[16:26:10] longeval: T=32768 prompt_len=4096 max_new=24 n=100 greedy=1 gpus=0,1
[16:26:10] haystack: /root/Adhi/BluTrain/dist/Context_Parallelism/lendata/haystack/b2val_haystack.bin
[16:26:10] prompt signature changed -- regenerating prompts and clearing state
[16:26:10] generating prompts: niah_single
[16:26:15] fan-out: 2 GPU(s), NP=1 -> 2 concurrent job(s)
[16:26:15] 2 job(s) to run
[16:26:15]   [gpu 0] ext32k / niah_single
[16:26:15]   [gpu 1] l2 / niah_single
[16:28:25]   scoring ext32k / niah_single

=== niah_single  (prompt_len=4096, n=100) ===
  accuracy: 75/100 = 75.0%
    depth      acc   (needle position in context)
     0.00  13/20   65.0%
     0.25  16/20   80.0%
     0.50  13/20   65.0%
     0.75  17/20   85.0%
     1.00  16/20   80.0%

  --- 2 sample generations ---
  [HIT ] expect=8056020
         got= 8056020.\n
  [HIT ] expect=5343902
         got= 5343902.\n

  wrote ext32k_len_4096/scores/ext32k_niah_single.csv
[16:28:25]   scoring l2 / niah_single

=== niah_single  (prompt_len=4096, n=100) ===
  accuracy: 76/100 = 76.0%
    depth      acc   (needle position in context)
     0.00  10/20   50.0%
     0.25  16/20   80.0%
     0.50  12/20   60.0%
     0.75  19/20   95.0%
     1.00  19/20   95.0%

  --- 2 sample generations ---
  [HIT ] expect=8056020
         got= 8056020.\n
  [HIT ] expect=5343902
         got= 5343902.\n

  wrote ext32k_len_4096/scores/l2_niah_single.csv
[16:28:26] building summary

=== per-task accuracy %  (secondary/descriptive: ~10 per depth cell) ===
  task                 ext32k       l2
  niah_single            75.0     76.0

=== PRIMARY: accuracy % pooled by depth across tasks (+-95% Wilson) ===
  depth             ext32k              l2
  0.00          65.0+-19.3      50.0+-20.1
  0.25          80.0+-16.8      80.0+-16.8
  0.50          65.0+-19.3      60.0+-19.7
  0.75          85.0+-15.4      95.0+-11.4
  1.00          80.0+-16.8      95.0+-11.4

  wrote ext32k_len_4096/summary.csv and ext32k_len_4096/depth_summary.csv
[16:28:26] DONE -> ext32k_len_4096/summary.csv