M3 Max Llama 2 7B Q4_0 decode (llama.cpp, siliconscore ref)
65.85 tok/s (single-request decode, per-request)
Fields
id
benchmarks/m3max-llama-2-7b-q4-0-llama-cpp-ctx512-tgMetric
decode_tok_s
Value
65.85
Unit
tok/s (single-request decode, per-request)
Model
Llama 2 7B
Methodology
Silicon Score benchmarks.csv reference run. M3 Max (40-core GPU, 48 GB): Llama 2 7B Q4_0 via llama.cpp. context_tokens=512. Decode tok/s single-request.
Measured by
third_party
Reproducible
yes
Thermal state
unknown
notes
siliconscore reference run; source_url=https://github.com/ggml-org/llama.cpp/discussions/4167. Chip variant 'M3 Max (40-core GPU, 48 GB)'.
Notes
siliconscore reference run; source_url=https://github.com/ggml-org/llama.cpp/discussions/4167. Chip variant 'M3 Max (40-core GPU, 48 GB)'.