vrr/prima.cpp
2
0
Fork 0
mirror of https://github.com/Lizonghang/prima.cpp.git synced 2025-09-06 11:59:02 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

Fixed the alignment display of device performance

Browse source
This commit is contained in:
leeetao  2025-03-07 07:46:30 +00:00
parent 54c4c1c26e
commit 6a416534c8
2 changed files with 111 additions and 111 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

36
common/profiler.cpp
View file

@ -1714,43 +1714,43 @@ void device_print_props(struct device_info * dev_info_set, int n, struct llama_m
} }
LOG_INF("\n"); LOG_INF("\n");
LOG_INF("| CPU flops (Q2K x F32, GFLOPS)"); LOG_INF("| CPU flops (Q2K x F32, GFLOPS) ");
for (int i = 0; i < n; ++i) { for (int i = 0; i < n; ++i) {
LOG_INF("| %-10.1f ", dev_info_set[i].cpu_props.flops_q2k_f32); LOG_INF("| %-10.1f ", dev_info_set[i].cpu_props.flops_q2k_f32);
} }
LOG_INF("\n"); LOG_INF("\n");
LOG_INF("| CPU flops (Q4K x F32, GFLOPS)"); LOG_INF("| CPU flops (Q4K x F32, GFLOPS) ");
for (int i = 0; i < n; ++i) { for (int i = 0; i < n; ++i) {
LOG_INF("| %-10.1f ", dev_info_set[i].cpu_props.flops_q4k_f32); LOG_INF("| %-10.1f ", dev_info_set[i].cpu_props.flops_q4k_f32);
} }
LOG_INF("\n"); LOG_INF("\n");
LOG_INF("| CPU flops (Q5K x F32, GFLOPS)"); LOG_INF("| CPU flops (Q5K x F32, GFLOPS) ");
for (int i = 0; i < n; ++i) { for (int i = 0; i < n; ++i) {
LOG_INF("| %-10.1f ", dev_info_set[i].cpu_props.flops_q5k_f32); LOG_INF("| %-10.1f ", dev_info_set[i].cpu_props.flops_q5k_f32);
} }
LOG_INF("\n"); LOG_INF("\n");
LOG_INF("| CPU flops (Q6K x F32, GFLOPS)"); LOG_INF("| CPU flops (Q6K x F32, GFLOPS) ");
for (int i = 0; i < n; ++i) { for (int i = 0; i < n; ++i) {
LOG_INF("| %-10.1f ", dev_info_set[i].cpu_props.flops_q6k_f32); LOG_INF("| %-10.1f ", dev_info_set[i].cpu_props.flops_q6k_f32);
} }
LOG_INF("\n"); LOG_INF("\n");
LOG_INF("| CPU flops (Q50 x F32, GFLOPS)"); LOG_INF("| CPU flops (Q50 x F32, GFLOPS) ");
for (int i = 0; i < n; ++i) { for (int i = 0; i < n; ++i) {
LOG_INF("| %-10.1f ", dev_info_set[i].cpu_props.flops_q50_f32); LOG_INF("| %-10.1f ", dev_info_set[i].cpu_props.flops_q50_f32);
} }
LOG_INF("\n"); LOG_INF("\n");
LOG_INF("| CPU flops (Q80 x F32, GFLOPS)"); LOG_INF("| CPU flops (Q80 x F32, GFLOPS) ");
for (int i = 0; i < n; ++i) { for (int i = 0; i < n; ++i) {
LOG_INF("| %-10.1f ", dev_info_set[i].cpu_props.flops_q80_f32); LOG_INF("| %-10.1f ", dev_info_set[i].cpu_props.flops_q80_f32);
} }
LOG_INF("\n"); LOG_INF("\n");
LOG_INF("| CPU flops (IQ1S x F32, GFLOPS)"); LOG_INF("| CPU flops (IQ1S x F32, GFLOPS) ");
for (int i = 0; i < n; ++i) { for (int i = 0; i < n; ++i) {
LOG_INF("| %-10.1f ", dev_info_set[i].cpu_props.flops_iq1s_f32); LOG_INF("| %-10.1f ", dev_info_set[i].cpu_props.flops_iq1s_f32);
} }
@ -1900,61 +1900,61 @@ void device_print_props(struct device_info * dev_info_set, int n, struct llama_m
} }
LOG_INF("\n"); LOG_INF("\n");
LOG_INF("| Metal KVCache Copy Time(ms/l)"); LOG_INF("| Metal KVCache Copy Time(ms/l) ");
for (int i = 0; i < n; ++i) { for (int i = 0; i < n; ++i) {
LOG_INF("| %-10.2f ", dev_info_set[i].gpu_props.metal_mem_cpy_delay); LOG_INF("| %-10.2f ", dev_info_set[i].gpu_props.metal_mem_cpy_delay);
} }
LOG_INF("\n"); LOG_INF("\n");
LOG_INF("| Metal flops (F32xF32, GFLOPS)"); LOG_INF("| Metal flops (F32xF32, GFLOPS) ");
for (int i = 0; i < n; ++i) { for (int i = 0; i < n; ++i) {
LOG_INF("| %-10.1f ", dev_info_set[i].gpu_props.metal_flops_f32_f32); LOG_INF("| %-10.1f ", dev_info_set[i].gpu_props.metal_flops_f32_f32);
} }
LOG_INF("\n"); LOG_INF("\n");
LOG_INF("| Metal flops (F16xF32, GFLOPS)"); LOG_INF("| Metal flops (F16xF32, GFLOPS) ");
for (int i = 0; i < n; ++i) { for (int i = 0; i < n; ++i) {
LOG_INF("| %-10.1f ", dev_info_set[i].gpu_props.metal_flops_f16_f32); LOG_INF("| %-10.1f ", dev_info_set[i].gpu_props.metal_flops_f16_f32);
} }
LOG_INF("\n"); LOG_INF("\n");
LOG_INF("| Metal flops (Q2KxF32, GFLOPS)"); LOG_INF("| Metal flops (Q2KxF32, GFLOPS) ");
for (int i = 0; i < n; ++i) { for (int i = 0; i < n; ++i) {
LOG_INF("| %-10.1f ", dev_info_set[i].gpu_props.metal_flops_q2k_f32); LOG_INF("| %-10.1f ", dev_info_set[i].gpu_props.metal_flops_q2k_f32);
} }
LOG_INF("\n"); LOG_INF("\n");
LOG_INF("| Metal flops (Q4KxF32, GFLOPS)"); LOG_INF("| Metal flops (Q4KxF32, GFLOPS) ");
for (int i = 0; i < n; ++i) { for (int i = 0; i < n; ++i) {
LOG_INF("| %-10.1f ", dev_info_set[i].gpu_props.metal_flops_q4k_f32); LOG_INF("| %-10.1f ", dev_info_set[i].gpu_props.metal_flops_q4k_f32);
} }
LOG_INF("\n"); LOG_INF("\n");
LOG_INF("| Metal flops (Q5KxF32, GFLOPS)"); LOG_INF("| Metal flops (Q5KxF32, GFLOPS) ");
for (int i = 0; i < n; ++i) { for (int i = 0; i < n; ++i) {
LOG_INF("| %-10.1f ", dev_info_set[i].gpu_props.metal_flops_q5k_f32); LOG_INF("| %-10.1f ", dev_info_set[i].gpu_props.metal_flops_q5k_f32);
} }
LOG_INF("\n"); LOG_INF("\n");
LOG_INF("| Metal flops (Q6KxF32, GFLOPS)"); LOG_INF("| Metal flops (Q6KxF32, GFLOPS) ");
for (int i = 0; i < n; ++i) { for (int i = 0; i < n; ++i) {
LOG_INF("| %-10.1f ", dev_info_set[i].gpu_props.metal_flops_q6k_f32); LOG_INF("| %-10.1f ", dev_info_set[i].gpu_props.metal_flops_q6k_f32);
} }
LOG_INF("\n"); LOG_INF("\n");
LOG_INF("| Metal flops (Q50xF32, GFLOPS)"); LOG_INF("| Metal flops (Q50xF32, GFLOPS) ");
for (int i = 0; i < n; ++i) { for (int i = 0; i < n; ++i) {
LOG_INF("| %-10.1f ", dev_info_set[i].gpu_props.metal_flops_q50_f32); LOG_INF("| %-10.1f ", dev_info_set[i].gpu_props.metal_flops_q50_f32);
} }
LOG_INF("\n"); LOG_INF("\n");
LOG_INF("| Metal flops (Q80xF32, GFLOPS)"); LOG_INF("| Metal flops (Q80xF32, GFLOPS) ");
for (int i = 0; i < n; ++i) { for (int i = 0; i < n; ++i) {
LOG_INF("| %-10.1f ", dev_info_set[i].gpu_props.metal_flops_q80_f32); LOG_INF("| %-10.1f ", dev_info_set[i].gpu_props.metal_flops_q80_f32);
} }
LOG_INF("\n"); LOG_INF("\n");
LOG_INF("| Metal flops (IQ1SxF32, GFLOPS)"); LOG_INF("| Metal flops (IQ1SxF32, GFLOPS) ");
for (int i = 0; i < n; ++i) { for (int i = 0; i < n; ++i) {
LOG_INF("| %-10.1f ", dev_info_set[i].gpu_props.metal_flops_iq1s_f32); LOG_INF("| %-10.1f ", dev_info_set[i].gpu_props.metal_flops_iq1s_f32);
} }
@ -1972,7 +1972,7 @@ void device_print_props(struct device_info * dev_info_set, int n, struct llama_m
} }
LOG_INF("\n"); LOG_INF("\n");
LOG_INF("| CUDA KVCache Copy Time (ms/l)"); LOG_INF("| CUDA KVCache Copy Time (ms/l) ");
for (int i = 0; i < n; ++i) { for (int i = 0; i < n; ++i) {
LOG_INF("| %-10.2f ", dev_info_set[i].gpu_props.cuda_mem_cpy_delay); LOG_INF("| %-10.2f ", dev_info_set[i].gpu_props.cuda_mem_cpy_delay);
} }