Apple M2 Max (30c, 32GB) llamafile: Llama-3.1-8B Q4_K_M decode
31.19 tok/s avg generation (batch-1); prefill 325.0 t/s
Fields
id
benchmarks/localscore-m2-max-30c-32gb-llama31-8b-q4km-decodeMetric
decode_tok_s
Value
31.19
Unit
tok/s avg generation (batch-1); prefill 325.0 t/s
Model
llama-3-1-8b-instruct Q4_K_M
Methodology
localscore.ai accelerator/1079 measured row via siliconscore CSV.
Measured by
third_party
notes
30-core/32GB variant; same silicon SKU, lower binned GPU. Useful for the used-MacBook market where 30c/32GB is the common config.
Notes
30-core/32GB variant; same silicon SKU, lower binned GPU. Useful for the used-MacBook market where 30c/32GB is the common config.