Can the Apple M1 Max (64 GB) run plamo-3-translate-31b?
plamo-3-translate-31b·31.8B·Apple M1 Max (64 GB)·64 GB unified memory
Yes. The Apple M1 Max (64 GB) runs plamo-3-translate-31b fully in unified memory at Q8_0 (about 38 GB of 48 GB).
- best quant
- Q8_0
- needs
- 38 GB
- usable
- 48 GB
Estimates from the model's size.Try other contexts on the plamo-3-translate-31b page →
Every quantisation of plamo-3-translate-31b on the Apple M1 Max (64 GB)
| Quant | Memory | Fit |
|---|---|---|
| Q4_K_M | 22 GB | fits |
| Q8_0 pick | 38 GB | fits |
| FP16 / BF16 | 71 GB | too big |
Speed is a bandwidth-based estimate for generating text (400 GB/s); prompt processing and long contexts are slower.
Run plamo-3-translate-31b on the Apple M1 Max (64 GB)
Using Q6_K, the best fit for Apple M1 Max (64 GB).
ollama run hf.co/mradermacher/plamo-3-translate-31b-i1-GGUF:Q6_K llama-server -hf mradermacher/plamo-3-translate-31b-i1-GGUF:Q6_Kbrew install llama.cpp ships with Metal. Downloads the GGUF on first run.
lms get mradermacher/plamo-3-translate-31b-i1-GGUF@Q6_KThen load it from the LM Studio app or with `lms load`.
from transformers import pipeline
pipe = pipeline("translation", model="pfnet/plamo-3-translate-31b")
result = pipe("Self-hosting puts you in control of your data.")
print(result)Translation models that fit the Apple M1 Max (64 GB) comfortably
All that fit →| Model | Params | Best quant | Memory |
|---|---|---|---|
| 30.1BA3B | Q8_0 | ≈36 GBFits in memory | |
| 30.4BA3B | Q8_0 | ≈37 GBFits in memory | |
| 9.7B | FP16 / BF16 | ≈22 GBFits in memory | |
| 8B | FP16 / BF16 | ≈18 GBFits in memory | |
| 8B | FP16 / BF16 | ≈18 GBFits in memory | |
| 8.5B | FP16 / BF16 | ≈20 GBFits in memory |
Other GPUs that run plamo-3-translate-31b
Frequently asked questions
- Can the Apple M1 Max (64 GB) run plamo-3-translate-31b?
- Yes. The Apple M1 Max (64 GB) runs plamo-3-translate-31b fully in unified memory at Q8_0 (about 38 GB of 48 GB).
- Which quantisation of plamo-3-translate-31b should I use on the Apple M1 Max (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.