.0 Fair 54.71 GB 0.19 GB 55.7 GB Insufficient
Q4_K_S 4.67 Good 63.88 GB 0.19 GB 64.87 GB Insufficient
GGUF 4.84 Good 66.2 GB 0.19 GB 67.19 GB Insufficient
Q4_K_M 4.98 Good 68.1 GB 0.19 GB 69.09 GB Insufficient
Q4_K_XL 4.99 Good 68.35 GB 0.19 GB 69.34 GB Insufficient
Q5_K_S 5.62 Very good 76.98 GB 0.19 GB 77.97 GB Insufficient
Q5_K_M 5.98 Very good 81.83 GB 0.19 GB 82.81 GB Insufficient
Q5_K_XL 5.99 Very good 82.02 GB 0.19 GB 83.01 GB Insufficient
Q6_K 6.66 Excellent 91.19 GB 0.19 GB 92.18 GB Insufficient
Q6_K_XL 7.29 Excellent 99.73 GB 0.19 GB 100.71 GB Insufficient
Q8_0 8.51 Excellent 116.44 GB 0.19 GB 117.42 GB Insufficient
Q8_K_XL 8.72 Excellent 119.31 GB 0.19 GB 120.3 GB Insufficient
BF16 16.01 Excellent 219.05 GB 0.19 GB 220.04 GB Insufficient

KV cache computed from the model's exact architecture. Speed is a rough estimate bounded by memory bandwidth.

Frequently asked questions

What kind of model is unsloth/Laguna-S-2.1-GGUF?

unsloth/Laguna-S-2.1-GGUF is a language model with 117.56 billion parameters, based on the laguna architecture. It is released under the openmdw-1.1 license and distributed as GGUF files for local inference.

How much VRAM do you need to run unsloth/Laguna-S-2.1-GGUF?

You need about 46.09 GB of VRAM to run unsloth/Laguna-S-2.1-GGUF entirely on the GPU using the IQ3_S quantization (at a 4,096-token context). Smaller quantizations lower the requirement at the cost of quality.

Can I run unsloth/Laguna-S-2.1-GGUF on an 8 GB GPU?

No. unsloth/Laguna-S-2.1-GGUF does not fit on an 8 GB GPU, even with the smallest quantization and system RAM offloading.

Can I run unsloth/Laguna-S-2.1-GGUF on a 16 GB GPU?

Partially. unsloth/Laguna-S-2.1-GGUF only fits on a 16 GB GPU by offloading part of it to system RAM (with IQ3_S), which runs but is slower.

Can I run unsloth/Laguna-S-2.1-GGUF on a 24 GB GPU?

Partially. unsloth/Laguna-S-2.1-GGUF only fits on a 24 GB GPU by offloading part of it to system RAM (with Q4_K_XL), which runs but is slower.

What context length does unsloth/Laguna-S-2.1-GGUF support?

unsloth/Laguna-S-2.1-GGUF supports a native context length of up to 262,144 tokens. A longer context grows the KV cache, so it increases the memory needed to run the model.

What is the best quantization for unsloth/Laguna-S-2.1-GGUF?

For unsloth/Laguna-S-2.1-GGUF, a strong default is Q4_K_M, which needs about 69.09 GB and keeps most of the quality while roughly halving the memory versus 8-bit. With VRAM to spare, Q5_K_M or Q6_K add a little more quality; if you are tight on memory, a smaller quantization still runs. Pick the highest quantization that fits your VRAM.