| [16:28:26] longeval: T=32768 prompt_len=8192 max_new=24 n=100 greedy=1 gpus=0,1 |
| [16:28:26] haystack: /root/Adhi/BluTrain/dist/Context_Parallelism/lendata/haystack/b2val_haystack.bin |
| [16:28:26] prompt signature changed -- regenerating prompts and clearing state |
| [16:28:26] generating prompts: niah_single |
| [16:28:26] fan-out: 2 GPU(s), NP=1 -> 2 concurrent job(s) |
| [16:28:26] 2 job(s) to run |
| [16:28:26] [gpu 0] ext32k / niah_single |
| [16:28:26] [gpu 1] l2 / niah_single |
| [16:32:02] scoring ext32k / niah_single |
|
|
| === niah_single (prompt_len=8192, n=100) === |
| accuracy: 60/100 = 60.0% |
| depth acc (needle position in context) |
| 0.00 7/20 35.0% |
| 0.25 13/20 65.0% |
| 0.50 11/20 55.0% |
| 0.75 13/20 65.0% |
| 1.00 16/20 80.0% |
|
|
| --- 2 sample generations --- |
| [HIT ] expect=8056020 |
| got= 8056020.\n |
| [HIT ] expect=5343902 |
| got= 5343902.\n |
|
|
| wrote ext32k_len_8192/scores/ext32k_niah_single.csv |
| [16:32:02] scoring l2 / niah_single |
|
|
| === niah_single (prompt_len=8192, n=100) === |
| accuracy: 36/100 = 36.0% |
| depth acc (needle position in context) |
| 0.00 0/20 0.0% |
| 0.25 1/20 5.0% |
| 0.50 7/20 35.0% |
| 0.75 9/20 45.0% |
| 1.00 19/20 95.0% |
|
|
| --- 2 sample generations --- |
| [MISS] expect=8056020 |
| got= the number of the moon.\n |
| [MISS] expect=5343902 |
| got= the magic number. It is a combination of the magic number and the special number of the plant.\n |
|
|
| wrote ext32k_len_8192/scores/l2_niah_single.csv |
| [16:32:02] building summary |
|
|
| === per-task accuracy % (secondary/descriptive: ~10 per depth cell) === |
| task ext32k l2 |
| niah_single 60.0 36.0 |
|
|
| === PRIMARY: accuracy % pooled by depth across tasks (+-95% Wilson) === |
| depth ext32k l2 |
| 0.00 35.0+-19.3 0.0+- 8.1 |
| 0.25 65.0+-19.3 5.0+-11.4 |
| 0.50 55.0+-20.0 35.0+-19.3 |
| 0.75 65.0+-19.3 45.0+-20.0 |
| 1.00 80.0+-16.8 95.0+-11.4 |
|
|
| wrote ext32k_len_8192/summary.csv and ext32k_len_8192/depth_summary.csv |
| [16:32:02] DONE -> ext32k_len_8192/summary.csv |
|
|
