selfhost.directory

The index of self-hostable projects — ranked, categorized and served with live updates.
Own your data and stop giving it to big tech.

inclusionAI logo

Can the Apple M4 Pro (48 GB) run ZwZ2-27B?

ZwZ2-27B·27.4B·Apple M4 Pro (48 GB)·48 GB unified memory

Saved in this browser, used on every model page.
Yes

Yes. The Apple M4 Pro (48 GB) runs ZwZ2-27B fully in unified memory at Q8_0 (about 30 GB of 36 GB), at roughly 5.2 tokens/s.

best quant
Q8_0
needs
30 GB
usable
36 GB
est. speed
~5.2 tok/s

Estimates for an 8K context; offload assumes 32 GB of system RAM.Try other contexts on the ZwZ2-27B page →

Every quantisation of ZwZ2-27B on the Apple M4 Pro (48 GB)

QuantQualityMemoryFitSpeed
Q2_K Heavy loss11 GBfits~17 tok/s
Q3_K_M Noticeable loss16 GBfits~11 tok/s
Q4_K_M Small loss19 GBfits~9.1 tok/s
Q5_K_M Near-lossless21 GBfits~7.7 tok/s
Q6_K Near-lossless24 GBfits~6.7 tok/s
Q8_0 pickLossless30 GBfits~5.2 tok/s
FP16 / BF16 Lossless55 GBtoo big—

Speed is a bandwidth-based estimate for generating text (273 GB/s); prompt processing and long contexts are slower.

Measured speed on the Apple M4 Pro (48 GB)

No community reports for this pair yet.

Report your own speed →

Run ZwZ2-27B on the Apple M4 Pro (48 GB)

Using Q8_0, the best fit for Apple M4 Pro (48 GB).

llama.cpp · OpenAI-compatible server with a web UIdocs
llama-server -hf inclusionAI/ZwZ2-27B-GGUF:Q8_0

brew install llama.cpp ships with Metal. Downloads the GGUF on first run.

Transformers · Ask about an imagedocs
from transformers import pipeline

pipe = pipeline("image-text-to-text", model="inclusionAI/ZwZ2-27B", dtype="auto", device_map="auto")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/bee.jpg"},
            {"type": "text", "text": "Describe this image."},
        ],
    }
]
out = pipe(text=messages, max_new_tokens=256)
print(out[0]["generated_text"][-1]["content"])

Vision models that fit the Apple M4 Pro (48 GB) comfortably

All that fit →
ModelParamsBest quantMemorySpeedLicence
27.8BQ8_0≈31 GBFits in memory~5.1 tok/sCommercial OK
27.8BQ8_0≈30 GBFits in memory~5.1 tok/sCommercial OK
25.8BA4BQ8_0≈29 GBFits in memory~35 tok/sCommercial OK
31.3BQ6_K≈33 GBFits in memory~5.8 tok/sCommercial OK
27.4BQ8_0≈32 GBFits in memory~5.2 tok/sConditions
27.8BQ8_0≈30 GBFits in memory~5.1 tok/sCommercial OK

Other GPUs that run ZwZ2-27B

Frequently asked questions

Can the Apple M4 Pro (48 GB) run ZwZ2-27B?
Yes. The Apple M4 Pro (48 GB) runs ZwZ2-27B fully in unified memory at Q8_0 (about 30 GB of 36 GB), at roughly 5.2 tokens/s.
How fast does ZwZ2-27B run on the Apple M4 Pro (48 GB)?
An estimate from memory bandwidth gives about 5.2 tokens/s at Q8_0.
Which quantisation of ZwZ2-27B should I use on the Apple M4 Pro (48 GB)?
Q8_0: the largest one that fits. Expected quality: lossless compared with the full-precision weights.

Memory and speed are estimates from the model's size and the card's memory bandwidth; real numbers depend on the runtime, drivers and context length. Community figures are medians of user reports.