Kimi-K2-Instruct GGUF size and VRAM requirements

License: other ⬇ 19,045 ❤ 229
Parameters1026.41B
Context131,072

unsloth/Kimi-K2-Instruct-GGUF is a very large instruction-tuned chat model with 1026.41 billion parameters, built on the deepseek2 architecture. It is released under the other license and has been downloaded 19,045 times.

To run unsloth/Kimi-K2-Instruct-GGUF locally at a 4,096-token context, its quantized versions need between 234.35 GB (Q1_0, lowest quality) and 1919.62 GB (BF16, highest quality) of memory, weights plus KV cache and a system margin included.

Available GGUF quantizations for unsloth/Kimi-K2-Instruct-GGUF include Q1_0, IQ1_S, IQ1_M, IQ2_XXS, IQ2_M, Q2_K, Q2_K_L, Q2_K_XL, IQ3_XXS, Q3_K_S, Q3_K_XL, Q3_K_M, IQ4_XS, IQ4_NL, Q4_0, Q4_K_S, Q4_K_XL, Q4_K_M, Q4_1, 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 131,072 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
Q1_0 1.9 Very low 226.87 GB 6.67 GB 234.35 GB Insufficient
IQ1_S 2.18 Very low 260.88 GB 6.67 GB 268.35 GB Insufficient
IQ1_M 2.37 Very low 283.34 GB 6.67 GB 290.81 GB Insufficient
IQ2_XXS 2.56 Very low 306.2 GB 6.67 GB 313.68 GB Insufficient
IQ2_M 2.71 Low 323.27 GB 6.67 GB 330.74 GB Insufficient
Q2_K 2.91 Low 347.55 GB 6.67 GB 355.02 GB Insufficient
Q2_K_L 2.91 Low 347.81 GB 6.67 GB 355.28 GB Insufficient
Q2_K_XL 2.98 Low 355.64 GB 6.67 GB 363.11 GB Insufficient
IQ3_XXS 3.25 Low 388.01 GB 6.67 GB 395.48 GB Insufficient
Q3_K_S 3.45 Fair 412.03 GB 6.67 GB 419.51 GB Insufficient
Q3_K_XL 3.52 Fair 421.03 GB 6.67 GB 428.5 GB Insufficient
Q3_K_M 3.81 Fair 455.77 GB 6.67 GB 463.24 GB Insufficient
IQ4_XS 4.26 Good 508.98 GB 6.67 GB 516.45 GB Insufficient
IQ4_NL 4.51 Good 538.76 GB 6.67 GB 546.23 GB Insufficient
Q4_0 4.53 Good 540.74 GB 6.67 GB 548.22 GB Insufficient
Q4_K_S 4.54 Good 542.73 GB 6.67 GB 550.2 GB Insufficient
Q4_K_XL 4.58 Good 546.79 GB 6.67 GB 554.26 GB Insufficient
Q4_K_M 4.84 Good 578.15 GB 6.67 GB 585.62 GB Insufficient
Q4_1 5.01 Very good 598.4 GB 6.67 GB 605.87 GB Insufficient
Q5_K_S 5.51 Very good 658.04 GB 6.67 GB 665.51 GB Insufficient
Q5_K_M 5.68 Very good 678.33 GB 6.67 GB 685.8 GB Insufficient
Q5_K_XL 5.69 Very good 680.4 GB 6.67 GB 687.88 GB Insufficient
Q6_K 6.57 Excellent 784.76 GB 6.67 GB 792.24 GB Insufficient
Q6_K_XL 6.85 Excellent 818.68 GB 6.67 GB 826.15 GB Insufficient
Q8_0 8.5 Excellent 1016.12 GB 6.67 GB 1023.59 GB Insufficient
Q8_K_XL 9.28 Excellent 1108.33 GB 6.67 GB 1115.8 GB Insufficient
BF16 16.0 Excellent 1912.15 GB 6.67 GB 1919.62 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/Kimi-K2-Instruct-GGUF?

unsloth/Kimi-K2-Instruct-GGUF is an instruction-tuned chat model with 1026.41 billion parameters, based on the deepseek2 architecture. It is released under the other license and distributed as GGUF files for local inference.

Can I run unsloth/Kimi-K2-Instruct-GGUF on an 8 GB GPU?

No. unsloth/Kimi-K2-Instruct-GGUF does not fit on an 8 GB GPU, even with the smallest quantization and system RAM offloading.

Can I run unsloth/Kimi-K2-Instruct-GGUF on a 16 GB GPU?

No. unsloth/Kimi-K2-Instruct-GGUF does not fit on a 16 GB GPU, even with the smallest quantization and system RAM offloading.

Can I run unsloth/Kimi-K2-Instruct-GGUF on a 24 GB GPU?

No. unsloth/Kimi-K2-Instruct-GGUF does not fit on a 24 GB GPU, even with the smallest quantization and system RAM offloading.

What context length does unsloth/Kimi-K2-Instruct-GGUF support?

unsloth/Kimi-K2-Instruct-GGUF supports a native context length of up to 131,072 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/Kimi-K2-Instruct-GGUF?

For unsloth/Kimi-K2-Instruct-GGUF, a strong default is Q4_K_M, which needs about 585.62 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.