Skip to content
localllm
日本語

meta-llama

Llama-4-Scout-17B-16E-Instruct

Image input

Parameters
107.8B
Max context length
10,485,760
Architecture
llama4
License
Conditional
Base model
meta-llama/Llama-4-Scout-17B-16E-Instruct
License
llama4
llama.cpp
b5074+

About Llama-4-Scout-17B-16E-Instruct

Llama-4-Scout-17B-16E-Instruct is a GGUF model based on meta-llama/Llama-4-Scout-17B-16E-Instruct on Hugging Face. GGUF files are published by unsloth/Llama-4-Scout-17B-16E-Instruct-GGUF. It has about 107.8B parameters, the llama4 architecture, a maximum context length of 10,485,760 tokens (values written in the GGUF).

There are 27 quantizations, with file sizes from 29.3 GB (UD-TQ1_0) to 216 GB (BF16) (actual sizes on Hugging Face).

Running Q4_K_M (65.4 GB) with an 8,192-token context needs an estimated 67.9 GB of memory, and an estimated 69.2 GB at 32,768 tokens. For 8,192 tokens, a GPU with 64 GB or more of VRAM, or a Mac with 64 GB or more of unified memory, is a rough guide (macOS caps how much memory the GPU can use, so leave some headroom). The smallest, UD-TQ1_0 (29.3 GB), needs an estimated 31.8 GB at 8,192 tokens.

The original model is licensed under Llama 4 Community License, which allows commercial use with conditions such as scale or purpose. Check the original license for the terms of use. An mmproj file for image input is also available.

File size and estimated memory per quantization

  • Up to 8 GB
  • Up to 16 GB
  • Up to 24 GB
  • Up to 48 GB
  • Up to 128 GB
  • More

File sizes are the actual values on Hugging Face. Memory is estimated for each context length (KV cache f16). The formula is in How memory is calculated.

All memory figures are Estimated (KV cache f16). It should run on a setup with at least this much VRAM (or Mac unified memory). Click a value to see the devices and models that fit that amount (Find by GPU / Mac).

Using image input: This model supports image input. To use images, an mmproj file is loaded as well.

About the "quality guide" (how far a quantization drifts from the original)

For each quantization type, how far the output drifts from the original model, grouped into six levels from values llama.cpp measured on Llama-3-8B.

The value is mean KLD (the difference between the original and quantized model in next-token probabilities); smaller means closer to the original. Quantizations made with an imatrix (importance data) come out slightly smaller. Levels use the value without an imatrix (or with one, for types that only have that).

These values were not measured on this model. Larger models and MoE models tend to drift less with the same type. Distributor-specific mixed types (such as UD-Q4_K_XL) cannot be matched by type name alone, so they show "-".

Quantization Quality guide KLD (no imatrix) KLD (imatrix) Size (Llama-3-8B)
Q8_0 Nearly identical 0.0014 - 7.96 GiB
Q6_K Very small 0.0055 - 6.14 GiB
Q5_K_M Very small 0.0108 - 5.33 GiB
Q5_K_S Small 0.0166 - 5.21 GiB
Q5_1 Small 0.0180 - 5.65 GiB
Q5_0 Small 0.0222 - 5.21 GiB
Q4_K_M Small 0.0313 0.0282 4.58 GiB
Q4_K_S Medium 0.0431 0.0320 4.37 GiB
IQ4_NL Small - 0.0357 4.35 GiB
IQ4_XS Small - 0.0363 4.14 GiB
Q4_1 Medium 0.0717 - 4.78 GiB
Q4_0 Medium 0.0719 - 4.34 GiB
Q3_K_L Medium 0.0872 0.0731 4.03 GiB
Q3_K_M Medium 0.1019 0.0844 3.74 GiB
IQ3_M Medium - 0.1025 3.53 GiB
IQ3_S Medium - 0.1113 3.42 GiB
IQ3_XS Large - 0.1387 3.28 GiB
IQ3_XXS Large - 0.1836 3.05 GiB
Q3_K_S Large 0.2282 0.1998 3.41 GiB
IQ2_M Large - 0.3260 2.74 GiB
Q2_K Very large 0.4451 0.3322 2.96 GiB
Q2_K_S Very large - 0.4034 2.96 GiB
IQ2_S Very large - 0.4392 2.56 GiB
IQ2_XS Very large - 0.5463 2.43 GiB
IQ2_XXS Very large - 0.8120 2.24 GiB
IQ1_M Very large - 1.3931 2.01 GiB
IQ1_S Very large - 2.2546 1.88 GiB

Source: llama.cpp (MIT), measured on Llama-3-8B against FP16, Wikitext-2, revision f364eb6f. Original table · Reading and choosing quantizations

GGUF source

unsloth/Llama-4-Scout-17B-16E-Instruct-GGUF

Converted and published by a third party About this label

Using the original model requires a request or agreement on Hugging Face (this does not affect downloading the GGUF files).

Last checked: Nov 11, 2026

Architecture
llama4
llama.cpp b5074 and later
Parameters
107.8B
Max context length
10,485,760
License
llama4 Conditional
File size and memory per quantization for Llama-4-Scout-17B-16E-Instruct (unsloth/Llama-4-Scout-17B-16E-Instruct-GGUF)
Quantization File size Measured Quality guide 4,096 8,192 16,384 32,768 65,536 131,072 262,144 10,485,760 max
UD-TQ1_0 29.3 GB - 31.0 GB 31.8 GB 32.3 GB 33.1 GB 34.7 GB 37.9 GB 44.4 GB 547 GB
UD-IQ1_S 32.5 GB - 34.2 GB 35.0 GB 35.5 GB 36.3 GB 37.9 GB 41.1 GB 47.6 GB 550 GB
UD-IQ1_M 35.0 GB - 36.7 GB 37.5 GB 38.0 GB 38.8 GB 40.4 GB 43.7 GB 50.1 GB 553 GB
UD-IQ2_XXS 37.4 GB - 39.2 GB 40.0 GB 40.4 GB 41.2 GB 42.9 GB 46.1 GB 52.5 GB 555 GB
UD-IQ2_M 39.1 GB - 40.8 GB 41.6 GB 42.1 GB 42.9 GB 44.5 GB 47.8 GB 54.2 GB 557 GB
Q2_K 39.6 GB Very large 41.3 GB 42.1 GB 42.6 GB 43.4 GB 45.0 GB 48.2 GB 54.7 GB 557 GB
Q2_K_L 39.8 GB - 41.6 GB 42.4 GB 42.8 GB 43.7 GB 45.3 GB 48.5 GB 54.9 GB 557 GB
UD-Q2_K_XL 42.4 GB - 44.1 GB 44.9 GB 45.4 GB 46.2 GB 47.8 GB 51.1 GB 57.5 GB 560 GB
UD-IQ3_XXS 45.7 GB - 47.5 GB 48.3 GB 48.8 GB 49.6 GB 51.2 GB 54.4 GB 60.8 GB 563 GB
Q3_K_S 46.7 GB Large 48.5 GB 49.3 GB 49.8 GB 50.6 GB 52.2 GB 55.4 GB 61.9 GB 564 GB
UD-Q3_K_XL 49.0 GB - 50.8 GB 51.6 GB 52.1 GB 52.9 GB 54.5 GB 57.7 GB 64.1 GB 567 GB
Q3_K_M (2 parts) 51.8 GB Medium 53.5 GB 54.3 GB 54.8 GB 55.6 GB 57.2 GB 60.4 GB 66.9 GB 569 GB
IQ4_XS (2 parts) 57.7 GB Small 59.4 GB 60.2 GB 60.7 GB 61.5 GB 63.1 GB 66.3 GB 72.8 GB 575 GB
IQ4_NL (2 parts) 60.9 GB Small 62.7 GB 63.5 GB 64.0 GB 64.8 GB 66.4 GB 69.6 GB 76.1 GB 579 GB
Q4_0 (2 parts) 61.2 GB Medium 62.9 GB 63.7 GB 64.2 GB 65.0 GB 66.6 GB 69.9 GB 76.3 GB 579 GB
Q4_K_S (2 parts) 61.5 GB Medium 63.2 GB 64.0 GB 64.5 GB 65.3 GB 66.9 GB 70.1 GB 76.6 GB 579 GB
UD-Q4_K_XL (2 parts) 62.0 GB - 63.8 GB 64.6 GB 65.0 GB 65.8 GB 67.5 GB 70.7 GB 77.1 GB 580 GB
Q4_K_M (2 parts) 65.4 GB Small 67.1 GB 67.9 GB 68.4 GB 69.2 GB 70.8 GB 74.0 GB 80.5 GB 583 GB
Q4_1 (2 parts) 67.6 GB Medium 69.3 GB 70.1 GB 70.6 GB 71.4 GB 73.0 GB 76.3 GB 82.7 GB 585 GB
Q5_K_S (2 parts) 74.3 GB Small 76.0 GB 76.8 GB 77.3 GB 78.1 GB 79.7 GB 82.9 GB 89.4 GB 592 GB
Q5_K_M (2 parts) 76.5 GB Very small 78.3 GB 79.1 GB 79.6 GB 80.4 GB 82.0 GB 85.2 GB 91.7 GB 594 GB
UD-Q5_K_XL (2 parts) 79.1 GB - 80.9 GB 81.7 GB 82.2 GB 83.0 GB 84.6 GB 87.8 GB 94.3 GB 597 GB
Q6_K (2 parts) 88.4 GB Very small 90.2 GB 91.0 GB 91.5 GB 92.3 GB 93.9 GB 97.1 GB 104 GB 606 GB
UD-Q6_K_XL (2 parts) 94.1 GB - 95.8 GB 96.6 GB 97.1 GB 97.9 GB 99.5 GB 103 GB 109 GB 612 GB
Q8_0 (3 parts) 115 GB Nearly identical 116 GB 117 GB 118 GB 118 GB 120 GB 123 GB 130 GB 632 GB
UD-Q8_K_XL (3 parts) 128 GB - 130 GB 131 GB 131 GB 132 GB 134 GB 137 GB 143 GB 646 GB
BF16 (5 parts) 216 GB Original precision 217 GB 218 GB 219 GB 219 GB 221 GB 224 GB 231 GB 733 GB

The mmproj file for image input (BF16) is 1.75 GB Measured . It is not included in the memory above, so add it if you use images. 

Ollama  ollama run hf.co/unsloth/Llama-4-Scout-17B-16E-Instruct-GGUF:Q4_K_M
llama.cpp  llama-server -hf unsloth/Llama-4-Scout-17B-16E-Instruct-GGUF:Q4_K_M

Third-party evaluations

Values from other organizations for the original model before quantization. Not this site's own rating. Values are for the original model before quantization. GGUF quantization (especially small quantizations) can lower performance.

Human preference (LMArena)

1,322 1,317–1,326

Llama 4 Scout 17B 16E Instruct · 29,777 votes · Values as of Nov 9, 2026

Source: LMArena (CC BY 4.0). Original leaderboard · See in ranking

Hard science (GPQA)

51.8% 45.6%–58.0%

Llama-4-Scout-17B-16E-Instruct · Values as of Apr 8, 2025

Source: GPQA Diamond (Epoch AI) (CC BY 4.0). Original leaderboard · See in ranking

License and commercial use

Llama 4 Community License Conditional

Commercial use is allowed. If, on the Llama 4 release date, you or your affiliates had more than 700 million monthly active users, you need a separate license from Meta. When redistributing, or offering a product or service that contains the model, include the license text and prominently display "Built with Llama". If you use it to create, train or fine-tune an AI model that you distribute, start that model's name with "Llama". You must follow the Acceptable Use Policy.

Based on the license on the original model's Hugging Face page. Original license text (checked Nov 1, 2026)

The summary states facts this site read from the original text and is not legal advice. Always check the original license and its latest terms before use. All licenses

How to run

Mainline llama.cpp can load this model's format (llama4) from b5074 (Apr 7, 2025) onward. Ollama and LM Studio bundle llama.cpp internally, so an older app fails to load it with errors such as "unknown model architecture". If that happens, update the app.

Example commands using Q4_K_M. To use another quantization, replace the name with one from the table above. How to run local LLMs

Ollama

Run it by pointing directly at the GGUF on Hugging Face.

ollama run hf.co/unsloth/Llama-4-Scout-17B-16E-Instruct-GGUF:Q4_K_M

llama.cpp

Pass the repository and quantization name to llama-server (or llama-cli) with -hf.

llama-server -hf unsloth/Llama-4-Scout-17B-16E-Instruct-GGUF:Q4_K_M

LM Studio

Paste unsloth/Llama-4-Scout-17B-16E-Instruct-GGUF into the search box, download it, and choose the quantization from the file list.

Frequently asked questions

How much memory do I need to run Llama-4-Scout-17B-16E-Instruct (Q4_K_M)?
The file size is 65.4 GB (measured). With a 8,192-token context, the estimated memory including the KV cache is 67.9 GB. The formula is published at https://localllm.az-arte.com/en/guide/memory-calculation.
Can Llama-4-Scout-17B-16E-Instruct be used commercially?
The original model's license is Llama 4 Community License (Commercial use with conditions).Commercial use is allowed. If, on the Llama 4 release date, you or your affiliates had more than 700 million monthly active users, you need a separate license from Meta. When redistributing, or offering a product or service that contains the model, include the license text and prominently display "Built with Llama". If you use it to create, train or fine-tune an AI model that you distribute, start that model's name with "Llama". You must follow the Acceptable Use Policy. Check the original license text before deciding.
Does Llama-4-Scout-17B-16E-Instruct run in 8 GB, 12 GB, 16 GB or 24 GB of VRAM?
8 GB: no quantization fits; 12 GB: no quantization fits; 16 GB: no quantization fits; 24 GB: no quantization fits (estimated at 8,192 tokens with an f16 KV cache, files from the most-downloaded source).
What is the command to run Llama-4-Scout-17B-16E-Instruct with Ollama?
Run "ollama run hf.co/unsloth/Llama-4-Scout-17B-16E-Instruct-GGUF:Q4_K_M" to use the GGUF on Hugging Face directly. To use another quantization, replace the name at the end.