Apple M3 Max (64GB) llama.cpp: Llama-3.3-70B Q4_K_M decode @32k ctx
6.13 tok/s single-stream @ 32,170 ctx
Fields
id
benchmarks/comm-apple-m3-max-llama33-70b-q4km-decode-32kctxMetric
decode_tok_s
Value
6.13
Unit
tok/s single-stream @ 32,170 ctx
Model
Llama-3.3-70B Q4_K_M
Methodology
Silicon Score CSV, same source row as the 258-ctx entry: decode 6.13 t/s at 32,170-token context, pp 50.32 t/s.
Measured by
third_party