LLM Memory Calculator for Mac
Effortlessly estimate how much unified memory your Mac (M1, M2, M3, or M4 series) will use when running Hugging Face models in MLX-based workflows.
Model configuration
- Number of layers
- —
- KV heads
- —
- Heads dimension
- —
- Precision
- —
Estimated unified memory
—
- Model weights
- —
- KV-Cache
- —
- Activation Cache
- —
- Inference overhead
- —
Enter a model ID to see the estimate
Paste any public Hugging Face repository that ships a config.json. Weights are read from its .safetensors files when available.
About this estimate
Just enter the model ID and context length and get immediate insights into memory demand, including activations and KV-cache, tailored to Apple Silicon's unified memory architecture.
Weights and KV cache also apply to Torch and Transformers workflows, but activations and overheads are calibrated on mlx-lm measurements: with a full-sequence prefill, expect a higher peak. Optimize your model deployment, avoid memory issues, and make the most of your Mac's AI potential.
Not working with a model? Missing feature? Let us know: support@insightkeeper.ai