llama.cpp Vulkan Radeon Pro W5700 llama-2-7b Q4_0 tg128 FA
71.3 tok/s (tg128)
Fields
id
benchmarks/llamacpp-vulkan-radeon-pro-w5700-llama27b-q40-fa-tg128Metric
decode_tok_s
Value
71.3
Unit
tok/s (tg128)
Model
llama-2-7b.Q4_0.gguf (ggml-org, 3.56 GiB, 6.74B params)
Methodology
COMMUNITY-SCOREBOARD (llama.cpp discussion #10879, maintainer-curated Vulkan backend). llama-bench llama-2-7b.Q4_0.gguf (ggml-org, 3.56 GiB, 6.74B params), pp512 and tg128, batch 1, full GPU offload, Vulkan. Per-request batch-1, not aggregate across concurrent requests. Flash attention on (-fa 1). Build 23bc779. Chip AMD Radeon Pro W5700. Printed mean 71.30 +/- 0.24. Generating 128 tokens, batch 1.
Measured by
third_party
Reproducible
yes
notes
Flash-attention scoreboard row. Decode is the faster of the two rows (71.30 vs 68.55).
Notes
Flash-attention scoreboard row. Decode is the faster of the two rows (71.30 vs 68.55).