MustaqiLLM — Hardware Requirements & GPU Compatibility
Chat
MustaqiLLM is a 5.2B-parameter open language model from NeuronUz. It supports a context window of up to 4,096 tokens. At BF16 it needs about 10.78 GB of VRAM — see which GPUs and Macs can run it below.
Runs well on a 24 GB GPU
Runs from 12 GB (BF16 · 4K)Best on 24 GB (BF16 · ~61 tok/s · 4K)Macs from 16 GB (BF16 · ~13 tok/s)
Apache 2.0
Reference GPU · RTX 4090 24 GB
Estimates, ±20% · how we estimate
Run MustaqiLLM
One recommended way. Every other method, OS and quant is in the step-by-step guide.
Recommended
vllm serve NeuronUz/MustaqiLLM --max-model-len 4096vLLM repo with its own quantisation
Download: Weights on Hugging Face
Other ways
each opens step-by-stepRun in cloud
Fits on RTX A4000 (16 GB) (5 GB headroom) · BF16
- Generation speed
- ~27 tok/s
- generation speed
- Cost per 1M output tokens
- $1.13
- per 1M output tokens
How Much VRAM Does MustaqiLLM Need?
Select a quantization level; the listed hardware adheres to it.
| Quantization | VRAM | +32K | On RTX 4090 (reference) | |
|---|---|---|---|---|
| 10.8 GB | — | 16-bit · bfloat — Brain floating point 16 — preferred for training | Fits: 4K context · ~61 tok/s |
Selected BF16
Brain floating point 1610.9GBat 4K context
- Weights
- 10.3 GB
- KV cache
- 0.3 GB
- Overhead
- 0.3 GB
Fits RTX 4090 (reference) · 13.1 GB free · ~61 tok/s
≥50 tok/s needs ≥828 GB/s · RTX 5090 ~108, RTX 3090 Ti ~61
Which Hardware Can Run MustaqiLLM?
BF16 · 10.8 GBMustaqiLLM (BF16) requires 10.8 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 15+ GB is recommended. Using the full 4K context window can add up to 0.2 GB, bringing total usage to 10.9 GB. 38 GPUs can run it, including NVIDIA GeForce RTX 5090, NVIDIA GeForce RTX 3090 Ti, NVIDIA GeForce RTX 5080. 150 devices with unified memory can run MustaqiLLM, including Mac Studio (M3 Ultra, 256GB), Mac Studio (M3 Ultra, 512GB), Mac Mini M2 Pro (16 GB).
12 GB
12 GB tier →fits · 1.1 GB free · up to 4K ctx
10 GPUs
RTX 3080 Ti~55Simulate output on RTX 3080 TiRTX 5070~41Simulate output on RTX 5070RTX 4070~30Simulate output on RTX 407020 GB
fits · 9.1 GB free · up to 4K ctx
1 GPU
RX 7900 XT~45Simulate output on RX 7900 XT32–40 GB
fits · 21.1 GB free · up to 4K ctx
1 GPU · 3 Mac classes · 4 device families
RTX 5090~108Simulate output on RTX 5090M5 Max36 GB~30Simulate output on Mac Studio M5 Max (36 GB)M532 GB~10Simulate output on MacBook Air 13" M5 (32 GB)Orin~12Simulate output on NVIDIA Jetson AGX Orin 32GBSnapdragon X Elite (X1E-80-100)~8Simulate output on Snapdragon X Elite Copilot+ PC
By VRAM tier · typical GPUs and Macs at BF16
Similar models
4| Model | Score | VRAM | tok/s | Why it is similar |
|---|---|---|---|---|
| Gemma 4 E2BGoogle | — | 3.5 GB | ~191 | same size |
| Molmo2 4BAllen AI | — | 4.7 GB | ~193 | same size |
| Gemma 3n E2B ITGoogle | — | 5.9 GB | ~183 | same size |
| Qwen2.5 Omni 3BAlibaba | — | 6 GB | ~180 | same size |
Specifications
- Publisher
- NeuronUz
- Parameters
- 5.2B
- Architecture
- NeuronLMForCausalLM
- Context Length
- 4,096 tokens
- Vocabulary Size
- 48,000
- Release Date
- 2026-08-13
- License
- Apache 2.0
Frequently Asked Questions
How much VRAM does MustaqiLLM need?
MustaqiLLM requires 10.8 GB of VRAM at BF16.
VRAM = Weights + KV Cache + Overhead
Weights = 5.2B × 16 bits ÷ 8 = 10.3 GB
KV Cache + Overhead ≈ 0.5 GB (at 2K context + ~0.3 GB framework)
Fit ratings and hardware model lists check this model with room for a 16K-token context, which needs a little more memory.
KV Cache + Overhead ≈ 0.6 GB (at full 4K context)
VRAM usage by quantization
Can I run MustaqiLLM on a Mac?
Yes — Mac Mini M2 Pro (16 GB) and 133 other Macs can run MustaqiLLM. Apple Silicon uses unified memory, so the model shares RAM with the system. At BF16 you need at least 10.8 GB of usable unified memory (RAM minus macOS overhead).
Can I run MustaqiLLM locally?
Yes — MustaqiLLM can run locally on consumer hardware. Popular tools include Ollama, LM Studio, and llama.cpp.
How fast is MustaqiLLM?
At BF16, MustaqiLLM can reach ~445 tok/s on AMD Instinct MI350X. On NVIDIA GeForce RTX 4090: ~61 tok/s. Speed depends mainly on GPU memory bandwidth. Real-world results typically within ±20%.
tok/s = (bandwidth GB/s ÷ model GB) × efficiency
Example: NVIDIA B200 → 8000 ÷ 10.78 × 0.65 = ~482 tok/s
Estimated speed at BF16 (10.8 GB)
Real-world results typically within ±20%. Speed depends on batch size, quantization kernel, and software stack.
What's the download size of MustaqiLLM?
At BF16, the download is about 10.33 GB.
Which GPUs can run MustaqiLLM?
38 consumer GPUs can run MustaqiLLM at BF16. Top options include AMD Radeon RX 7900 XT, AMD Radeon RX 7900 XTX, NVIDIA GeForce RTX 3090, AMD Radeon RX 6700 XT. 8 GPUs have plenty of headroom for comfortable inference.
Which devices can run MustaqiLLM?
150 devices with unified memory can run MustaqiLLM at BF16 (10.8 GB), including AMD Ryzen AI 9 HX 370 (Strix Point) Laptop, ASUS Ascent GX10, Beelink GTR9 Pro (Ryzen AI Max+ 395, 128 GB), Framework Desktop (Ryzen AI Max+ 395, 128 GB). Apple Silicon Macs use unified memory shared between CPU and GPU, making them well-suited for local LLM inference.