File size: 2,118 Bytes
a2d6f5f | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 | [16:42:35] longeval: T=16384 prompt_len=8192 max_new=24 n=100 greedy=1 gpus=0,1
[16:42:35] haystack: /root/Adhi/BluTrain/dist/Context_Parallelism/lendata/haystack/b2val_haystack.bin
[16:42:35] fan-out: 2 GPU(s), NP=1 -> 4 concurrent job(s)
[16:42:35] armA_c/niah_single already generated
[16:42:35] l1/niah_single already generated
[16:42:35] l2/niah_single already generated
[16:42:35] l3/niah_single already generated
[16:42:35] l3long/niah_single already generated
[16:42:35] 1 job(s) to run
[16:42:35] [gpu 0] l3long2k / niah_single
[16:44:43] scoring l3long2k / niah_single
=== niah_single (prompt_len=8192, n=100) ===
accuracy: 54/100 = 54.0%
depth acc (needle position in context)
0.00 6/20 30.0%
0.25 12/20 60.0%
0.50 10/20 50.0%
0.75 9/20 45.0%
1.00 17/20 85.0%
--- 2 sample generations ---
[HIT ] expect=8056020
got= 8056020.\n
[HIT ] expect=5343902
got= 5343902.\n
wrote decide_len_8192/scores/l3long2k_niah_single.csv
[16:44:43] building summary
=== per-task accuracy % (secondary/descriptive: ~10 per depth cell) ===
task armA_c l1 l2 l3 l3long l3long2k
niah_single 53.0 48.0 62.0 64.0 55.0 54.0
=== PRIMARY: accuracy % pooled by depth across tasks (+-95% Wilson) ===
depth armA_c l1 l2 l3 l3long l3long2k
0.00 15.0+-15.4 5.0+-11.4 55.0+-20.0 60.0+-19.7 30.0+-18.7 30.0+-18.7
0.25 45.0+-20.0 35.0+-19.3 55.0+-20.0 60.0+-19.7 60.0+-19.7 60.0+-19.7
0.50 55.0+-20.0 40.0+-19.7 50.0+-20.1 55.0+-20.0 40.0+-19.7 50.0+-20.1
0.75 55.0+-20.0 70.0+-18.7 55.0+-20.0 55.0+-20.0 50.0+-20.1 45.0+-20.0
1.00 95.0+-11.4 90.0+-13.7 95.0+-11.4 90.0+-13.7 95.0+-11.4 85.0+-15.4
wrote decide_len_8192/summary.csv and decide_len_8192/depth_summary.csv
[16:44:43] DONE -> decide_len_8192/summary.csv
|