GGUF
conversational
Experiment_1 / opt /prof_decode.txt
HelloSun's picture
Upload opt/prof_decode.txt with huggingface_hub
73fbc37 verified
Raw History Blame Contribute Delete
1.13 kB
"avg_ts": 32.210311
PROF total compute 3571.2 ms, barrier-wait 353.2 ms
PROF ADD n= 26316 compute 34.30 ms barrier 18.01 ms (1.3 us/op compute, 0.7 us/op barrier)
PROF RMS_NORM n= 8901 compute 0.35 ms barrier 3.25 ms (0.0 us/op compute, 0.4 us/op barrier)
PROF GET_ROWS n= 387 compute 0.61 ms barrier 0.19 ms (1.6 us/op compute, 0.5 us/op barrier)
PROF SET_ROWS n= 8772 compute 11.75 ms barrier 5.78 ms (1.3 us/op compute, 0.7 us/op barrier)
PROF ROPE n= 8772 compute 12.65 ms barrier 22.85 ms (1.4 us/op compute, 2.6 us/op barrier)
PROF FLASH_ATTN_EXT n= 4386 compute 82.19 ms barrier 13.96 ms (18.7 us/op compute, 3.2 us/op barrier)
PROF GLU n= 4386 compute 40.82 ms barrier 3.01 ms (9.3 us/op compute, 0.7 us/op barrier)
PROF MUL_MAT q4_K n= 26316 compute 2470.86 ms barrier 236.91 ms (93.9 us/op compute, 9.0 us/op barrier)
PROF MUL_MAT q6_K n= 4515 compute 917.64 ms barrier 49.19 ms (203.2 us/op compute, 10.9 us/op barrier)
PROF_DONE