Llama-4-Maverick-17B-128E-Instruct needs about 244.9 GB of memory at Q4_K_M (a 242.8 GB file) with an 8K context window, which is more than a 48 GB GPU holds. This repository has 26 quants, from 119.4 GB (IQ1_S) to 801.5 GB (BF16). Longer context needs more: at 32K tokens the KV cache alone adds about 6.4 GB.
17B Llama‑4 Maverick with 128 experts, multimodal, 1M context length
Llama‑4 Maverick is a 17 billion‑parameter model with 128 experts. It supports text and image inputs and outputs multilingual text and code. The model offers a 1 million‑token context window and is trained on a large multilingual dataset.
Good at
- Multimodal text and image understanding
- Large context handling
- High‑capacity language generation
Limits
- No vision support in current GGUF version
- Requires large VRAM for higher‑bit quantizations
- License is other, not open source
Best for
- Chatbots needing long context
- Image‑guided content creation
- Multilingual code assistance
Who it’s for. Developers building applications that require extensive context and multimodal input. Researchers exploring large‑scale multimodal models may also find it useful.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| IQ1_S | 119.4 GB (3 parts) | 121.5 GB | Too big | Get file → | |
| IQ1_M | 126.2 GB (3 parts) | 128.4 GB | Too big | Get file → | |
| IQ2_XXS | 134.1 GB (3 parts) | 136.2 GB | Too big | Get file → | |
| Q2_K | 145.6 GB (3 parts) | 147.8 GB | Too big | Get file → | |
| Q2_K_L | 145.9 GB (3 parts) | 148.0 GB | Too big | Get file → | |
| Q2_K_XL | 151.6 GB (4 parts) | 153.7 GB | Too big | Get file → | |
| IQ3_XXS | 168.7 GB (4 parts) | 170.8 GB | Too big | Get file → | |
| Q3_K_S | 172.7 GB (4 parts) | 174.8 GB | Too big | Get file → | |
| Q3_K_XL | 178.9 GB (4 parts) | 181.1 GB | Too big | Get file → | |
| Q3_K_M | 191.1 GB (4 parts) | 193.2 GB | Too big | Get file → | |
| IQ4_XS | 214.3 GB (5 parts) | 216.4 GB | Too big | Get file → | |
| IQ4_XS | 220.7 GB (5 parts) | 222.8 GB | Too big | Get file → | |
| IQ4_NL | 225.8 GB (5 parts) | 227.9 GB | Too big | Get file → | |
| Q4_K_S | 227.8 GB (5 parts) | 229.9 GB | Too big | Get file → | |
| Q4_K_XL | 232.1 GB (5 parts) | 234.2 GB | Too big | Get file → | |
| Q4_K_M | 242.8 GB (5 parts) | 244.9 GB | Too big | Get file → | |
| Q4_1 | 250.7 GB (6 parts) | 252.8 GB | Too big | Get file → | |
| Q5_K_S | 275.7 GB (6 parts) | 277.8 GB | Too big | Get file → | |
| Q5_K_M | 284.5 GB (6 parts) | 286.6 GB | Too big | Get file → | |
| Q5_K_XL | 287.0 GB (6 parts) | 289.1 GB | Too big | Get file → | |
| IQ2_M | 293.8 GB (7 parts) | 295.9 GB | Too big | Get file → | |
| Q6_K | 328.8 GB (7 parts) | 330.9 GB | Too big | Get file → | |
| Q6_K_XL | 341.1 GB (8 parts) | 343.2 GB | Too big | Get file → | |
| Q8_0 | 425.8 GB (9 parts) | 427.9 GB | Too big | Get file → | |
| Q8_K_XL | 460.0 GB (10 parts) | 462.1 GB | Too big | Get file → | |
| BF16 | 801.5 GB (18 parts) | 803.6 GB | Too big | Get file → |
For 8 GB VRAM, use Q4_K_S or Q4_K_M (≈230 GB file, ~230 GB memory). For 16 GB VRAM, Q5_K_S or Q5_K_M are suitable. For 24 GB VRAM, Q6_K or Q6_K_XL can be used. Memory estimates are approximate.
What GPU do you need for Llama-4-Maverick-17B-128E-Instruct?
These figures add the file size, an fp16 KV cache at an 8K context window (worked out from this model’s own header) and about 0.5 GB for the runtime. They are estimates, so leave a little headroom, and expect to use less context or a smaller quant if you are right at the limit.
- 4 GB VRAM: no quant fits at an 8K context.
- 6 GB VRAM: no quant fits at an 8K context.
- 8 GB VRAM: no quant fits at an 8K context.
- 12 GB VRAM: no quant fits at an 8K context.
- 16 GB VRAM: no quant fits at an 8K context.
- 24 GB VRAM: no quant fits at an 8K context.
- 32 GB VRAM: no quant fits at an 8K context.
- 48 GB VRAM: no quant fits at an 8K context.
Which Llama-4-Maverick-17B-128E-Instruct quant should you choose?
Q4_K_M is the usual starting point for Llama-4-Maverick-17B-128E-Instruct: it balances file size and quality well. Here it is a 242.8 GB file.
The Q8_0 file is 425.8 GB, about 1.8 times the size of Q4_K_M. It keeps the weights closer to the original, which only matters if you have the memory to spare.
If memory is tight, the smallest file is IQ1_S at 119.4 GB, with the quality trade-offs that come with fewer bits. Use it only when memory is very tight. Expect a clear drop in quality, which is usually worse on small models and easier to live with on large ones.
New to quant names? Read GGUF quantization explained.
Specifications
Architecture from the GGUF header
Frequently asked questions about Llama-4-Maverick-17B-128E-Instruct
How much VRAM does Llama-4-Maverick-17B-128E-Instruct need?
About 244.9 GB at Q4_K_M with an 8K context window. That is the 242.8 GB file, plus the KV cache and roughly 0.5 GB of runtime overhead. Smaller quants need less and longer context needs more.
Can I run Llama-4-Maverick-17B-128E-Instruct on an 8 GB GPU?
Not at an 8K context. The smallest file needs about 121.5 GB, so you would need a larger GPU or CPU offloading.
What is the file size of Llama-4-Maverick-17B-128E-Instruct GGUF?
It depends on the quant: IQ1_S is 119.4 GB, IQ1_M is 126.2 GB, IQ2_XXS is 134.1 GB, Q2_K is 145.6 GB, Q2_K_L is 145.9 GB, Q2_K_XL is 151.6 GB and 20 more.
What context length does Llama-4-Maverick-17B-128E-Instruct support?
The GGUF header lists a trained context length of 1,048,576 tokens (1024K). The KV cache grows with the context you use: about 1.6 GB at 8K and 6.4 GB at 32K.
Where can I download Llama-4-Maverick-17B-128E-Instruct GGUF?
The files are on Hugging Face in the unsloth/Llama-4-Maverick-17B-128E-Instruct-GGUF-UD repository. The download buttons on this page open the file on Hugging Face. ggufmodels.com does not host any model files.
What license does Llama-4-Maverick-17B-128E-Instruct use?
Hugging Face lists the license as other. Licenses can differ between the quantized files and the original model, so check the repository and the base model before commercial use.
Can I run Llama‑4 Maverick on 8 GB VRAM?
Use a 4‑bit quantization such as Q4_K_S or Q4_K_M; memory usage will be around 230 GB, which is not feasible on 8 GB VRAM.
Does the model support image input?
Yes, the model is multimodal and can process image inputs.
What is the maximum context length?
The model supports up to 1 million tokens of context.