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.

InternScience logo

Can the Apple M4 Max 40-core (64 GB) run MLExpert?

MLExpert·34.7B (3.5B active)·Apple M4 Max 40-core (64 GB)·64 GB unified memory

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

Yes. The Apple M4 Max 40-core (64 GB) runs MLExpert fully in unified memory at Q8_0 (about 37 GB of 48 GB), at roughly 82 tokens/s.

best quant
Q8_0
needs
37 GB
usable
48 GB
est. speed
~82 tok/s

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

Every quantisation of MLExpert on the Apple M4 Max 40-core (64 GB)

QuantQualityMemoryFitSpeed
Q2_K Heavy loss15 GBfits~264 tok/s
Q3_K_M Noticeable loss18 GBfits~178 tok/s
Q4_K_M Small loss23 GBfits~143 tok/s
Q5_K_M Near-lossless28 GBfits~122 tok/s
Q6_K Near-lossless31 GBfits~106 tok/s
Q8_0 pickLossless37 GBfits~82 tok/s
FP16 / BF16 Lossless69 GBtoo big—

Speed is a bandwidth-based estimate for generating text (546 GB/s); prompt processing and long contexts are slower.

Measured speed on the Apple M4 Max 40-core (64 GB)

No community reports for this pair yet.

Report your own speed →

Run MLExpert on the Apple M4 Max 40-core (64 GB)

Using Q8_0, the best fit for Apple M4 Max 40-core (64 GB).

Ollama · Q8_0 straight from Hugging Facedocs
ollama run hf.co/bartowski/InternScience_MLExpert-GGUF:Q8_0
llama.cpp · OpenAI-compatible server with a web UIdocs
llama-server -hf bartowski/InternScience_MLExpert-GGUF:Q8_0

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

LM Studio · Download with the lms CLIdocs
lms get bartowski/InternScience_MLExpert-GGUF@Q8_0

Then load it from the LM Studio app or with `lms load`.

Transformers · Chat with the pipeline APIdocs
from transformers import pipeline

pipe = pipeline("text-generation", model="InternScience/MLExpert", dtype="auto", device_map="auto")
messages = [{"role": "user", "content": "Explain self-hosting in one paragraph."}]
out = pipe(messages, max_new_tokens=512)
print(out[0]["generated_text"][-1]["content"])

Chat models that fit the Apple M4 Max 40-core (64 GB) comfortably

All that fit →
ModelParamsBest quantMemorySpeedLicence
36BA3BQ8_0≈38 GBFits in memory~94 tok/sCommercial OK
31.2BA4.6BQ8_0≈39 GBFits in memory~61 tok/sCommercial OK
32.8BQ8_0≈37 GBFits in memory~8.6 tok/sCommercial OK
RecursiveSI logoflintRecursiveSI
39.8BA1.3BQ8_0≈43 GBFits in memory~226 tok/sCommercial OK
20.9BA4.2BQ8_0≈13 GBFits in memory~67 tok/sCommercial OK
30.5BA3BQ8_0≈33 GBFits in memory~94 tok/sCommercial OK

Other GPUs that run MLExpert

Frequently asked questions

Can the Apple M4 Max 40-core (64 GB) run MLExpert?
Yes. The Apple M4 Max 40-core (64 GB) runs MLExpert fully in unified memory at Q8_0 (about 37 GB of 48 GB), at roughly 82 tokens/s.
How fast does MLExpert run on the Apple M4 Max 40-core (64 GB)?
An estimate from memory bandwidth gives about 82 tokens/s at Q8_0.
Which quantisation of MLExpert should I use on the Apple M4 Max 40-core (64 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.