TokenBender commited on
Commit
418f34e
·
verified ·
1 Parent(s): a8ada5d

Add legacy pod log artifacts

Browse files
This view is limited to 50 files because it contains too many changes.   See raw diff
Files changed (50) hide show
  1. .gitattributes +86 -0
  2. circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/data/qwen25_math_1p5b_rank_ladder_r16_correct_addition.json +0 -0
  3. circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/data/qwen25_math_1p5b_rank_ladder_r1_correct_addition.json +0 -0
  4. circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/data/qwen25_math_1p5b_rank_ladder_r2_correct_addition.json +0 -0
  5. circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/data/qwen25_math_1p5b_rank_ladder_r4_correct_addition.json +0 -0
  6. circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/data/qwen25_math_1p5b_rank_ladder_r8_correct_addition.json +0 -0
  7. circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_16_correct_dataset.log +26 -0
  8. circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_16_direct_greedy.log +179 -0
  9. circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_16_direct_rank_shard0.log +68 -0
  10. circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_16_direct_rank_shard1.log +68 -0
  11. circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_16_direct_rank_shard2.log +68 -0
  12. circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_16_direct_rank_shard3.log +68 -0
  13. circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_16_fresh_build.log +7 -0
  14. circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_16_fresh_target90_eval.log +6 -0
  15. circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_16_fresh_target90_generation.log +4 -0
  16. circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_16_merge.log +9 -0
  17. circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_16_position_trace.log +64 -0
  18. circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_16_prep_stage.log +5 -0
  19. circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_16_target90_search.log +271 -0
  20. circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_16_train.log +9 -0
  21. circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_1_correct_dataset.log +26 -0
  22. circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_1_direct_rank_shard0.log +68 -0
  23. circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_1_direct_rank_shard1.log +68 -0
  24. circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_1_direct_rank_shard2.log +68 -0
  25. circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_1_direct_rank_shard3.log +68 -0
  26. circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_1_fresh_build.log +7 -0
  27. circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_1_fresh_target90_eval.log +6 -0
  28. circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_1_fresh_target90_generation.log +4 -0
  29. circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_1_merge.log +9 -0
  30. circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_1_position_trace.log +64 -0
  31. circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_1_prep_stage.log +5 -0
  32. circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_1_target90_search.log +370 -0
  33. circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_1_train.log +9 -0
  34. circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_2_correct_dataset.log +26 -0
  35. circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_2_direct_rank_shard0.log +68 -0
  36. circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_2_direct_rank_shard1.log +68 -0
  37. circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_2_direct_rank_shard2.log +68 -0
  38. circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_2_direct_rank_shard3.log +68 -0
  39. circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_2_fresh_build.log +7 -0
  40. circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_2_fresh_target90_eval.log +6 -0
  41. circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_2_fresh_target90_generation.log +4 -0
  42. circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_2_merge.log +9 -0
  43. circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_2_position_trace.log +64 -0
  44. circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_2_prep_stage.log +5 -0
  45. circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_2_target90_search.log +377 -0
  46. circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_2_train.log +9 -0
  47. circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_4_correct_dataset.log +26 -0
  48. circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_4_direct_rank_shard0.log +68 -0
  49. circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_4_direct_rank_shard1.log +68 -0
  50. circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_4_direct_rank_shard2.log +68 -0
.gitattributes CHANGED
@@ -36,3 +36,89 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
36
  adapters/adapter-a/tokenizer.json filter=lfs diff=lfs merge=lfs -text
37
  checkpoints/checkpoint-a/tokenizer.json filter=lfs diff=lfs merge=lfs -text
38
  checkpoints/checkpoint-b/tokenizer.json filter=lfs diff=lfs merge=lfs -text
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
36
  adapters/adapter-a/tokenizer.json filter=lfs diff=lfs merge=lfs -text
37
  checkpoints/checkpoint-a/tokenizer.json filter=lfs diff=lfs merge=lfs -text
38
  checkpoints/checkpoint-b/tokenizer.json filter=lfs diff=lfs merge=lfs -text
39
+ circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_1/lora_r1_beta005/adapter/tokenizer.json filter=lfs diff=lfs merge=lfs -text
40
+ circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_16/lora_r16_beta005/adapter/tokenizer.json filter=lfs diff=lfs merge=lfs -text
41
+ circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_2/lora_r2_beta005/adapter/tokenizer.json filter=lfs diff=lfs merge=lfs -text
42
+ circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_4/lora_r4_beta005/adapter/tokenizer.json filter=lfs diff=lfs merge=lfs -text
43
+ circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_8/lora_r8_beta005/adapter/tokenizer.json filter=lfs diff=lfs merge=lfs -text
44
+ circuit-shotting/artifacts/pod_logs/gentle-comet-ef/full/qwen25_math_1p5b_2digit_full_sft_issue10/full_sft_ce_lr1e5/train/model/tokenizer.json filter=lfs diff=lfs merge=lfs -text
45
+ circuit-shotting/artifacts/pod_logs/gentle-comet-ef/full/qwen25_math_1p5b_2digit_full_sft_issue10/full_sft_kl_beta005_lr1e5/train/model/tokenizer.json filter=lfs diff=lfs merge=lfs -text
46
+ circuit-shotting/artifacts/pod_logs/steed-medium/full/qwen25_math_1p5b_2digit_lora_kl_v1/train/adapter/tokenizer.json filter=lfs diff=lfs merge=lfs -text
47
+ circuit-shotting/artifacts/pod_logs/steed-medium/full/qwen25_math_1p5b_2digit_max_recovery_v1/lora_r32_beta005/adapter/tokenizer.json filter=lfs diff=lfs merge=lfs -text
48
+ circuit-shotting/artifacts/pod_logs/steed-medium/full/qwen25_math_1p5b_3digit_adacs_split_v1/results/adacs_split_scout_v1/carry_ones_to_tens/answer_len/3/pairs_carry_ones_to_tens.json filter=lfs diff=lfs merge=lfs -text
49
+ circuit-shotting/artifacts/pod_logs/steed-medium/full/qwen25_math_1p5b_3digit_adacs_split_v1/results/adacs_split_scout_v1/carry_ones_to_tens/answer_len/4/pairs_carry_ones_to_tens.json filter=lfs diff=lfs merge=lfs -text
50
+ circuit-shotting/artifacts/pod_logs/steed-medium/full/qwen25_math_1p5b_3digit_adacs_split_v1/results/adacs_split_scout_v1/carry_ones_to_tens/d0_ones/0/pairs_carry_ones_to_tens.json filter=lfs diff=lfs merge=lfs -text
51
+ circuit-shotting/artifacts/pod_logs/steed-medium/full/qwen25_math_1p5b_3digit_adacs_split_v1/results/adacs_split_scout_v1/carry_ones_to_tens/d0_ones/1/pairs_carry_ones_to_tens.json filter=lfs diff=lfs merge=lfs -text
52
+ circuit-shotting/artifacts/pod_logs/steed-medium/full/qwen25_math_1p5b_3digit_adacs_split_v1/results/adacs_split_scout_v1/carry_ones_to_tens/d0_ones/2/pairs_carry_ones_to_tens.json filter=lfs diff=lfs merge=lfs -text
53
+ circuit-shotting/artifacts/pod_logs/steed-medium/full/qwen25_math_1p5b_3digit_adacs_split_v1/results/adacs_split_scout_v1/carry_ones_to_tens/d0_ones/3/pairs_carry_ones_to_tens.json filter=lfs diff=lfs merge=lfs -text
54
+ circuit-shotting/artifacts/pod_logs/steed-medium/full/qwen25_math_1p5b_3digit_adacs_split_v1/results/adacs_split_scout_v1/carry_ones_to_tens/d0_ones/4/pairs_carry_ones_to_tens.json filter=lfs diff=lfs merge=lfs -text
55
+ circuit-shotting/artifacts/pod_logs/steed-medium/full/qwen25_math_1p5b_3digit_adacs_split_v1/results/adacs_split_scout_v1/carry_ones_to_tens/d0_ones/5/pairs_carry_ones_to_tens.json filter=lfs diff=lfs merge=lfs -text
56
+ circuit-shotting/artifacts/pod_logs/steed-medium/full/qwen25_math_1p5b_3digit_adacs_split_v1/results/adacs_split_scout_v1/carry_ones_to_tens/d0_ones/6/pairs_carry_ones_to_tens.json filter=lfs diff=lfs merge=lfs -text
57
+ circuit-shotting/artifacts/pod_logs/steed-medium/full/qwen25_math_1p5b_3digit_adacs_split_v1/results/adacs_split_scout_v1/carry_ones_to_tens/d0_ones/7/pairs_carry_ones_to_tens.json filter=lfs diff=lfs merge=lfs -text
58
+ circuit-shotting/artifacts/pod_logs/steed-medium/full/qwen25_math_1p5b_3digit_adacs_split_v1/results/adacs_split_scout_v1/carry_ones_to_tens/d0_ones/8/pairs_carry_ones_to_tens.json filter=lfs diff=lfs merge=lfs -text
59
+ circuit-shotting/artifacts/pod_logs/steed-medium/full/qwen25_math_1p5b_3digit_adacs_split_v1/results/adacs_split_scout_v1/carry_ones_to_tens/d0_ones/9/pairs_carry_ones_to_tens.json filter=lfs diff=lfs merge=lfs -text
60
+ circuit-shotting/artifacts/pod_logs/steed-medium/full/qwen25_math_1p5b_3digit_adacs_split_v1/results/adacs_split_scout_v1/carry_ones_to_tens/d1_tens/1/pairs_carry_ones_to_tens.json filter=lfs diff=lfs merge=lfs -text
61
+ circuit-shotting/artifacts/pod_logs/steed-medium/full/qwen25_math_1p5b_3digit_adacs_split_v1/results/adacs_split_scout_v1/carry_ones_to_tens/d1_tens/2/pairs_carry_ones_to_tens.json filter=lfs diff=lfs merge=lfs -text
62
+ circuit-shotting/artifacts/pod_logs/steed-medium/full/qwen25_math_1p5b_3digit_adacs_split_v1/results/adacs_split_scout_v1/carry_ones_to_tens/d1_tens/3/pairs_carry_ones_to_tens.json filter=lfs diff=lfs merge=lfs -text
63
+ circuit-shotting/artifacts/pod_logs/steed-medium/full/qwen25_math_1p5b_3digit_adacs_split_v1/results/adacs_split_scout_v1/carry_ones_to_tens/d1_tens/4/pairs_carry_ones_to_tens.json filter=lfs diff=lfs merge=lfs -text
64
+ circuit-shotting/artifacts/pod_logs/steed-medium/full/qwen25_math_1p5b_3digit_adacs_split_v1/results/adacs_split_scout_v1/carry_ones_to_tens/d1_tens/5/pairs_carry_ones_to_tens.json filter=lfs diff=lfs merge=lfs -text
65
+ circuit-shotting/artifacts/pod_logs/steed-medium/full/qwen25_math_1p5b_3digit_adacs_split_v1/results/adacs_split_scout_v1/carry_ones_to_tens/d1_tens/6/pairs_carry_ones_to_tens.json filter=lfs diff=lfs merge=lfs -text
66
+ circuit-shotting/artifacts/pod_logs/steed-medium/full/qwen25_math_1p5b_3digit_adacs_split_v1/results/adacs_split_scout_v1/carry_ones_to_tens/d1_tens/7/pairs_carry_ones_to_tens.json filter=lfs diff=lfs merge=lfs -text
67
+ circuit-shotting/artifacts/pod_logs/steed-medium/full/qwen25_math_1p5b_3digit_adacs_split_v1/results/adacs_split_scout_v1/carry_ones_to_tens/d1_tens/8/pairs_carry_ones_to_tens.json filter=lfs diff=lfs merge=lfs -text
68
+ circuit-shotting/artifacts/pod_logs/steed-medium/full/qwen25_math_1p5b_3digit_adacs_split_v1/results/adacs_split_scout_v1/carry_ones_to_tens/ones_digit_bucket/0_2/pairs_carry_ones_to_tens.json filter=lfs diff=lfs merge=lfs -text
69
+ circuit-shotting/artifacts/pod_logs/steed-medium/full/qwen25_math_1p5b_3digit_adacs_split_v1/results/adacs_split_scout_v1/carry_ones_to_tens/ones_digit_bucket/3_5/pairs_carry_ones_to_tens.json filter=lfs diff=lfs merge=lfs -text
70
+ circuit-shotting/artifacts/pod_logs/steed-medium/full/qwen25_math_1p5b_3digit_adacs_split_v1/results/adacs_split_scout_v1/carry_ones_to_tens/ones_digit_bucket/6_9/pairs_carry_ones_to_tens.json filter=lfs diff=lfs merge=lfs -text
71
+ circuit-shotting/artifacts/pod_logs/steed-medium/full/qwen25_math_1p5b_3digit_adacs_split_v1/results/adacs_split_scout_v1/carry_tens_to_hundreds/answer_len/3/pairs_carry_tens_to_hundreds.json filter=lfs diff=lfs merge=lfs -text
72
+ circuit-shotting/artifacts/pod_logs/steed-medium/full/qwen25_math_1p5b_3digit_adacs_split_v1/results/adacs_split_scout_v1/carry_tens_to_hundreds/answer_len/4/pairs_carry_tens_to_hundreds.json filter=lfs diff=lfs merge=lfs -text
73
+ circuit-shotting/artifacts/pod_logs/steed-medium/full/qwen25_math_1p5b_3digit_adacs_split_v1/results/adacs_split_scout_v1/carry_tens_to_hundreds/c1_ones_to_tens/false/pairs_carry_tens_to_hundreds.json filter=lfs diff=lfs merge=lfs -text
74
+ circuit-shotting/artifacts/pod_logs/steed-medium/full/qwen25_math_1p5b_3digit_adacs_split_v1/results/adacs_split_scout_v1/carry_tens_to_hundreds/c1_ones_to_tens/true/pairs_carry_tens_to_hundreds.json filter=lfs diff=lfs merge=lfs -text
75
+ circuit-shotting/artifacts/pod_logs/steed-medium/full/qwen25_math_1p5b_3digit_adacs_split_v1/results/adacs_split_scout_v1/carry_tens_to_hundreds/d1_tens/0/pairs_carry_tens_to_hundreds.json filter=lfs diff=lfs merge=lfs -text
76
+ circuit-shotting/artifacts/pod_logs/steed-medium/full/qwen25_math_1p5b_3digit_adacs_split_v1/results/adacs_split_scout_v1/carry_tens_to_hundreds/d1_tens/1/pairs_carry_tens_to_hundreds.json filter=lfs diff=lfs merge=lfs -text
77
+ circuit-shotting/artifacts/pod_logs/steed-medium/full/qwen25_math_1p5b_3digit_adacs_split_v1/results/adacs_split_scout_v1/carry_tens_to_hundreds/d1_tens/2/pairs_carry_tens_to_hundreds.json filter=lfs diff=lfs merge=lfs -text
78
+ circuit-shotting/artifacts/pod_logs/steed-medium/full/qwen25_math_1p5b_3digit_adacs_split_v1/results/adacs_split_scout_v1/carry_tens_to_hundreds/d1_tens/3/pairs_carry_tens_to_hundreds.json filter=lfs diff=lfs merge=lfs -text
79
+ circuit-shotting/artifacts/pod_logs/steed-medium/full/qwen25_math_1p5b_3digit_adacs_split_v1/results/adacs_split_scout_v1/carry_tens_to_hundreds/d1_tens/4/pairs_carry_tens_to_hundreds.json filter=lfs diff=lfs merge=lfs -text
80
+ circuit-shotting/artifacts/pod_logs/steed-medium/full/qwen25_math_1p5b_3digit_adacs_split_v1/results/adacs_split_scout_v1/carry_tens_to_hundreds/d1_tens/5/pairs_carry_tens_to_hundreds.json filter=lfs diff=lfs merge=lfs -text
81
+ circuit-shotting/artifacts/pod_logs/steed-medium/full/qwen25_math_1p5b_3digit_adacs_split_v1/results/adacs_split_scout_v1/carry_tens_to_hundreds/d1_tens/6/pairs_carry_tens_to_hundreds.json filter=lfs diff=lfs merge=lfs -text
82
+ circuit-shotting/artifacts/pod_logs/steed-medium/full/qwen25_math_1p5b_3digit_adacs_split_v1/results/adacs_split_scout_v1/carry_tens_to_hundreds/d1_tens/7/pairs_carry_tens_to_hundreds.json filter=lfs diff=lfs merge=lfs -text
83
+ circuit-shotting/artifacts/pod_logs/steed-medium/full/qwen25_math_1p5b_3digit_adacs_split_v1/results/adacs_split_scout_v1/carry_tens_to_hundreds/d1_tens/8/pairs_carry_tens_to_hundreds.json filter=lfs diff=lfs merge=lfs -text
84
+ circuit-shotting/artifacts/pod_logs/steed-medium/full/qwen25_math_1p5b_3digit_adacs_split_v1/results/adacs_split_scout_v1/carry_tens_to_hundreds/d1_tens/9/pairs_carry_tens_to_hundreds.json filter=lfs diff=lfs merge=lfs -text
85
+ circuit-shotting/artifacts/pod_logs/steed-medium/full/qwen25_math_1p5b_3digit_adacs_split_v1/results/adacs_split_scout_v1/carry_tens_to_hundreds/d2_hundreds/0/pairs_carry_tens_to_hundreds.json filter=lfs diff=lfs merge=lfs -text
86
+ circuit-shotting/artifacts/pod_logs/steed-medium/full/qwen25_math_1p5b_3digit_adacs_split_v1/results/adacs_split_scout_v1/carry_tens_to_hundreds/d2_hundreds/1/pairs_carry_tens_to_hundreds.json filter=lfs diff=lfs merge=lfs -text
87
+ circuit-shotting/artifacts/pod_logs/steed-medium/full/qwen25_math_1p5b_3digit_adacs_split_v1/results/adacs_split_scout_v1/carry_tens_to_hundreds/d2_hundreds/5/pairs_carry_tens_to_hundreds.json filter=lfs diff=lfs merge=lfs -text
88
+ circuit-shotting/artifacts/pod_logs/steed-medium/full/qwen25_math_1p5b_3digit_adacs_split_v1/results/adacs_split_scout_v1/carry_tens_to_hundreds/d2_hundreds/6/pairs_carry_tens_to_hundreds.json filter=lfs diff=lfs merge=lfs -text
89
+ circuit-shotting/artifacts/pod_logs/steed-medium/full/qwen25_math_1p5b_3digit_adacs_split_v1/results/adacs_split_scout_v1/carry_tens_to_hundreds/d2_hundreds/7/pairs_carry_tens_to_hundreds.json filter=lfs diff=lfs merge=lfs -text
90
+ circuit-shotting/artifacts/pod_logs/steed-medium/full/qwen25_math_1p5b_3digit_adacs_split_v1/results/adacs_split_scout_v1/carry_tens_to_hundreds/d2_hundreds/8/pairs_carry_tens_to_hundreds.json filter=lfs diff=lfs merge=lfs -text
91
+ circuit-shotting/artifacts/pod_logs/steed-medium/full/qwen25_math_1p5b_3digit_adacs_split_v1/results/adacs_split_scout_v1/carry_tens_to_hundreds/d2_hundreds/9/pairs_carry_tens_to_hundreds.json filter=lfs diff=lfs merge=lfs -text
92
+ circuit-shotting/artifacts/pod_logs/steed-medium/full/qwen25_math_1p5b_3digit_adacs_split_v1/results/adacs_split_scout_v1/hundreds_digit/answer_len/3/pairs_hundreds_digit.json filter=lfs diff=lfs merge=lfs -text
93
+ circuit-shotting/artifacts/pod_logs/steed-medium/full/qwen25_math_1p5b_3digit_adacs_split_v1/results/adacs_split_scout_v1/hundreds_digit/answer_len/4/pairs_hundreds_digit.json filter=lfs diff=lfs merge=lfs -text
94
+ circuit-shotting/artifacts/pod_logs/steed-medium/full/qwen25_math_1p5b_3digit_adacs_split_v1/results/adacs_split_scout_v1/hundreds_digit/c2_tens_to_hundreds/false/pairs_hundreds_digit.json filter=lfs diff=lfs merge=lfs -text
95
+ circuit-shotting/artifacts/pod_logs/steed-medium/full/qwen25_math_1p5b_3digit_adacs_split_v1/results/adacs_split_scout_v1/hundreds_digit/c2_tens_to_hundreds/true/pairs_hundreds_digit.json filter=lfs diff=lfs merge=lfs -text
96
+ circuit-shotting/artifacts/pod_logs/steed-medium/full/qwen25_math_1p5b_3digit_adacs_split_v1/results/adacs_split_scout_v1/hundreds_digit/c3_hundreds_to_thousands/false/pairs_hundreds_digit.json filter=lfs diff=lfs merge=lfs -text
97
+ circuit-shotting/artifacts/pod_logs/steed-medium/full/qwen25_math_1p5b_3digit_adacs_split_v1/results/adacs_split_scout_v1/hundreds_digit/c3_hundreds_to_thousands/true/pairs_hundreds_digit.json filter=lfs diff=lfs merge=lfs -text
98
+ circuit-shotting/artifacts/pod_logs/steed-medium/full/qwen25_math_1p5b_3digit_adacs_split_v1/results/adacs_split_scout_v1/hundreds_digit/hundreds_digit_bucket/0_2/pairs_hundreds_digit.json filter=lfs diff=lfs merge=lfs -text
99
+ circuit-shotting/artifacts/pod_logs/steed-medium/full/qwen25_math_1p5b_3digit_adacs_split_v1/results/adacs_split_scout_v1/hundreds_digit/hundreds_digit_bucket/3_5/pairs_hundreds_digit.json filter=lfs diff=lfs merge=lfs -text
100
+ circuit-shotting/artifacts/pod_logs/steed-medium/full/qwen25_math_1p5b_3digit_adacs_split_v1/results/adacs_split_scout_v1/hundreds_digit/hundreds_digit_bucket/6_9/pairs_hundreds_digit.json filter=lfs diff=lfs merge=lfs -text
101
+ circuit-shotting/artifacts/pod_logs/steed-medium/full/qwen25_math_1p5b_3digit_adacs_split_v1/results/adacs_split_scout_v1/ones_digit/answer_len/3/pairs_ones_digit.json filter=lfs diff=lfs merge=lfs -text
102
+ circuit-shotting/artifacts/pod_logs/steed-medium/full/qwen25_math_1p5b_3digit_adacs_split_v1/results/adacs_split_scout_v1/ones_digit/answer_len/4/pairs_ones_digit.json filter=lfs diff=lfs merge=lfs -text
103
+ circuit-shotting/artifacts/pod_logs/steed-medium/full/qwen25_math_1p5b_3digit_adacs_split_v1/results/adacs_split_scout_v1/ones_digit/c1_ones_to_tens/false/pairs_ones_digit.json filter=lfs diff=lfs merge=lfs -text
104
+ circuit-shotting/artifacts/pod_logs/steed-medium/full/qwen25_math_1p5b_3digit_adacs_split_v1/results/adacs_split_scout_v1/ones_digit/c1_ones_to_tens/true/pairs_ones_digit.json filter=lfs diff=lfs merge=lfs -text
105
+ circuit-shotting/artifacts/pod_logs/steed-medium/full/qwen25_math_1p5b_3digit_adacs_split_v1/results/adacs_split_scout_v1/ones_digit/ones_digit_bucket/0_2/pairs_ones_digit.json filter=lfs diff=lfs merge=lfs -text
106
+ circuit-shotting/artifacts/pod_logs/steed-medium/full/qwen25_math_1p5b_3digit_adacs_split_v1/results/adacs_split_scout_v1/ones_digit/ones_digit_bucket/3_5/pairs_ones_digit.json filter=lfs diff=lfs merge=lfs -text
107
+ circuit-shotting/artifacts/pod_logs/steed-medium/full/qwen25_math_1p5b_3digit_adacs_split_v1/results/adacs_split_scout_v1/ones_digit/ones_digit_bucket/6_9/pairs_ones_digit.json filter=lfs diff=lfs merge=lfs -text
108
+ circuit-shotting/artifacts/pod_logs/steed-medium/full/qwen25_math_1p5b_3digit_adacs_split_v1/results/adacs_split_scout_v1/tens_digit/answer_len/3/pairs_tens_digit.json filter=lfs diff=lfs merge=lfs -text
109
+ circuit-shotting/artifacts/pod_logs/steed-medium/full/qwen25_math_1p5b_3digit_adacs_split_v1/results/adacs_split_scout_v1/tens_digit/answer_len/4/pairs_tens_digit.json filter=lfs diff=lfs merge=lfs -text
110
+ circuit-shotting/artifacts/pod_logs/steed-medium/full/qwen25_math_1p5b_3digit_adacs_split_v1/results/adacs_split_scout_v1/tens_digit/c1_ones_to_tens/false/pairs_tens_digit.json filter=lfs diff=lfs merge=lfs -text
111
+ circuit-shotting/artifacts/pod_logs/steed-medium/full/qwen25_math_1p5b_3digit_adacs_split_v1/results/adacs_split_scout_v1/tens_digit/c1_ones_to_tens/true/pairs_tens_digit.json filter=lfs diff=lfs merge=lfs -text
112
+ circuit-shotting/artifacts/pod_logs/steed-medium/full/qwen25_math_1p5b_3digit_adacs_split_v1/results/adacs_split_scout_v1/tens_digit/c2_tens_to_hundreds/false/pairs_tens_digit.json filter=lfs diff=lfs merge=lfs -text
113
+ circuit-shotting/artifacts/pod_logs/steed-medium/full/qwen25_math_1p5b_3digit_adacs_split_v1/results/adacs_split_scout_v1/tens_digit/c2_tens_to_hundreds/true/pairs_tens_digit.json filter=lfs diff=lfs merge=lfs -text
114
+ circuit-shotting/artifacts/pod_logs/steed-medium/full/qwen25_math_1p5b_3digit_adacs_split_v1/results/adacs_split_scout_v1/tens_digit/tens_digit_bucket/0_2/pairs_tens_digit.json filter=lfs diff=lfs merge=lfs -text
115
+ circuit-shotting/artifacts/pod_logs/steed-medium/full/qwen25_math_1p5b_3digit_adacs_split_v1/results/adacs_split_scout_v1/tens_digit/tens_digit_bucket/3_5/pairs_tens_digit.json filter=lfs diff=lfs merge=lfs -text
116
+ circuit-shotting/artifacts/pod_logs/steed-medium/full/qwen25_math_1p5b_3digit_adacs_split_v1/results/adacs_split_scout_v1/tens_digit/tens_digit_bucket/6_9/pairs_tens_digit.json filter=lfs diff=lfs merge=lfs -text
117
+ circuit-shotting/artifacts/pod_logs/steed-medium/full/qwen25_math_1p5b_3digit_adacs_v1/results/adacs_milestone1_full/pairs_answer_length.json filter=lfs diff=lfs merge=lfs -text
118
+ circuit-shotting/artifacts/pod_logs/steed-medium/full/qwen25_math_1p5b_3digit_adacs_v1/results/adacs_milestone1_full/pairs_carry_into_thousands.json filter=lfs diff=lfs merge=lfs -text
119
+ circuit-shotting/artifacts/pod_logs/steed-medium/full/qwen25_math_1p5b_3digit_adacs_v1/results/adacs_milestone1_full/pairs_carry_ones_to_tens.json filter=lfs diff=lfs merge=lfs -text
120
+ circuit-shotting/artifacts/pod_logs/steed-medium/full/qwen25_math_1p5b_3digit_adacs_v1/results/adacs_milestone1_full/pairs_carry_tens_to_hundreds.json filter=lfs diff=lfs merge=lfs -text
121
+ circuit-shotting/artifacts/pod_logs/steed-medium/full/qwen25_math_1p5b_3digit_adacs_v1/results/adacs_milestone1_full/pairs_hundreds_digit.json filter=lfs diff=lfs merge=lfs -text
122
+ circuit-shotting/artifacts/pod_logs/steed-medium/full/qwen25_math_1p5b_3digit_adacs_v1/results/adacs_milestone1_full/pairs_ones_digit.json filter=lfs diff=lfs merge=lfs -text
123
+ circuit-shotting/artifacts/pod_logs/steed-medium/full/qwen25_math_1p5b_3digit_adacs_v1/results/adacs_milestone1_full/pairs_tens_digit.json filter=lfs diff=lfs merge=lfs -text
124
+ circuit-shotting/artifacts/pod_logs/steed-medium/full/qwen25_math_1p5b_3digit_success_vs_failure_v1_repro_backbone_0_1_2_26_27/paired_dataset.json filter=lfs diff=lfs merge=lfs -text
circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/data/qwen25_math_1p5b_rank_ladder_r16_correct_addition.json ADDED
The diff for this file is too large to render. See raw diff
 
circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/data/qwen25_math_1p5b_rank_ladder_r1_correct_addition.json ADDED
The diff for this file is too large to render. See raw diff
 
circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/data/qwen25_math_1p5b_rank_ladder_r2_correct_addition.json ADDED
The diff for this file is too large to render. See raw diff
 
circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/data/qwen25_math_1p5b_rank_ladder_r4_correct_addition.json ADDED
The diff for this file is too large to render. See raw diff
 
circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/data/qwen25_math_1p5b_rank_ladder_r8_correct_addition.json ADDED
The diff for this file is too large to render. See raw diff
 
circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_16_correct_dataset.log ADDED
@@ -0,0 +1,26 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ Device: cuda dtype: bfloat16
2
+ Loading results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_16/lora_r16_beta005_merged ...
3
+
4
+ Pool: 8,100 unique (a, b) pairs in [10, 99] shard=1/1
5
+ 512/8100 correct= 504 acc=98.4% 644.8 ex/s eta 0.2 min
6
+ 1024/8100 correct= 1006 acc=98.2% 836.3 ex/s eta 0.1 min
7
+ 1536/8100 correct= 1511 acc=98.4% 930.4 ex/s eta 0.1 min
8
+ 2048/8100 correct= 2014 acc=98.3% 982.3 ex/s eta 0.1 min
9
+ 2560/8100 correct= 2521 acc=98.5% 1013.6 ex/s eta 0.1 min
10
+ 3072/8100 correct= 3026 acc=98.5% 1035.5 ex/s eta 0.1 min
11
+ 3584/8100 correct= 3533 acc=98.6% 1052.1 ex/s eta 0.1 min
12
+ 4096/8100 correct= 4037 acc=98.6% 1064.1 ex/s eta 0.1 min
13
+ 4608/8100 correct= 4538 acc=98.5% 1073.1 ex/s eta 0.1 min
14
+ 5120/8100 correct= 5043 acc=98.5% 1079.3 ex/s eta 0.0 min
15
+ 5632/8100 correct= 5548 acc=98.5% 1085.7 ex/s eta 0.0 min
16
+ 6144/8100 correct= 6048 acc=98.4% 1091.2 ex/s eta 0.0 min
17
+ 6656/8100 correct= 6553 acc=98.5% 1094.2 ex/s eta 0.0 min
18
+ 7168/8100 correct= 7060 acc=98.5% 1096.4 ex/s eta 0.0 min
19
+ 7680/8100 correct= 7559 acc=98.4% 1099.4 ex/s eta 0.0 min
20
+ 8100/8100 correct= 7971 acc=98.4% 1100.4 ex/s eta 0.0 min
21
+
22
+ =======================================================
23
+ Tested: 8,100
24
+ Correct: 7,971 (98.4%)
25
+ Correct saved: data/qwen25_math_1p5b_rank_ladder_r16_correct_addition.json
26
+ =======================================================
circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_16_direct_greedy.log ADDED
@@ -0,0 +1,179 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ [records] results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_16/position_adacs positions=hundreds,tens,ones
2
+ records=1,500 skipped=0
3
+ [model] results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_16/lora_r16_beta005_merged device=cuda dtype=bfloat16 batch=256
4
+
5
+ topk_100 mlp= 2099 exact=12.13%
6
+ rel_0.05 mlp= 3042 exact=21.13%
7
+ topk_500 mlp= 13729 exact=37.87%
8
+ rel_0.01 mlp= 25570 exact=35.13%
9
+ topk_2000 mlp= 55061 exact=29.00%
10
+ rel_0.001 mlp= 224944 exact=98.13%
11
+ greedy_init mlp= 13729 exact=37.87%
12
+ greedy_step_1_add__rel_0.001_minus_topk_500__layer_06 mlp= 22493 exact=52.27%
13
+ greedy_step_1_add__rel_0.001_minus_topk_500__layer_08 mlp= 22496 exact=50.27%
14
+ greedy_step_1_add__rel_0.001_minus_topk_500__layer_07 mlp= 22471 exact=48.80%
15
+ greedy_step_1_add__rel_0.001_minus_topk_500__layer_05 mlp= 20396 exact=47.00%
16
+ greedy_step_1_add__rel_0.001_minus_topk_500__layer_04 mlp= 21626 exact=44.87%
17
+ greedy_step_1_add__rel_0.001_minus_topk_500__layer_02 mlp= 17486 exact=44.27%
18
+ greedy_step_1_add__rel_0.001_minus_topk_500__layer_00 mlp= 19999 exact=43.93%
19
+ greedy_step_1_add__rel_0.001_minus_topk_500__layer_12 mlp= 22162 exact=43.93%
20
+ greedy_step_1_add__rel_0.001_minus_topk_500__layer_15 mlp= 22188 exact=43.60%
21
+ greedy_step_1_drop__topk_500__layer_10 mlp= 13650 exact=43.27%
22
+ greedy_step_1_add__rel_0.001_minus_topk_500__layer_10 mlp= 22392 exact=43.20%
23
+ greedy_step_1_drop__topk_500__layer_05 mlp= 13600 exact=42.73%
24
+ greedy_step_1_add__rel_0.001_minus_topk_500__layer_03 mlp= 17439 exact=42.60%
25
+ greedy_step_1_drop__topk_500__layer_12 mlp= 13596 exact=42.33%
26
+ greedy_step_1_add__rel_0.001_minus_topk_500__layer_13 mlp= 22197 exact=42.00%
27
+ greedy_step_1_add__rel_0.001_minus_topk_500__layer_14 mlp= 22276 exact=42.00%
28
+ greedy_step_1_add__rel_0.001_minus_topk_500__layer_09 mlp= 22473 exact=42.00%
29
+ greedy_step_1_add__rel_0.001_minus_topk_500__layer_16 mlp= 22362 exact=41.93%
30
+ greedy_step_1_drop__topk_500__layer_08 mlp= 13618 exact=41.67%
31
+ greedy_step_1_add__rel_0.001_minus_topk_500__layer_17 mlp= 22462 exact=41.67%
32
+ greedy_step_1_add__topk_2000_minus_topk_500__layer_07 mlp= 14430 exact=41.60%
33
+ greedy_step_1_drop__topk_500__layer_17 mlp= 13611 exact=41.47%
34
+ greedy_step_1_add__rel_0.001_minus_topk_500__layer_21 mlp= 22117 exact=41.20%
35
+ greedy_step_1_add__rel_0.001_minus_topk_500__layer_19 mlp= 22315 exact=41.13%
36
+ greedy_step_1_drop__topk_500__layer_14 mlp= 13589 exact=40.93%
37
+ greedy_step_1_add__rel_0.01_minus_topk_500__layer_21 mlp= 14505 exact=40.87%
38
+ greedy_step_1_add__rel_0.001_minus_topk_500__layer_01 mlp= 15978 exact=40.87%
39
+ greedy_step_1_add__topk_2000_minus_topk_500__layer_00 mlp= 14044 exact=40.47%
40
+ greedy_step_1_add__topk_2000_minus_topk_500__layer_02 mlp= 13990 exact=40.40%
41
+ greedy_step_1_drop__topk_500__layer_15 mlp= 13633 exact=40.33%
42
+ greedy_step_1_drop__topk_500__layer_09 mlp= 13629 exact=40.33%
43
+ greedy_step_1_add__rel_0.01_minus_topk_500__layer_16 mlp= 13968 exact=40.33%
44
+ greedy_step_2_add__rel_0.001_minus_topk_500__layer_08 mlp= 31260 exact=57.40%
45
+ greedy_step_2_add__rel_0.001_minus_topk_500__layer_07 mlp= 31235 exact=56.20%
46
+ greedy_step_2_add__rel_0.001_minus_topk_500__layer_05 mlp= 29160 exact=61.00%
47
+ greedy_step_2_add__rel_0.001_minus_topk_500__layer_04 mlp= 30390 exact=65.40%
48
+ greedy_step_2_add__rel_0.001_minus_topk_500__layer_02 mlp= 26250 exact=55.87%
49
+ greedy_step_2_add__rel_0.001_minus_topk_500__layer_00 mlp= 28763 exact=57.27%
50
+ greedy_step_2_add__rel_0.001_minus_topk_500__layer_12 mlp= 30926 exact=57.33%
51
+ greedy_step_2_add__rel_0.001_minus_topk_500__layer_15 mlp= 30952 exact=59.80%
52
+ greedy_step_2_drop__topk_500__layer_10 mlp= 22414 exact=54.27%
53
+ greedy_step_2_add__rel_0.001_minus_topk_500__layer_10 mlp= 31156 exact=59.40%
54
+ greedy_step_2_drop__topk_500__layer_05 mlp= 22364 exact=54.13%
55
+ greedy_step_2_add__rel_0.001_minus_topk_500__layer_03 mlp= 26203 exact=57.07%
56
+ greedy_step_2_drop__topk_500__layer_12 mlp= 22360 exact=52.93%
57
+ greedy_step_2_add__rel_0.001_minus_topk_500__layer_13 mlp= 30961 exact=53.27%
58
+ greedy_step_2_add__rel_0.001_minus_topk_500__layer_14 mlp= 31040 exact=56.53%
59
+ greedy_step_2_add__rel_0.001_minus_topk_500__layer_09 mlp= 31237 exact=59.47%
60
+ greedy_step_2_add__rel_0.001_minus_topk_500__layer_16 mlp= 31126 exact=50.87%
61
+ greedy_step_2_drop__topk_500__layer_08 mlp= 22382 exact=57.20%
62
+ greedy_step_2_add__rel_0.001_minus_topk_500__layer_17 mlp= 31226 exact=57.20%
63
+ greedy_step_2_add__topk_2000_minus_topk_500__layer_07 mlp= 23194 exact=53.87%
64
+ greedy_step_2_drop__topk_500__layer_17 mlp= 22375 exact=55.47%
65
+ greedy_step_2_add__rel_0.001_minus_topk_500__layer_21 mlp= 30881 exact=56.07%
66
+ greedy_step_2_add__rel_0.001_minus_topk_500__layer_19 mlp= 31079 exact=56.13%
67
+ greedy_step_2_drop__topk_500__layer_14 mlp= 22353 exact=53.20%
68
+ greedy_step_2_add__rel_0.01_minus_topk_500__layer_21 mlp= 23269 exact=55.40%
69
+ greedy_step_2_add__rel_0.001_minus_topk_500__layer_01 mlp= 24742 exact=54.67%
70
+ greedy_step_2_add__topk_2000_minus_topk_500__layer_00 mlp= 22808 exact=55.27%
71
+ greedy_step_2_add__topk_2000_minus_topk_500__layer_02 mlp= 22754 exact=53.40%
72
+ greedy_step_2_drop__topk_500__layer_15 mlp= 22397 exact=55.60%
73
+ greedy_step_2_drop__topk_500__layer_09 mlp= 22393 exact=53.93%
74
+ greedy_step_2_add__rel_0.01_minus_topk_500__layer_16 mlp= 22732 exact=51.93%
75
+ greedy_step_3_add__rel_0.001_minus_topk_500__layer_08 mlp= 39157 exact=66.53%
76
+ greedy_step_3_add__rel_0.001_minus_topk_500__layer_07 mlp= 39132 exact=71.53%
77
+ greedy_step_3_add__rel_0.001_minus_topk_500__layer_05 mlp= 37057 exact=66.07%
78
+ greedy_step_3_add__rel_0.001_minus_topk_500__layer_02 mlp= 34147 exact=64.87%
79
+ greedy_step_3_add__rel_0.001_minus_topk_500__layer_00 mlp= 36660 exact=68.67%
80
+ greedy_step_3_add__rel_0.001_minus_topk_500__layer_12 mlp= 38823 exact=68.33%
81
+ greedy_step_3_add__rel_0.001_minus_topk_500__layer_15 mlp= 38849 exact=71.13%
82
+ greedy_step_3_drop__topk_500__layer_10 mlp= 30311 exact=67.53%
83
+ greedy_step_3_add__rel_0.001_minus_topk_500__layer_10 mlp= 39053 exact=70.20%
84
+ greedy_step_3_drop__topk_500__layer_05 mlp= 30261 exact=66.67%
85
+ greedy_step_3_add__rel_0.001_minus_topk_500__layer_03 mlp= 34100 exact=67.33%
86
+ greedy_step_3_drop__topk_500__layer_12 mlp= 30257 exact=67.20%
87
+ greedy_step_3_add__rel_0.001_minus_topk_500__layer_13 mlp= 38858 exact=68.13%
88
+ greedy_step_3_add__rel_0.001_minus_topk_500__layer_14 mlp= 38937 exact=70.47%
89
+ greedy_step_3_add__rel_0.001_minus_topk_500__layer_09 mlp= 39134 exact=67.47%
90
+ greedy_step_3_add__rel_0.001_minus_topk_500__layer_16 mlp= 39023 exact=68.27%
91
+ greedy_step_3_drop__topk_500__layer_08 mlp= 30279 exact=68.60%
92
+ greedy_step_3_add__rel_0.001_minus_topk_500__layer_17 mlp= 39123 exact=68.87%
93
+ greedy_step_3_add__topk_2000_minus_topk_500__layer_07 mlp= 31091 exact=67.00%
94
+ greedy_step_3_drop__topk_500__layer_17 mlp= 30272 exact=69.20%
95
+ greedy_step_3_add__rel_0.001_minus_topk_500__layer_21 mlp= 38778 exact=69.40%
96
+ greedy_step_3_add__rel_0.001_minus_topk_500__layer_19 mlp= 38976 exact=69.07%
97
+ greedy_step_3_drop__topk_500__layer_14 mlp= 30250 exact=66.40%
98
+ greedy_step_3_add__rel_0.01_minus_topk_500__layer_21 mlp= 31166 exact=68.07%
99
+ greedy_step_3_add__rel_0.001_minus_topk_500__layer_01 mlp= 32639 exact=63.73%
100
+ greedy_step_3_add__topk_2000_minus_topk_500__layer_00 mlp= 30705 exact=66.27%
101
+ greedy_step_3_add__topk_2000_minus_topk_500__layer_02 mlp= 30651 exact=64.00%
102
+ greedy_step_3_drop__topk_500__layer_15 mlp= 30294 exact=69.80%
103
+ greedy_step_3_drop__topk_500__layer_09 mlp= 30290 exact=66.80%
104
+ greedy_step_3_add__rel_0.01_minus_topk_500__layer_16 mlp= 30629 exact=65.47%
105
+ greedy_step_4_add__rel_0.001_minus_topk_500__layer_08 mlp= 47899 exact=69.47%
106
+ greedy_step_4_add__rel_0.001_minus_topk_500__layer_05 mlp= 45799 exact=73.27%
107
+ greedy_step_4_add__rel_0.001_minus_topk_500__layer_02 mlp= 42889 exact=73.53%
108
+ greedy_step_4_add__rel_0.001_minus_topk_500__layer_00 mlp= 45402 exact=74.07%
109
+ greedy_step_4_add__rel_0.001_minus_topk_500__layer_12 mlp= 47565 exact=74.47%
110
+ greedy_step_4_add__rel_0.001_minus_topk_500__layer_15 mlp= 47591 exact=77.20%
111
+ greedy_step_4_drop__topk_500__layer_10 mlp= 39053 exact=73.00%
112
+ greedy_step_4_add__rel_0.001_minus_topk_500__layer_10 mlp= 47795 exact=73.73%
113
+ greedy_step_4_drop__topk_500__layer_05 mlp= 39003 exact=74.20%
114
+ greedy_step_4_add__rel_0.001_minus_topk_500__layer_03 mlp= 42842 exact=74.80%
115
+ greedy_step_4_drop__topk_500__layer_12 mlp= 38999 exact=73.93%
116
+ greedy_step_4_add__rel_0.001_minus_topk_500__layer_13 mlp= 47600 exact=72.33%
117
+ greedy_step_4_add__rel_0.001_minus_topk_500__layer_14 mlp= 47679 exact=72.93%
118
+ greedy_step_4_add__rel_0.001_minus_topk_500__layer_09 mlp= 47876 exact=76.13%
119
+ greedy_step_4_add__rel_0.001_minus_topk_500__layer_16 mlp= 47765 exact=74.13%
120
+ greedy_step_4_drop__topk_500__layer_08 mlp= 39021 exact=73.27%
121
+ greedy_step_4_add__rel_0.001_minus_topk_500__layer_17 mlp= 47865 exact=78.73%
122
+ greedy_step_4_drop__topk_500__layer_17 mlp= 39014 exact=74.73%
123
+ greedy_step_4_add__rel_0.001_minus_topk_500__layer_21 mlp= 47520 exact=75.00%
124
+ greedy_step_4_add__rel_0.001_minus_topk_500__layer_19 mlp= 47718 exact=77.00%
125
+ greedy_step_4_drop__topk_500__layer_14 mlp= 38992 exact=70.40%
126
+ greedy_step_4_add__rel_0.01_minus_topk_500__layer_21 mlp= 39908 exact=74.67%
127
+ greedy_step_4_add__rel_0.001_minus_topk_500__layer_01 mlp= 41381 exact=71.47%
128
+ greedy_step_4_add__topk_2000_minus_topk_500__layer_00 mlp= 39447 exact=73.93%
129
+ greedy_step_4_add__topk_2000_minus_topk_500__layer_02 mlp= 39393 exact=73.13%
130
+ greedy_step_4_drop__topk_500__layer_15 mlp= 39036 exact=74.07%
131
+ greedy_step_4_drop__topk_500__layer_09 mlp= 39032 exact=70.27%
132
+ greedy_step_4_add__rel_0.01_minus_topk_500__layer_16 mlp= 39371 exact=72.33%
133
+ greedy_step_5_add__rel_0.001_minus_topk_500__layer_08 mlp= 56632 exact=76.93%
134
+ greedy_step_5_add__rel_0.001_minus_topk_500__layer_05 mlp= 54532 exact=80.60%
135
+ greedy_step_5_add__rel_0.001_minus_topk_500__layer_02 mlp= 51622 exact=80.53%
136
+ greedy_step_5_add__rel_0.001_minus_topk_500__layer_00 mlp= 54135 exact=79.40%
137
+ greedy_step_5_add__rel_0.001_minus_topk_500__layer_12 mlp= 56298 exact=81.00%
138
+ greedy_step_5_add__rel_0.001_minus_topk_500__layer_15 mlp= 56324 exact=82.33%
139
+ greedy_step_5_drop__topk_500__layer_10 mlp= 47786 exact=76.93%
140
+ greedy_step_5_add__rel_0.001_minus_topk_500__layer_10 mlp= 56528 exact=83.00%
141
+ greedy_step_5_drop__topk_500__layer_05 mlp= 47736 exact=81.00%
142
+ greedy_step_5_add__rel_0.001_minus_topk_500__layer_03 mlp= 51575 exact=80.00%
143
+ greedy_step_5_drop__topk_500__layer_12 mlp= 47732 exact=78.93%
144
+ greedy_step_5_add__rel_0.001_minus_topk_500__layer_13 mlp= 56333 exact=77.00%
145
+ greedy_step_5_add__rel_0.001_minus_topk_500__layer_14 mlp= 56412 exact=76.60%
146
+ greedy_step_5_add__rel_0.001_minus_topk_500__layer_09 mlp= 56609 exact=82.53%
147
+ greedy_step_5_add__rel_0.001_minus_topk_500__layer_16 mlp= 56498 exact=83.73%
148
+ greedy_step_5_drop__topk_500__layer_08 mlp= 47754 exact=81.07%
149
+ greedy_step_5_drop__topk_500__layer_17 mlp= 47747 exact=81.07%
150
+ greedy_step_5_add__rel_0.001_minus_topk_500__layer_21 mlp= 56253 exact=83.13%
151
+ greedy_step_5_add__rel_0.001_minus_topk_500__layer_19 mlp= 56451 exact=80.27%
152
+ greedy_step_5_drop__topk_500__layer_14 mlp= 47725 exact=76.47%
153
+ greedy_step_5_add__rel_0.01_minus_topk_500__layer_21 mlp= 48641 exact=81.53%
154
+ greedy_step_5_add__rel_0.001_minus_topk_500__layer_01 mlp= 50114 exact=78.20%
155
+ greedy_step_5_add__topk_2000_minus_topk_500__layer_00 mlp= 48180 exact=78.93%
156
+ greedy_step_5_add__topk_2000_minus_topk_500__layer_02 mlp= 48126 exact=79.20%
157
+ greedy_step_5_drop__topk_500__layer_15 mlp= 47769 exact=77.20%
158
+ greedy_step_5_drop__topk_500__layer_09 mlp= 47765 exact=77.40%
159
+ greedy_step_5_add__rel_0.01_minus_topk_500__layer_16 mlp= 48104 exact=80.13%
160
+ greedy_step_6_add__rel_0.001_minus_topk_500__layer_08 mlp= 65265 exact=81.20%
161
+ greedy_step_6_add__rel_0.001_minus_topk_500__layer_05 mlp= 63165 exact=84.73%
162
+ greedy_step_6_add__rel_0.001_minus_topk_500__layer_02 mlp= 60255 exact=84.60%
163
+ greedy_step_6_add__rel_0.001_minus_topk_500__layer_00 mlp= 62768 exact=84.53%
164
+ greedy_step_6_add__rel_0.001_minus_topk_500__layer_12 mlp= 64931 exact=81.40%
165
+ greedy_step_6_add__rel_0.001_minus_topk_500__layer_15 mlp= 64957 exact=85.73%
166
+ greedy_step_6_drop__topk_500__layer_10 mlp= 56419 exact=84.93%
167
+ greedy_step_6_add__rel_0.001_minus_topk_500__layer_10 mlp= 65161 exact=84.40%
168
+ greedy_step_6_drop__topk_500__layer_05 mlp= 56369 exact=85.20%
169
+ greedy_step_6_add__rel_0.001_minus_topk_500__layer_03 mlp= 60208 exact=85.40%
170
+ greedy_step_6_drop__topk_500__layer_12 mlp= 56365 exact=84.40%
171
+ greedy_step_6_add__rel_0.001_minus_topk_500__layer_13 mlp= 64966 exact=82.27%
172
+ greedy_step_6_add__rel_0.001_minus_topk_500__layer_14 mlp= 65045 exact=80.27%
173
+ greedy_step_6_add__rel_0.001_minus_topk_500__layer_09 mlp= 65242 exact=83.80%
174
+ greedy_step_6_drop__topk_500__layer_08 mlp= 56387 exact=85.67%
175
+ greedy_step_6_drop__topk_500__layer_17 mlp= 56380 exact=85.40%
176
+ greedy_step_6_add__rel_0.001_minus_topk_500__layer_21 mlp= 64886 exact=87.60%
177
+ greedy_step_6_add__rel_0.001_minus_topk_500__layer_19 mlp= 65084 exact=85.67%
178
+ greedy_step_6_drop__topk_500__layer_14 mlp= 56358 exact=82.73%
179
+ greedy_step_6_add__rel_0.01_minus_topk_500__layer_21 mlp= 57274 exact=85.47%
circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_16_direct_rank_shard0.log ADDED
@@ -0,0 +1,68 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ [records] results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_16/position_adacs positions=hundreds,tens,ones
2
+ records=1,500 skipped=0
3
+ [model] results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_16/lora_r16_beta005_merged device=cuda dtype=bfloat16 batch=256
4
+
5
+ topk_100 mlp= 2099 exact=12.13%
6
+ rel_0.05 mlp= 3042 exact=21.13%
7
+ topk_500 mlp= 13729 exact=37.87%
8
+ rel_0.01 mlp= 25570 exact=35.13%
9
+ topk_2000 mlp= 55061 exact=29.00%
10
+ rel_0.001 mlp= 224944 exact=98.13%
11
+ [rank] todo=28 shard=0/4
12
+ drop__topk_500__layer_00 mlp= 13620 exact=6.40%
13
+ delta=-31.47% per1k=-288.685%
14
+ drop__topk_500__layer_04 mlp= 13527 exact=15.13%
15
+ delta=-22.73% per1k=-112.541%
16
+ drop__topk_500__layer_08 mlp= 13618 exact=41.67%
17
+ delta=+3.80% per1k=+34.234%
18
+ drop__topk_500__layer_12 mlp= 13596 exact=42.33%
19
+ delta=+4.47% per1k=+33.584%
20
+ drop__topk_500__layer_16 mlp= 13603 exact=36.53%
21
+ delta=-1.33% per1k=-10.582%
22
+ drop__topk_500__layer_20 mlp= 13339 exact=32.07%
23
+ delta=-5.80% per1k=-14.872%
24
+ drop__topk_500__layer_24 mlp= 10885 exact=34.13%
25
+ delta=-3.73% per1k=-1.313%
26
+ add__rel_0.01_minus_topk_500__layer_00 mlp= 13794 exact=39.60%
27
+ delta=+1.73% per1k=+26.667%
28
+ add__rel_0.01_minus_topk_500__layer_04 mlp= 13874 exact=38.67%
29
+ delta=+0.80% per1k=+5.517%
30
+ add__rel_0.01_minus_topk_500__layer_08 mlp= 13903 exact=36.93%
31
+ delta=-0.93% per1k=-5.364%
32
+ add__rel_0.01_minus_topk_500__layer_12 mlp= 13897 exact=37.33%
33
+ delta=-0.53% per1k=-3.175%
34
+ add__rel_0.01_minus_topk_500__layer_16 mlp= 13968 exact=40.33%
35
+ delta=+2.47% per1k=+10.321%
36
+ add__rel_0.01_minus_topk_500__layer_20 mlp= 14288 exact=37.73%
37
+ delta=-0.13% per1k=-0.239%
38
+ add__rel_0.01_minus_topk_500__layer_24 mlp= 15357 exact=38.07%
39
+ delta=+0.20% per1k=+0.123%
40
+ add__topk_2000_minus_topk_500__layer_00 mlp= 14044 exact=40.47%
41
+ delta=+2.60% per1k=+8.254%
42
+ add__topk_2000_minus_topk_500__layer_04 mlp= 14179 exact=40.27%
43
+ delta=+2.40% per1k=+5.333%
44
+ add__topk_2000_minus_topk_500__layer_08 mlp= 14303 exact=36.27%
45
+ delta=-1.60% per1k=-2.787%
46
+ add__topk_2000_minus_topk_500__layer_12 mlp= 14237 exact=36.93%
47
+ delta=-0.93% per1k=-1.837%
48
+ add__topk_2000_minus_topk_500__layer_16 mlp= 14416 exact=38.13%
49
+ delta=+0.27% per1k=+0.388%
50
+ add__topk_2000_minus_topk_500__layer_20 mlp= 16168 exact=37.27%
51
+ delta=-0.60% per1k=-0.246%
52
+ add__topk_2000_minus_topk_500__layer_24 mlp= 18572 exact=38.00%
53
+ delta=+0.13% per1k=+0.028%
54
+ add__rel_0.001_minus_topk_500__layer_00 mlp= 19999 exact=43.93%
55
+ delta=+6.07% per1k=+0.968%
56
+ add__rel_0.001_minus_topk_500__layer_04 mlp= 21626 exact=44.87%
57
+ delta=+7.00% per1k=+0.886%
58
+ add__rel_0.001_minus_topk_500__layer_08 mlp= 22496 exact=50.27%
59
+ delta=+12.40% per1k=+1.414%
60
+ add__rel_0.001_minus_topk_500__layer_12 mlp= 22162 exact=43.93%
61
+ delta=+6.07% per1k=+0.719%
62
+ add__rel_0.001_minus_topk_500__layer_16 mlp= 22362 exact=41.93%
63
+ delta=+4.07% per1k=+0.471%
64
+ add__rel_0.001_minus_topk_500__layer_20 mlp= 22296 exact=38.27%
65
+ delta=+0.40% per1k=+0.047%
66
+ add__rel_0.001_minus_topk_500__layer_24 mlp= 19845 exact=38.20%
67
+ delta=+0.33% per1k=+0.055%
68
+ [done] rank-only -> results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_16/direct_group_rank_shard0.json
circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_16_direct_rank_shard1.log ADDED
@@ -0,0 +1,68 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ [records] results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_16/position_adacs positions=hundreds,tens,ones
2
+ records=1,500 skipped=0
3
+ [model] results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_16/lora_r16_beta005_merged device=cuda dtype=bfloat16 batch=256
4
+
5
+ topk_100 mlp= 2099 exact=12.13%
6
+ rel_0.05 mlp= 3042 exact=21.13%
7
+ topk_500 mlp= 13729 exact=37.87%
8
+ rel_0.01 mlp= 25570 exact=35.13%
9
+ topk_2000 mlp= 55061 exact=29.00%
10
+ rel_0.001 mlp= 224944 exact=98.13%
11
+ [rank] todo=28 shard=1/4
12
+ drop__topk_500__layer_01 mlp= 13647 exact=26.60%
13
+ delta=-11.27% per1k=-137.398%
14
+ drop__topk_500__layer_05 mlp= 13600 exact=42.73%
15
+ delta=+4.87% per1k=+37.726%
16
+ drop__topk_500__layer_09 mlp= 13629 exact=40.33%
17
+ delta=+2.47% per1k=+24.667%
18
+ drop__topk_500__layer_13 mlp= 13607 exact=40.00%
19
+ delta=+2.13% per1k=+17.486%
20
+ drop__topk_500__layer_17 mlp= 13611 exact=41.47%
21
+ delta=+3.60% per1k=+30.508%
22
+ drop__topk_500__layer_21 mlp= 13160 exact=13.40%
23
+ delta=-24.47% per1k=-42.999%
24
+ drop__topk_500__layer_25 mlp= 10564 exact=32.13%
25
+ delta=-5.73% per1k=-1.811%
26
+ add__rel_0.01_minus_topk_500__layer_01 mlp= 13791 exact=39.00%
27
+ delta=+1.13% per1k=+18.280%
28
+ add__rel_0.01_minus_topk_500__layer_05 mlp= 13817 exact=37.87%
29
+ delta=+0.00% per1k=+0.000%
30
+ add__rel_0.01_minus_topk_500__layer_09 mlp= 13898 exact=36.67%
31
+ delta=-1.20% per1k=-7.101%
32
+ add__rel_0.01_minus_topk_500__layer_13 mlp= 13933 exact=36.80%
33
+ delta=-1.07% per1k=-5.229%
34
+ add__rel_0.01_minus_topk_500__layer_17 mlp= 13983 exact=37.67%
35
+ delta=-0.20% per1k=-0.787%
36
+ add__rel_0.01_minus_topk_500__layer_21 mlp= 14505 exact=40.87%
37
+ delta=+3.00% per1k=+3.866%
38
+ add__rel_0.01_minus_topk_500__layer_25 mlp= 15462 exact=38.40%
39
+ delta=+0.53% per1k=+0.308%
40
+ add__topk_2000_minus_topk_500__layer_01 mlp= 13919 exact=38.93%
41
+ delta=+1.07% per1k=+5.614%
42
+ add__topk_2000_minus_topk_500__layer_05 mlp= 14091 exact=39.27%
43
+ delta=+1.40% per1k=+3.867%
44
+ add__topk_2000_minus_topk_500__layer_09 mlp= 14284 exact=37.60%
45
+ delta=-0.27% per1k=-0.480%
46
+ add__topk_2000_minus_topk_500__layer_13 mlp= 14367 exact=34.60%
47
+ delta=-3.27% per1k=-5.120%
48
+ add__topk_2000_minus_topk_500__layer_17 mlp= 14544 exact=38.87%
49
+ delta=+1.00% per1k=+1.227%
50
+ add__topk_2000_minus_topk_500__layer_21 mlp= 16942 exact=40.07%
51
+ delta=+2.20% per1k=+0.685%
52
+ add__topk_2000_minus_topk_500__layer_25 mlp= 18365 exact=38.33%
53
+ delta=+0.47% per1k=+0.101%
54
+ add__rel_0.001_minus_topk_500__layer_01 mlp= 15978 exact=40.87%
55
+ delta=+3.00% per1k=+1.334%
56
+ add__rel_0.001_minus_topk_500__layer_05 mlp= 20396 exact=47.00%
57
+ delta=+9.13% per1k=+1.370%
58
+ add__rel_0.001_minus_topk_500__layer_09 mlp= 22473 exact=42.00%
59
+ delta=+4.13% per1k=+0.473%
60
+ add__rel_0.001_minus_topk_500__layer_13 mlp= 22197 exact=42.00%
61
+ delta=+4.13% per1k=+0.488%
62
+ add__rel_0.001_minus_topk_500__layer_17 mlp= 22462 exact=41.67%
63
+ delta=+3.80% per1k=+0.435%
64
+ add__rel_0.001_minus_topk_500__layer_21 mlp= 22117 exact=41.20%
65
+ delta=+3.33% per1k=+0.397%
66
+ add__rel_0.001_minus_topk_500__layer_25 mlp= 19524 exact=38.47%
67
+ delta=+0.60% per1k=+0.104%
68
+ [done] rank-only -> results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_16/direct_group_rank_shard1.json
circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_16_direct_rank_shard2.log ADDED
@@ -0,0 +1,68 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ [records] results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_16/position_adacs positions=hundreds,tens,ones
2
+ records=1,500 skipped=0
3
+ [model] results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_16/lora_r16_beta005_merged device=cuda dtype=bfloat16 batch=256
4
+
5
+ topk_100 mlp= 2099 exact=12.13%
6
+ rel_0.05 mlp= 3042 exact=21.13%
7
+ topk_500 mlp= 13729 exact=37.87%
8
+ rel_0.01 mlp= 25570 exact=35.13%
9
+ topk_2000 mlp= 55061 exact=29.00%
10
+ rel_0.001 mlp= 224944 exact=98.13%
11
+ [rank] todo=28 shard=2/4
12
+ drop__topk_500__layer_02 mlp= 13640 exact=29.00%
13
+ delta=-8.87% per1k=-99.625%
14
+ drop__topk_500__layer_06 mlp= 13589 exact=39.47%
15
+ delta=+1.60% per1k=+11.429%
16
+ drop__topk_500__layer_10 mlp= 13650 exact=43.27%
17
+ delta=+5.40% per1k=+68.354%
18
+ drop__topk_500__layer_14 mlp= 13589 exact=40.93%
19
+ delta=+3.07% per1k=+21.905%
20
+ drop__topk_500__layer_18 mlp= 13568 exact=33.13%
21
+ delta=-4.73% per1k=-29.400%
22
+ drop__topk_500__layer_22 mlp= 12680 exact=5.07%
23
+ delta=-32.80% per1k=-31.268%
24
+ drop__topk_500__layer_26 mlp= 12588 exact=37.33%
25
+ delta=-0.53% per1k=-0.467%
26
+ add__rel_0.01_minus_topk_500__layer_02 mlp= 13800 exact=38.27%
27
+ delta=+0.40% per1k=+5.634%
28
+ add__rel_0.01_minus_topk_500__layer_06 mlp= 13922 exact=39.00%
29
+ delta=+1.13% per1k=+5.872%
30
+ add__rel_0.01_minus_topk_500__layer_10 mlp= 13888 exact=39.33%
31
+ delta=+1.47% per1k=+9.224%
32
+ add__rel_0.01_minus_topk_500__layer_14 mlp= 13941 exact=37.13%
33
+ delta=-0.73% per1k=-3.459%
34
+ add__rel_0.01_minus_topk_500__layer_18 mlp= 14051 exact=38.80%
35
+ delta=+0.93% per1k=+2.899%
36
+ add__rel_0.01_minus_topk_500__layer_22 mlp= 14759 exact=38.20%
37
+ delta=+0.33% per1k=+0.324%
38
+ add__rel_0.01_minus_topk_500__layer_26 mlp= 14634 exact=37.87%
39
+ delta=+0.00% per1k=+0.000%
40
+ add__topk_2000_minus_topk_500__layer_02 mlp= 13990 exact=40.40%
41
+ delta=+2.53% per1k=+9.706%
42
+ add__topk_2000_minus_topk_500__layer_06 mlp= 14361 exact=39.53%
43
+ delta=+1.67% per1k=+2.637%
44
+ add__topk_2000_minus_topk_500__layer_10 mlp= 14217 exact=38.13%
45
+ delta=+0.27% per1k=+0.546%
46
+ add__topk_2000_minus_topk_500__layer_14 mlp= 14394 exact=37.27%
47
+ delta=-0.60% per1k=-0.902%
48
+ add__topk_2000_minus_topk_500__layer_18 mlp= 14681 exact=38.67%
49
+ delta=+0.80% per1k=+0.840%
50
+ add__topk_2000_minus_topk_500__layer_22 mlp= 18448 exact=38.80%
51
+ delta=+0.93% per1k=+0.198%
52
+ add__topk_2000_minus_topk_500__layer_26 mlp= 17305 exact=38.00%
53
+ delta=+0.13% per1k=+0.037%
54
+ add__rel_0.001_minus_topk_500__layer_02 mlp= 17486 exact=44.27%
55
+ delta=+6.40% per1k=+1.703%
56
+ add__rel_0.001_minus_topk_500__layer_06 mlp= 22493 exact=52.27%
57
+ delta=+14.40% per1k=+1.643%
58
+ add__rel_0.001_minus_topk_500__layer_10 mlp= 22392 exact=43.20%
59
+ delta=+5.33% per1k=+0.616%
60
+ add__rel_0.001_minus_topk_500__layer_14 mlp= 22276 exact=42.00%
61
+ delta=+4.13% per1k=+0.484%
62
+ add__rel_0.001_minus_topk_500__layer_18 mlp= 22376 exact=39.93%
63
+ delta=+2.07% per1k=+0.239%
64
+ add__rel_0.001_minus_topk_500__layer_22 mlp= 21639 exact=39.73%
65
+ delta=+1.87% per1k=+0.236%
66
+ add__rel_0.001_minus_topk_500__layer_26 mlp= 21545 exact=38.20%
67
+ delta=+0.33% per1k=+0.043%
68
+ [done] rank-only -> results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_16/direct_group_rank_shard2.json
circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_16_direct_rank_shard3.log ADDED
@@ -0,0 +1,68 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ [records] results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_16/position_adacs positions=hundreds,tens,ones
2
+ records=1,500 skipped=0
3
+ [model] results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_16/lora_r16_beta005_merged device=cuda dtype=bfloat16 batch=256
4
+
5
+ topk_100 mlp= 2099 exact=12.13%
6
+ rel_0.05 mlp= 3042 exact=21.13%
7
+ topk_500 mlp= 13729 exact=37.87%
8
+ rel_0.01 mlp= 25570 exact=35.13%
9
+ topk_2000 mlp= 55061 exact=29.00%
10
+ rel_0.001 mlp= 224944 exact=98.13%
11
+ [rank] todo=28 shard=3/4
12
+ drop__topk_500__layer_03 mlp= 13662 exact=29.07%
13
+ delta=-8.80% per1k=-131.343%
14
+ drop__topk_500__layer_07 mlp= 13599 exact=36.87%
15
+ delta=-1.00% per1k=-7.692%
16
+ drop__topk_500__layer_11 mlp= 13647 exact=37.93%
17
+ delta=+0.07% per1k=+0.813%
18
+ drop__topk_500__layer_15 mlp= 13633 exact=40.33%
19
+ delta=+2.47% per1k=+25.694%
20
+ drop__topk_500__layer_19 mlp= 13416 exact=26.20%
21
+ delta=-11.67% per1k=-37.274%
22
+ drop__topk_500__layer_23 mlp= 12196 exact=5.93%
23
+ delta=-31.93% per1k=-20.831%
24
+ drop__topk_500__layer_27 mlp= 13220 exact=37.27%
25
+ delta=-0.60% per1k=-1.179%
26
+ add__rel_0.01_minus_topk_500__layer_03 mlp= 13794 exact=39.87%
27
+ delta=+2.00% per1k=+30.769%
28
+ add__rel_0.01_minus_topk_500__layer_07 mlp= 13949 exact=40.07%
29
+ delta=+2.20% per1k=+10.000%
30
+ add__rel_0.01_minus_topk_500__layer_11 mlp= 13869 exact=38.27%
31
+ delta=+0.40% per1k=+2.857%
32
+ add__rel_0.01_minus_topk_500__layer_15 mlp= 13935 exact=38.00%
33
+ delta=+0.13% per1k=+0.647%
34
+ add__rel_0.01_minus_topk_500__layer_19 mlp= 14199 exact=38.07%
35
+ delta=+0.20% per1k=+0.426%
36
+ add__rel_0.01_minus_topk_500__layer_23 mlp= 14960 exact=38.13%
37
+ delta=+0.27% per1k=+0.217%
38
+ add__rel_0.01_minus_topk_500__layer_27 mlp= 14118 exact=37.87%
39
+ delta=+0.00% per1k=+0.000%
40
+ add__topk_2000_minus_topk_500__layer_03 mlp= 13895 exact=38.87%
41
+ delta=+1.00% per1k=+6.024%
42
+ add__topk_2000_minus_topk_500__layer_07 mlp= 14430 exact=41.60%
43
+ delta=+3.73% per1k=+5.326%
44
+ add__topk_2000_minus_topk_500__layer_11 mlp= 14170 exact=37.20%
45
+ delta=-0.67% per1k=-1.512%
46
+ add__topk_2000_minus_topk_500__layer_15 mlp= 14325 exact=38.13%
47
+ delta=+0.27% per1k=+0.447%
48
+ add__topk_2000_minus_topk_500__layer_19 mlp= 15168 exact=37.67%
49
+ delta=-0.20% per1k=-0.139%
50
+ add__topk_2000_minus_topk_500__layer_23 mlp= 18628 exact=38.67%
51
+ delta=+0.80% per1k=+0.163%
52
+ add__topk_2000_minus_topk_500__layer_27 mlp= 15301 exact=38.13%
53
+ delta=+0.27% per1k=+0.170%
54
+ add__rel_0.001_minus_topk_500__layer_03 mlp= 17439 exact=42.60%
55
+ delta=+4.73% per1k=+1.276%
56
+ add__rel_0.001_minus_topk_500__layer_07 mlp= 22471 exact=48.80%
57
+ delta=+10.93% per1k=+1.251%
58
+ add__rel_0.001_minus_topk_500__layer_11 mlp= 22274 exact=40.07%
59
+ delta=+2.20% per1k=+0.257%
60
+ add__rel_0.001_minus_topk_500__layer_15 mlp= 22188 exact=43.60%
61
+ delta=+5.73% per1k=+0.678%
62
+ add__rel_0.001_minus_topk_500__layer_19 mlp= 22315 exact=41.13%
63
+ delta=+3.27% per1k=+0.380%
64
+ add__rel_0.001_minus_topk_500__layer_23 mlp= 21156 exact=38.47%
65
+ delta=+0.60% per1k=+0.081%
66
+ add__rel_0.001_minus_topk_500__layer_27 mlp= 21644 exact=37.87%
67
+ delta=+0.00% per1k=+0.000%
68
+ [done] rank-only -> results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_16/direct_group_rank_shard3.json
circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_16_fresh_build.log ADDED
@@ -0,0 +1,7 @@
 
 
 
 
 
 
 
 
1
+ [dataset] data/qwen25_math_1p5b_rank_ladder_r16_correct_addition.json
2
+ [pairing] loading tokenizer for prompt-length matching
3
+ enriched examples=7,971
4
+ hundreds pairs= 3160 buckets= 79
5
+ tens pairs= 4000 buckets= 20
6
+ ones pairs= 4000 buckets= 18
7
+ [done] build-only artifacts -> results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_16/fresh_pairs_seed123
circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_16_fresh_target90_eval.log ADDED
@@ -0,0 +1,6 @@
 
 
 
 
 
 
 
1
+ [records] results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_16/fresh_pairs_seed123
2
+ records=1,500 skipped=0
3
+ [model] results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_16/lora_r16_beta005_merged device=cuda dtype=bfloat16 batch=256
4
+
5
+ target90 mlp= 207027 exact=90.80%
6
+ [done] results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_16/fresh_target90_eval.json
circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_16_fresh_target90_generation.log ADDED
@@ -0,0 +1,4 @@
 
 
 
 
 
1
+ [records] 1,500 skipped=0
2
+
3
+ target90 mlp= 207027 gen_exact=90.67%
4
+ [done] results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_16/fresh_target90_generation.json
circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_16_merge.log ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ Warning: You are sending unauthenticated requests to the HF Hub. Please set a HF_TOKEN to enable higher rate limits and faster downloads.
2
+ [model] Qwen/Qwen2.5-Math-1.5B adapter=results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_16/lora_r16_beta005/adapter device=cuda
3
+
4
+ scale=0.45 kl=0.893586 acc=99.98% correct=8098/8100
5
+ scale=0.55 kl=0.965352 acc=99.94% correct=8095/8100
6
+ scale=0.75 kl=1.08896 acc=99.98% correct=8098/8100
7
+ scale=1 kl=1.23818 acc=99.85% correct=8088/8100
8
+
9
+ [done] selected scale=0.45 rule=lowest_kl_at_or_above_min_accuracy merged=results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_16/lora_r16_beta005_merged
circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_16_position_trace.log ADDED
@@ -0,0 +1,64 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ [worker gpu=0 positions=hundreds]
2
+ [dataset] data/qwen25_math_1p5b_rank_ladder_r16_correct_addition.json
3
+ [pairing] loading tokenizer for prompt-length matching
4
+ enriched examples=7,971
5
+ hundreds pairs= 3160 buckets= 79
6
+ [model] results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_16/lora_r16_beta005_merged
7
+ device=cuda dtype=bfloat16 batch_size=256
8
+
9
+ Loading weights: 0%| | 0/338 [00:00<?, ?it/s]
10
+ Loading weights: 100%|██████████| 338/338 [00:00<00:00, 6288.37it/s]
11
+ layers=28 d_ffn=8960 heads=12
12
+ [hundreds] attr_pairs=2660 test_pairs=500
13
+ hundreds positive mlp= 248182 zero=100.00% cf=100.00%
14
+ hundreds rel_0.001 mlp= 167002 zero=100.00% cf=100.00%
15
+ hundreds rel_0.01 mlp= 9008 zero=88.20% cf=77.00%
16
+ hundreds rel_0.05 mlp= 1174 zero=23.40% cf=4.80%
17
+ hundreds topk_100 mlp= 867 zero=4.60% cf=2.80%
18
+ hundreds topk_500 mlp= 5739 zero=82.80% cf=70.00%
19
+ hundreds topk_2000 mlp= 25499 zero=96.40% cf=85.00%
20
+ [done] artifacts -> results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_16/position_adacs/_workers/gpu0
21
+
22
+ [worker gpu=1 positions=tens]
23
+ [dataset] data/qwen25_math_1p5b_rank_ladder_r16_correct_addition.json
24
+ [pairing] loading tokenizer for prompt-length matching
25
+ enriched examples=7,971
26
+ tens pairs= 4000 buckets= 20
27
+ [model] results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_16/lora_r16_beta005_merged
28
+ device=cuda dtype=bfloat16 batch_size=256
29
+
30
+ Loading weights: 0%| | 0/338 [00:00<?, ?it/s]
31
+ Loading weights: 100%|██████████| 338/338 [00:00<00:00, 7374.43it/s]
32
+ layers=28 d_ffn=8960 heads=12
33
+ [tens] attr_pairs=3500 test_pairs=500
34
+ tens positive mlp= 250345 zero=100.00% cf=100.00%
35
+ tens rel_0.001 mlp= 200693 zero=100.00% cf=100.00%
36
+ tens rel_0.01 mlp= 15895 zero=21.40% cf=74.00%
37
+ tens rel_0.05 mlp= 2048 zero=8.60% cf=31.00%
38
+ tens topk_100 mlp= 1562 zero=11.20% cf=19.00%
39
+ tens topk_500 mlp= 9811 zero=14.00% cf=77.40%
40
+ tens topk_2000 mlp= 41986 zero=31.20% cf=65.20%
41
+ [done] artifacts -> results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_16/position_adacs/_workers/gpu1
42
+
43
+ [worker gpu=2 positions=ones]
44
+ [dataset] data/qwen25_math_1p5b_rank_ladder_r16_correct_addition.json
45
+ [pairing] loading tokenizer for prompt-length matching
46
+ enriched examples=7,971
47
+ ones pairs= 4000 buckets= 18
48
+ [model] results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_16/lora_r16_beta005_merged
49
+ device=cuda dtype=bfloat16 batch_size=256
50
+
51
+ Loading weights: 0%| | 0/338 [00:00<?, ?it/s]
52
+ Loading weights: 100%|██████████| 338/338 [00:00<00:00, 7279.24it/s]
53
+ layers=28 d_ffn=8960 heads=12
54
+ [ones] attr_pairs=3500 test_pairs=500
55
+ ones positive mlp= 250880 zero=98.00% cf=98.00%
56
+ ones rel_0.001 mlp= 222542 zero=97.80% cf=98.00%
57
+ ones rel_0.01 mlp= 19522 zero=24.40% cf=49.40%
58
+ ones rel_0.05 mlp= 2367 zero=8.20% cf=38.00%
59
+ ones topk_100 mlp= 1592 zero=9.00% cf=24.00%
60
+ ones topk_500 mlp= 8714 zero=18.60% cf=59.40%
61
+ ones topk_2000 mlp= 42957 zero=43.80% cf=54.80%
62
+ [done] artifacts -> results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_16/position_adacs/_workers/gpu2
63
+
64
+ [done] merged multi-gpu artifacts -> results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_16/position_adacs
circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_16_prep_stage.log ADDED
@@ -0,0 +1,5 @@
 
 
 
 
 
 
1
+ [rank 16] prep start on gpu 0
2
+ [rank 16] train cached
3
+ [rank 16] merge cached
4
+ [rank 16] correct dataset cached
5
+ [rank 16] prep done
circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_16_target90_search.log ADDED
@@ -0,0 +1,271 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+
2
+ topk_100 mlp= 2099 exact=12.13%
3
+ rel_0.05 mlp= 3042 exact=21.13%
4
+ topk_500 mlp= 13729 exact=37.87%
5
+ rel_0.01 mlp= 25570 exact=35.13%
6
+ topk_2000 mlp= 55061 exact=29.00%
7
+ rel_0.001 mlp= 224944 exact=98.13%
8
+ add_step_1_add__rel_0.001_minus_topk_500__layer_06 mlp= 22493 exact=52.27%
9
+ add_step_1_add__rel_0.001_minus_topk_500__layer_08 mlp= 22496 exact=50.27%
10
+ add_step_1_add__rel_0.001_minus_topk_500__layer_07 mlp= 22471 exact=48.80%
11
+ add_step_1_add__rel_0.001_minus_topk_500__layer_05 mlp= 20396 exact=47.00%
12
+ add_step_1_add__rel_0.001_minus_topk_500__layer_04 mlp= 21626 exact=44.87%
13
+ add_step_1_add__rel_0.001_minus_topk_500__layer_02 mlp= 17486 exact=44.27%
14
+ add_step_1_add__rel_0.001_minus_topk_500__layer_00 mlp= 19999 exact=43.93%
15
+ add_step_1_add__rel_0.001_minus_topk_500__layer_12 mlp= 22162 exact=43.93%
16
+ add_step_1_add__rel_0.001_minus_topk_500__layer_15 mlp= 22188 exact=43.60%
17
+ add_step_1_add__rel_0.001_minus_topk_500__layer_10 mlp= 22392 exact=43.20%
18
+ add_step_1_add__rel_0.001_minus_topk_500__layer_03 mlp= 17439 exact=42.60%
19
+ add_step_1_add__rel_0.001_minus_topk_500__layer_13 mlp= 22197 exact=42.00%
20
+ add_step_1_add__rel_0.001_minus_topk_500__layer_14 mlp= 22276 exact=42.00%
21
+ add_step_1_add__rel_0.001_minus_topk_500__layer_09 mlp= 22473 exact=42.00%
22
+ add_step_1_add__rel_0.001_minus_topk_500__layer_16 mlp= 22362 exact=41.93%
23
+ add_step_1_add__rel_0.001_minus_topk_500__layer_17 mlp= 22462 exact=41.67%
24
+ add_step_1_add__topk_2000_minus_topk_500__layer_07 mlp= 14430 exact=41.60%
25
+ add_step_1_add__rel_0.001_minus_topk_500__layer_21 mlp= 22117 exact=41.20%
26
+ add_step_1_add__rel_0.001_minus_topk_500__layer_19 mlp= 22315 exact=41.13%
27
+ add_step_1_add__rel_0.01_minus_topk_500__layer_21 mlp= 14505 exact=40.87%
28
+ add_step_1_add__rel_0.001_minus_topk_500__layer_01 mlp= 15978 exact=40.87%
29
+ add_step_1_add__topk_2000_minus_topk_500__layer_00 mlp= 14044 exact=40.47%
30
+ add_step_1_add__topk_2000_minus_topk_500__layer_02 mlp= 13990 exact=40.40%
31
+ add_step_1_add__rel_0.01_minus_topk_500__layer_16 mlp= 13968 exact=40.33%
32
+ add_step_1_add__topk_2000_minus_topk_500__layer_04 mlp= 14179 exact=40.27%
33
+ add_step_1_add__rel_0.01_minus_topk_500__layer_07 mlp= 13949 exact=40.07%
34
+ add_step_1_add__topk_2000_minus_topk_500__layer_21 mlp= 16942 exact=40.07%
35
+ add_step_1_add__rel_0.001_minus_topk_500__layer_11 mlp= 22274 exact=40.07%
36
+ add_step_1_add__rel_0.001_minus_topk_500__layer_18 mlp= 22376 exact=39.93%
37
+ add_step_1_add__rel_0.01_minus_topk_500__layer_03 mlp= 13794 exact=39.87%
38
+ add_step_1_add__rel_0.001_minus_topk_500__layer_22 mlp= 21639 exact=39.73%
39
+ add_step_1_add__rel_0.01_minus_topk_500__layer_00 mlp= 13794 exact=39.60%
40
+ add_step_1_add__topk_2000_minus_topk_500__layer_06 mlp= 14361 exact=39.53%
41
+ add_step_1_add__rel_0.01_minus_topk_500__layer_10 mlp= 13888 exact=39.33%
42
+ add_step_1_add__topk_2000_minus_topk_500__layer_05 mlp= 14091 exact=39.27%
43
+ add_step_1_add__rel_0.01_minus_topk_500__layer_01 mlp= 13791 exact=39.00%
44
+ add_step_1_add__rel_0.01_minus_topk_500__layer_06 mlp= 13922 exact=39.00%
45
+ add_step_1_add__topk_2000_minus_topk_500__layer_01 mlp= 13919 exact=38.93%
46
+ add_step_1_add__topk_2000_minus_topk_500__layer_03 mlp= 13895 exact=38.87%
47
+ add_step_1_add__topk_2000_minus_topk_500__layer_17 mlp= 14544 exact=38.87%
48
+ add_step_1_add__rel_0.01_minus_topk_500__layer_18 mlp= 14051 exact=38.80%
49
+ add_step_1_add__topk_2000_minus_topk_500__layer_22 mlp= 18448 exact=38.80%
50
+ add_step_1_add__rel_0.01_minus_topk_500__layer_04 mlp= 13874 exact=38.67%
51
+ add_step_1_add__topk_2000_minus_topk_500__layer_18 mlp= 14681 exact=38.67%
52
+ add_step_1_add__topk_2000_minus_topk_500__layer_23 mlp= 18628 exact=38.67%
53
+ add_step_1_add__rel_0.001_minus_topk_500__layer_25 mlp= 19524 exact=38.47%
54
+ add_step_1_add__rel_0.001_minus_topk_500__layer_23 mlp= 21156 exact=38.47%
55
+ add_step_1_add__rel_0.01_minus_topk_500__layer_25 mlp= 15462 exact=38.40%
56
+ add_step_1_accepted mlp= 22493 exact=52.27%
57
+ add_step_2_add__rel_0.001_minus_topk_500__layer_08 mlp= 31260 exact=57.40%
58
+ add_step_2_add__rel_0.001_minus_topk_500__layer_07 mlp= 31235 exact=56.20%
59
+ add_step_2_add__rel_0.001_minus_topk_500__layer_05 mlp= 29160 exact=61.00%
60
+ add_step_2_add__rel_0.001_minus_topk_500__layer_04 mlp= 30390 exact=65.40%
61
+ add_step_2_add__rel_0.001_minus_topk_500__layer_02 mlp= 26250 exact=55.87%
62
+ add_step_2_add__rel_0.001_minus_topk_500__layer_00 mlp= 28763 exact=57.27%
63
+ add_step_2_add__rel_0.001_minus_topk_500__layer_12 mlp= 30926 exact=57.33%
64
+ add_step_2_add__rel_0.001_minus_topk_500__layer_15 mlp= 30952 exact=59.80%
65
+ add_step_2_add__rel_0.001_minus_topk_500__layer_10 mlp= 31156 exact=59.40%
66
+ add_step_2_add__rel_0.001_minus_topk_500__layer_03 mlp= 26203 exact=57.07%
67
+ add_step_2_add__rel_0.001_minus_topk_500__layer_13 mlp= 30961 exact=53.27%
68
+ add_step_2_add__rel_0.001_minus_topk_500__layer_14 mlp= 31040 exact=56.53%
69
+ add_step_2_add__rel_0.001_minus_topk_500__layer_09 mlp= 31237 exact=59.47%
70
+ add_step_2_add__rel_0.001_minus_topk_500__layer_16 mlp= 31126 exact=50.87%
71
+ add_step_2_add__rel_0.001_minus_topk_500__layer_17 mlp= 31226 exact=57.20%
72
+ add_step_2_add__topk_2000_minus_topk_500__layer_07 mlp= 23194 exact=53.87%
73
+ add_step_2_add__rel_0.001_minus_topk_500__layer_21 mlp= 30881 exact=56.07%
74
+ add_step_2_add__rel_0.001_minus_topk_500__layer_19 mlp= 31079 exact=56.13%
75
+ add_step_2_add__rel_0.01_minus_topk_500__layer_21 mlp= 23269 exact=55.40%
76
+ add_step_2_add__rel_0.001_minus_topk_500__layer_01 mlp= 24742 exact=54.67%
77
+ add_step_2_add__topk_2000_minus_topk_500__layer_00 mlp= 22808 exact=55.27%
78
+ add_step_2_add__topk_2000_minus_topk_500__layer_02 mlp= 22754 exact=53.40%
79
+ add_step_2_add__rel_0.01_minus_topk_500__layer_16 mlp= 22732 exact=51.93%
80
+ add_step_2_add__topk_2000_minus_topk_500__layer_04 mlp= 22943 exact=52.60%
81
+ add_step_2_add__rel_0.01_minus_topk_500__layer_07 mlp= 22713 exact=51.87%
82
+ add_step_2_add__topk_2000_minus_topk_500__layer_21 mlp= 25706 exact=54.73%
83
+ add_step_2_add__rel_0.001_minus_topk_500__layer_11 mlp= 31038 exact=58.27%
84
+ add_step_2_add__rel_0.001_minus_topk_500__layer_18 mlp= 31140 exact=55.00%
85
+ add_step_2_add__rel_0.01_minus_topk_500__layer_03 mlp= 22558 exact=53.80%
86
+ add_step_2_add__rel_0.001_minus_topk_500__layer_22 mlp= 30403 exact=53.40%
87
+ add_step_2_add__rel_0.01_minus_topk_500__layer_00 mlp= 22558 exact=52.13%
88
+ add_step_2_add__rel_0.01_minus_topk_500__layer_10 mlp= 22652 exact=51.20%
89
+ add_step_2_add__topk_2000_minus_topk_500__layer_05 mlp= 22855 exact=51.80%
90
+ add_step_2_add__rel_0.01_minus_topk_500__layer_01 mlp= 22555 exact=52.67%
91
+ add_step_2_add__topk_2000_minus_topk_500__layer_01 mlp= 22683 exact=52.33%
92
+ add_step_2_add__topk_2000_minus_topk_500__layer_03 mlp= 22659 exact=53.00%
93
+ add_step_2_add__topk_2000_minus_topk_500__layer_17 mlp= 23308 exact=51.27%
94
+ add_step_2_add__rel_0.01_minus_topk_500__layer_18 mlp= 22815 exact=51.87%
95
+ add_step_2_add__topk_2000_minus_topk_500__layer_22 mlp= 27212 exact=52.73%
96
+ add_step_2_add__rel_0.01_minus_topk_500__layer_04 mlp= 22638 exact=52.80%
97
+ add_step_2_add__topk_2000_minus_topk_500__layer_18 mlp= 23445 exact=51.20%
98
+ add_step_2_add__topk_2000_minus_topk_500__layer_23 mlp= 27392 exact=52.47%
99
+ add_step_2_add__rel_0.001_minus_topk_500__layer_25 mlp= 28288 exact=52.33%
100
+ add_step_2_add__rel_0.001_minus_topk_500__layer_23 mlp= 29920 exact=52.20%
101
+ add_step_2_add__rel_0.01_minus_topk_500__layer_25 mlp= 24226 exact=51.67%
102
+ add_step_2_accepted mlp= 30390 exact=65.40%
103
+ add_step_3_add__rel_0.001_minus_topk_500__layer_08 mlp= 39157 exact=66.53%
104
+ add_step_3_add__rel_0.001_minus_topk_500__layer_07 mlp= 39132 exact=71.53%
105
+ add_step_3_add__rel_0.001_minus_topk_500__layer_05 mlp= 37057 exact=66.07%
106
+ add_step_3_add__rel_0.001_minus_topk_500__layer_02 mlp= 34147 exact=64.87%
107
+ add_step_3_add__rel_0.001_minus_topk_500__layer_00 mlp= 36660 exact=68.67%
108
+ add_step_3_add__rel_0.001_minus_topk_500__layer_12 mlp= 38823 exact=68.33%
109
+ add_step_3_add__rel_0.001_minus_topk_500__layer_15 mlp= 38849 exact=71.13%
110
+ add_step_3_add__rel_0.001_minus_topk_500__layer_10 mlp= 39053 exact=70.20%
111
+ add_step_3_add__rel_0.001_minus_topk_500__layer_03 mlp= 34100 exact=67.33%
112
+ add_step_3_add__rel_0.001_minus_topk_500__layer_13 mlp= 38858 exact=68.13%
113
+ add_step_3_add__rel_0.001_minus_topk_500__layer_14 mlp= 38937 exact=70.47%
114
+ add_step_3_add__rel_0.001_minus_topk_500__layer_09 mlp= 39134 exact=67.47%
115
+ add_step_3_add__rel_0.001_minus_topk_500__layer_16 mlp= 39023 exact=68.27%
116
+ add_step_3_add__rel_0.001_minus_topk_500__layer_17 mlp= 39123 exact=68.87%
117
+ add_step_3_add__topk_2000_minus_topk_500__layer_07 mlp= 31091 exact=67.00%
118
+ add_step_3_add__rel_0.001_minus_topk_500__layer_21 mlp= 38778 exact=69.40%
119
+ add_step_3_add__rel_0.001_minus_topk_500__layer_19 mlp= 38976 exact=69.07%
120
+ add_step_3_add__rel_0.01_minus_topk_500__layer_21 mlp= 31166 exact=68.07%
121
+ add_step_3_add__rel_0.001_minus_topk_500__layer_01 mlp= 32639 exact=63.73%
122
+ add_step_3_add__topk_2000_minus_topk_500__layer_00 mlp= 30705 exact=66.27%
123
+ add_step_3_add__topk_2000_minus_topk_500__layer_02 mlp= 30651 exact=64.00%
124
+ add_step_3_add__rel_0.01_minus_topk_500__layer_16 mlp= 30629 exact=65.47%
125
+ add_step_3_add__rel_0.01_minus_topk_500__layer_07 mlp= 30610 exact=66.00%
126
+ add_step_3_add__topk_2000_minus_topk_500__layer_21 mlp= 33603 exact=68.60%
127
+ add_step_3_add__rel_0.001_minus_topk_500__layer_11 mlp= 38935 exact=74.80%
128
+ add_step_3_add__rel_0.001_minus_topk_500__layer_18 mlp= 39037 exact=67.13%
129
+ add_step_3_add__rel_0.01_minus_topk_500__layer_03 mlp= 30455 exact=62.93%
130
+ add_step_3_add__rel_0.001_minus_topk_500__layer_22 mlp= 38300 exact=66.60%
131
+ add_step_3_add__rel_0.01_minus_topk_500__layer_00 mlp= 30455 exact=65.47%
132
+ add_step_3_add__rel_0.01_minus_topk_500__layer_10 mlp= 30549 exact=65.93%
133
+ add_step_3_add__topk_2000_minus_topk_500__layer_05 mlp= 30752 exact=65.40%
134
+ add_step_3_add__rel_0.01_minus_topk_500__layer_01 mlp= 30452 exact=65.53%
135
+ add_step_3_add__topk_2000_minus_topk_500__layer_01 mlp= 30580 exact=65.40%
136
+ add_step_3_add__topk_2000_minus_topk_500__layer_03 mlp= 30556 exact=63.47%
137
+ add_step_3_add__topk_2000_minus_topk_500__layer_17 mlp= 31205 exact=65.87%
138
+ add_step_3_add__rel_0.01_minus_topk_500__layer_18 mlp= 30712 exact=64.27%
139
+ add_step_3_add__topk_2000_minus_topk_500__layer_22 mlp= 35109 exact=65.80%
140
+ add_step_3_add__topk_2000_minus_topk_500__layer_18 mlp= 31342 exact=64.60%
141
+ add_step_3_add__topk_2000_minus_topk_500__layer_23 mlp= 35289 exact=65.20%
142
+ add_step_3_add__rel_0.001_minus_topk_500__layer_25 mlp= 36185 exact=65.07%
143
+ add_step_3_add__rel_0.001_minus_topk_500__layer_23 mlp= 37817 exact=65.07%
144
+ add_step_3_add__rel_0.01_minus_topk_500__layer_25 mlp= 32123 exact=65.33%
145
+ add_step_3_accepted mlp= 38935 exact=74.80%
146
+ add_step_4_add__rel_0.001_minus_topk_500__layer_08 mlp= 47702 exact=72.47%
147
+ add_step_4_add__rel_0.001_minus_topk_500__layer_07 mlp= 47677 exact=76.93%
148
+ add_step_4_add__rel_0.001_minus_topk_500__layer_05 mlp= 45602 exact=76.20%
149
+ add_step_4_add__rel_0.001_minus_topk_500__layer_02 mlp= 42692 exact=73.47%
150
+ add_step_4_add__rel_0.001_minus_topk_500__layer_00 mlp= 45205 exact=77.40%
151
+ add_step_4_add__rel_0.001_minus_topk_500__layer_12 mlp= 47368 exact=76.27%
152
+ add_step_4_add__rel_0.001_minus_topk_500__layer_15 mlp= 47394 exact=80.47%
153
+ add_step_4_add__rel_0.001_minus_topk_500__layer_10 mlp= 47598 exact=73.80%
154
+ add_step_4_add__rel_0.001_minus_topk_500__layer_03 mlp= 42645 exact=72.93%
155
+ add_step_4_add__rel_0.001_minus_topk_500__layer_13 mlp= 47403 exact=73.00%
156
+ add_step_4_add__rel_0.001_minus_topk_500__layer_14 mlp= 47482 exact=74.87%
157
+ add_step_4_add__rel_0.001_minus_topk_500__layer_09 mlp= 47679 exact=74.27%
158
+ add_step_4_add__rel_0.001_minus_topk_500__layer_16 mlp= 47568 exact=77.53%
159
+ add_step_4_add__rel_0.001_minus_topk_500__layer_17 mlp= 47668 exact=77.27%
160
+ add_step_4_add__topk_2000_minus_topk_500__layer_07 mlp= 39636 exact=76.73%
161
+ add_step_4_add__rel_0.001_minus_topk_500__layer_21 mlp= 47323 exact=77.87%
162
+ add_step_4_add__rel_0.001_minus_topk_500__layer_19 mlp= 47521 exact=78.67%
163
+ add_step_4_add__rel_0.01_minus_topk_500__layer_21 mlp= 39711 exact=76.07%
164
+ add_step_4_add__rel_0.001_minus_topk_500__layer_01 mlp= 41184 exact=72.73%
165
+ add_step_4_add__topk_2000_minus_topk_500__layer_00 mlp= 39250 exact=75.13%
166
+ add_step_4_add__topk_2000_minus_topk_500__layer_02 mlp= 39196 exact=73.53%
167
+ add_step_4_add__rel_0.01_minus_topk_500__layer_16 mlp= 39174 exact=76.27%
168
+ add_step_4_add__rel_0.01_minus_topk_500__layer_07 mlp= 39155 exact=78.00%
169
+ add_step_4_add__topk_2000_minus_topk_500__layer_21 mlp= 42148 exact=77.07%
170
+ add_step_4_add__rel_0.001_minus_topk_500__layer_18 mlp= 47582 exact=75.07%
171
+ add_step_4_add__rel_0.01_minus_topk_500__layer_03 mlp= 39000 exact=73.80%
172
+ add_step_4_add__rel_0.001_minus_topk_500__layer_22 mlp= 46845 exact=77.13%
173
+ add_step_4_add__rel_0.01_minus_topk_500__layer_00 mlp= 39000 exact=74.07%
174
+ add_step_4_add__rel_0.01_minus_topk_500__layer_10 mlp= 39094 exact=74.80%
175
+ add_step_4_add__topk_2000_minus_topk_500__layer_05 mlp= 39297 exact=76.47%
176
+ add_step_4_add__rel_0.01_minus_topk_500__layer_01 mlp= 38997 exact=74.20%
177
+ add_step_4_add__topk_2000_minus_topk_500__layer_01 mlp= 39125 exact=74.20%
178
+ add_step_4_add__topk_2000_minus_topk_500__layer_03 mlp= 39101 exact=73.13%
179
+ add_step_4_add__topk_2000_minus_topk_500__layer_17 mlp= 39750 exact=77.27%
180
+ add_step_4_add__rel_0.01_minus_topk_500__layer_18 mlp= 39257 exact=73.07%
181
+ add_step_4_add__topk_2000_minus_topk_500__layer_22 mlp= 43654 exact=76.33%
182
+ add_step_4_add__topk_2000_minus_topk_500__layer_18 mlp= 39887 exact=72.87%
183
+ add_step_4_add__topk_2000_minus_topk_500__layer_23 mlp= 43834 exact=75.20%
184
+ add_step_4_add__rel_0.001_minus_topk_500__layer_25 mlp= 44730 exact=75.00%
185
+ add_step_4_add__rel_0.001_minus_topk_500__layer_23 mlp= 46362 exact=75.20%
186
+ add_step_4_add__rel_0.01_minus_topk_500__layer_25 mlp= 40668 exact=75.07%
187
+ add_step_4_accepted mlp= 47394 exact=80.47%
188
+ baseline_rel_0.001_compress_1_drop_00 mlp= 218565 exact=30.27%
189
+ baseline_rel_0.001_compress_1_drop_01 mlp= 222613 exact=68.33%
190
+ baseline_rel_0.001_compress_1_drop_02 mlp= 221098 exact=93.73%
191
+ baseline_rel_0.001_compress_1_drop_03 mlp= 221167 exact=96.40%
192
+ baseline_rel_0.001_compress_1_drop_04 mlp= 216845 exact=82.00%
193
+ baseline_rel_0.001_compress_1_drop_05 mlp= 218148 exact=97.07%
194
+ baseline_rel_0.001_compress_1_drop_06 mlp= 216040 exact=96.33%
195
+ baseline_rel_0.001_compress_1_drop_07 mlp= 216072 exact=96.53%
196
+ baseline_rel_0.001_compress_1_drop_08 mlp= 216066 exact=99.27%
197
+ baseline_rel_0.001_compress_1_drop_09 mlp= 216100 exact=98.20%
198
+ baseline_rel_0.001_compress_1_drop_10 mlp= 216202 exact=96.60%
199
+ baseline_rel_0.001_compress_1_drop_11 mlp= 216317 exact=98.87%
200
+ baseline_rel_0.001_compress_1_drop_12 mlp= 216378 exact=97.53%
201
+ baseline_rel_0.001_compress_1_drop_13 mlp= 216354 exact=98.07%
202
+ baseline_rel_0.001_compress_1_drop_14 mlp= 216257 exact=96.00%
203
+ baseline_rel_0.001_compress_1_drop_15 mlp= 216389 exact=97.00%
204
+ baseline_rel_0.001_compress_1_drop_16 mlp= 216185 exact=99.00%
205
+ baseline_rel_0.001_compress_1_drop_17 mlp= 216093 exact=98.73%
206
+ baseline_rel_0.001_compress_1_drop_18 mlp= 216136 exact=95.13%
207
+ baseline_rel_0.001_compress_1_drop_19 mlp= 216045 exact=98.73%
208
+ baseline_rel_0.001_compress_1_drop_20 mlp= 215987 exact=98.00%
209
+ baseline_rel_0.001_compress_1_drop_21 mlp= 215987 exact=71.73%
210
+ baseline_rel_0.001_compress_1_drop_22 mlp= 215985 exact=11.80%
211
+ baseline_rel_0.001_compress_1_drop_23 mlp= 215984 exact=15.00%
212
+ baseline_rel_0.001_compress_1_drop_24 mlp= 215984 exact=93.00%
213
+ baseline_rel_0.001_compress_1_drop_25 mlp= 215984 exact=86.13%
214
+ baseline_rel_0.001_compress_1_drop_26 mlp= 215987 exact=97.87%
215
+ baseline_rel_0.001_compress_1_drop_27 mlp= 216520 exact=96.73%
216
+ baseline_rel_0.001_compress_1_accepted mlp= 215984 exact=93.00%
217
+ baseline_rel_0.001_compress_2_drop_00 mlp= 209605 exact=27.00%
218
+ baseline_rel_0.001_compress_2_drop_01 mlp= 213653 exact=63.53%
219
+ baseline_rel_0.001_compress_2_drop_02 mlp= 212138 exact=86.27%
220
+ baseline_rel_0.001_compress_2_drop_03 mlp= 212207 exact=88.40%
221
+ baseline_rel_0.001_compress_2_drop_04 mlp= 207885 exact=71.27%
222
+ baseline_rel_0.001_compress_2_drop_05 mlp= 209188 exact=91.13%
223
+ baseline_rel_0.001_compress_2_drop_06 mlp= 207080 exact=90.53%
224
+ baseline_rel_0.001_compress_2_drop_07 mlp= 207112 exact=90.93%
225
+ baseline_rel_0.001_compress_2_drop_08 mlp= 207106 exact=94.33%
226
+ baseline_rel_0.001_compress_2_drop_09 mlp= 207140 exact=92.87%
227
+ baseline_rel_0.001_compress_2_drop_10 mlp= 207242 exact=91.93%
228
+ baseline_rel_0.001_compress_2_drop_11 mlp= 207357 exact=92.13%
229
+ baseline_rel_0.001_compress_2_drop_12 mlp= 207418 exact=90.73%
230
+ baseline_rel_0.001_compress_2_drop_13 mlp= 207394 exact=92.87%
231
+ baseline_rel_0.001_compress_2_drop_14 mlp= 207297 exact=91.00%
232
+ baseline_rel_0.001_compress_2_drop_15 mlp= 207429 exact=91.40%
233
+ baseline_rel_0.001_compress_2_drop_16 mlp= 207225 exact=94.53%
234
+ baseline_rel_0.001_compress_2_drop_17 mlp= 207133 exact=93.27%
235
+ baseline_rel_0.001_compress_2_drop_18 mlp= 207176 exact=87.73%
236
+ baseline_rel_0.001_compress_2_drop_19 mlp= 207085 exact=89.07%
237
+ baseline_rel_0.001_compress_2_drop_20 mlp= 207027 exact=90.07%
238
+ baseline_rel_0.001_compress_2_drop_21 mlp= 207027 exact=63.27%
239
+ baseline_rel_0.001_compress_2_drop_22 mlp= 207025 exact=11.13%
240
+ baseline_rel_0.001_compress_2_drop_23 mlp= 207024 exact=8.00%
241
+ baseline_rel_0.001_compress_2_drop_25 mlp= 207024 exact=50.60%
242
+ baseline_rel_0.001_compress_2_drop_26 mlp= 207027 exact=87.87%
243
+ baseline_rel_0.001_compress_2_drop_27 mlp= 207560 exact=89.53%
244
+ baseline_rel_0.001_compress_2_accepted mlp= 207027 exact=90.07%
245
+ baseline_rel_0.001_compress_3_drop_00 mlp= 200648 exact=23.53%
246
+ baseline_rel_0.001_compress_3_drop_01 mlp= 204696 exact=61.40%
247
+ baseline_rel_0.001_compress_3_drop_02 mlp= 203181 exact=82.13%
248
+ baseline_rel_0.001_compress_3_drop_03 mlp= 203250 exact=81.87%
249
+ baseline_rel_0.001_compress_3_drop_04 mlp= 198928 exact=58.67%
250
+ baseline_rel_0.001_compress_3_drop_05 mlp= 200231 exact=88.20%
251
+ baseline_rel_0.001_compress_3_drop_06 mlp= 198123 exact=88.47%
252
+ baseline_rel_0.001_compress_3_drop_07 mlp= 198155 exact=87.53%
253
+ baseline_rel_0.001_compress_3_drop_08 mlp= 198149 exact=89.47%
254
+ baseline_rel_0.001_compress_3_drop_09 mlp= 198183 exact=89.13%
255
+ baseline_rel_0.001_compress_3_drop_10 mlp= 198285 exact=89.00%
256
+ baseline_rel_0.001_compress_3_drop_11 mlp= 198400 exact=86.00%
257
+ baseline_rel_0.001_compress_3_drop_12 mlp= 198461 exact=84.47%
258
+ baseline_rel_0.001_compress_3_drop_13 mlp= 198437 exact=89.20%
259
+ baseline_rel_0.001_compress_3_drop_14 mlp= 198340 exact=88.20%
260
+ baseline_rel_0.001_compress_3_drop_15 mlp= 198472 exact=86.67%
261
+ baseline_rel_0.001_compress_3_drop_16 mlp= 198268 exact=88.87%
262
+ baseline_rel_0.001_compress_3_drop_17 mlp= 198176 exact=86.67%
263
+ baseline_rel_0.001_compress_3_drop_18 mlp= 198219 exact=82.53%
264
+ baseline_rel_0.001_compress_3_drop_19 mlp= 198128 exact=77.33%
265
+ baseline_rel_0.001_compress_3_drop_21 mlp= 198070 exact=42.67%
266
+ baseline_rel_0.001_compress_3_drop_22 mlp= 198068 exact=7.60%
267
+ baseline_rel_0.001_compress_3_drop_23 mlp= 198067 exact=7.73%
268
+ baseline_rel_0.001_compress_3_drop_25 mlp= 198067 exact=50.60%
269
+ baseline_rel_0.001_compress_3_drop_26 mlp= 198070 exact=80.87%
270
+ baseline_rel_0.001_compress_3_drop_27 mlp= 198603 exact=83.33%
271
+ [done] results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_16/target90_search.json
circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_16_train.log ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ [model] Qwen/Qwen2.5-Math-1.5B device=cuda dtype=bfloat16
2
+ Warning: You are sending unauthenticated requests to the HF Hub. Please set a HF_TOKEN to enable higher rate limits and faster downloads.
3
+
4
+ trainable params: 18,464,768 || all params: 1,562,179,072 || trainable%: 1.1820
5
+ {"step": 1, "loss": 1.5390625, "ce": 1.5390625, "kl": -9.489059448242188e-05, "lr": 1.3333333333333333e-05, "elapsed_s": 0.812575101852417, "compact_exact": {"n": 8100, "correct": 7590, "accuracy": 0.937037037037037, "wrong_sample": [{"prompt": "11 + 13 =", "answer": "24", "pred": "35"}, {"prompt": "11 + 21 =", "answer": "32", "pred": "132"}, {"prompt": "11 + 31 =", "answer": "42", "pred": "142"}, {"prompt": "11 + 41 =", "answer": "52", "pred": "152"}, {"prompt": "11 + 47 =", "answer": "58", "pred": "158"}, {"prompt": "11 + 48 =", "answer": "59", "pred": "159"}, {"prompt": "11 + 49 =", "answer": "60", "pred": "160"}, {"prompt": "11 + 50 =", "answer": "61", "pred": "161"}, {"prompt": "11 + 51 =", "answer": "62", "pred": "162"}, {"prompt": "11 + 53 =", "answer": "64", "pred": "164"}, {"prompt": "11 + 57 =", "answer": "68", "pred": "168"}, {"prompt": "11 + 58 =", "answer": "69", "pred": "169"}, {"prompt": "11 + 59 =", "answer": "70", "pred": "170"}, {"prompt": "11 + 61 =", "answer": "72", "pred": "172"}, {"prompt": "11 + 62 =", "answer": "73", "pred": "173"}, {"prompt": "11 + 67 =", "answer": "78", "pred": "178"}, {"prompt": "11 + 68 =", "answer": "79", "pred": "179"}, {"prompt": "11 + 69 =", "answer": "80", "pred": "180"}, {"prompt": "11 + 70 =", "answer": "81", "pred": "181"}, {"prompt": "11 + 71 =", "answer": "82", "pred": "182"}, {"prompt": "11 + 75 =", "answer": "86", "pred": "186"}, {"prompt": "11 + 76 =", "answer": "87", "pred": "187"}, {"prompt": "11 + 78 =", "answer": "89", "pred": "189"}, {"prompt": "11 + 79 =", "answer": "90", "pred": "190"}, {"prompt": "11 + 80 =", "answer": "91", "pred": "191"}]}}
6
+ {"step": 100, "loss": 0.15492942116477273, "ce": 0.1362332816075797, "kl": 0.37419204519252586, "lr": 0.00015922352526649803, "elapsed_s": 69.5666401386261, "compact_exact": {"n": 8100, "correct": 7923, "accuracy": 0.9781481481481481, "wrong_sample": [{"prompt": "11 + 13 =", "answer": "24", "pred": "14"}, {"prompt": "11 + 16 =", "answer": "27", "pred": "17"}, {"prompt": "11 + 17 =", "answer": "28", "pred": "18"}, {"prompt": "11 + 18 =", "answer": "29", "pred": "19"}, {"prompt": "11 + 19 =", "answer": "30", "pred": "20"}, {"prompt": "11 + 20 =", "answer": "31", "pred": "21"}, {"prompt": "11 + 23 =", "answer": "34", "pred": "24"}, {"prompt": "11 + 24 =", "answer": "35", "pred": "25"}, {"prompt": "11 + 25 =", "answer": "36", "pred": "26"}, {"prompt": "11 + 26 =", "answer": "37", "pred": "27"}, {"prompt": "11 + 29 =", "answer": "40", "pred": "30"}, {"prompt": "11 + 49 =", "answer": "60", "pred": "50"}, {"prompt": "21 + 29 =", "answer": "50", "pred": "40"}, {"prompt": "21 + 39 =", "answer": "60", "pred": "50"}, {"prompt": "21 + 69 =", "answer": "90", "pred": "80"}, {"prompt": "22 + 13 =", "answer": "35", "pred": "15"}, {"prompt": "22 + 49 =", "answer": "71", "pred": "51"}, {"prompt": "24 + 71 =", "answer": "95", "pred": "115"}, {"prompt": "24 + 77 =", "answer": "101", "pred": "121"}, {"prompt": "24 + 78 =", "answer": "102", "pred": "122"}, {"prompt": "24 + 81 =", "answer": "105", "pred": "125"}, {"prompt": "28 + 72 =", "answer": "100", "pred": "110"}, {"prompt": "28 + 77 =", "answer": "105", "pred": "115"}, {"prompt": "29 + 71 =", "answer": "100", "pred": "110"}, {"prompt": "29 + 73 =", "answer": "102", "pred": "112"}]}}
7
+ {"step": 200, "loss": 0.03357421875, "ce": 0.016380691528320314, "kl": 0.343701171875, "lr": 5.484666416891109e-05, "elapsed_s": 126.36040806770325, "compact_exact": {"n": 8100, "correct": 8089, "accuracy": 0.998641975308642, "wrong_sample": [{"prompt": "13 + 79 =", "answer": "92", "pred": "922"}, {"prompt": "14 + 78 =", "answer": "92", "pred": "922"}, {"prompt": "18 + 74 =", "answer": "92", "pred": "922"}, {"prompt": "19 + 73 =", "answer": "92", "pred": "922"}, {"prompt": "58 + 42 =", "answer": "100", "pred": "90"}, {"prompt": "59 + 41 =", "answer": "100", "pred": "90"}, {"prompt": "77 + 54 =", "answer": "131", "pred": "121"}, {"prompt": "77 + 63 =", "answer": "140", "pred": "130"}, {"prompt": "99 + 61 =", "answer": "160", "pred": "150"}, {"prompt": "99 + 71 =", "answer": "170", "pred": "160"}, {"prompt": "99 + 81 =", "answer": "180", "pred": "170"}]}}
8
+ {"step": 300, "loss": 0.0200146484375, "ce": 0.004040641784667969, "kl": 0.3195703125, "lr": 0.0, "elapsed_s": 183.05117917060852, "compact_exact": {"n": 8100, "correct": 8080, "accuracy": 0.9975308641975309, "wrong_sample": [{"prompt": "11 + 31 =", "answer": "42", "pred": "422"}, {"prompt": "11 + 33 =", "answer": "44", "pred": "444"}, {"prompt": "11 + 34 =", "answer": "45", "pred": "455"}, {"prompt": "11 + 35 =", "answer": "46", "pred": "466"}, {"prompt": "11 + 37 =", "answer": "48", "pred": "488"}, {"prompt": "11 + 38 =", "answer": "49", "pred": "499999"}, {"prompt": "13 + 29 =", "answer": "42", "pred": "422"}, {"prompt": "13 + 79 =", "answer": "92", "pred": "922"}, {"prompt": "14 + 78 =", "answer": "92", "pred": "922"}, {"prompt": "18 + 74 =", "answer": "92", "pred": "922"}, {"prompt": "19 + 73 =", "answer": "92", "pred": "922"}, {"prompt": "77 + 10 =", "answer": "87", "pred": "877"}, {"prompt": "77 + 11 =", "answer": "88", "pred": "888888"}, {"prompt": "77 + 14 =", "answer": "91", "pred": "911"}, {"prompt": "77 + 19 =", "answer": "96", "pred": "966"}, {"prompt": "77 + 34 =", "answer": "111", "pred": "1111"}, {"prompt": "77 + 38 =", "answer": "115", "pred": "1155"}, {"prompt": "88 + 13 =", "answer": "101", "pred": "1011"}, {"prompt": "88 + 27 =", "answer": "115", "pred": "1155"}, {"prompt": "88 + 83 =", "answer": "171", "pred": "1711"}]}}
9
+ [done] adapter=results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_16/lora_r16_beta005/adapter accuracy=99.7531%
circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_1_correct_dataset.log ADDED
@@ -0,0 +1,26 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ Device: cuda dtype: bfloat16
2
+ Loading results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_1/lora_r1_beta005_merged ...
3
+
4
+ Pool: 8,100 unique (a, b) pairs in [10, 99] shard=1/1
5
+ 512/8100 correct= 507 acc=99.0% 648.1 ex/s eta 0.2 min
6
+ 1024/8100 correct= 1016 acc=99.2% 840.3 ex/s eta 0.1 min
7
+ 1536/8100 correct= 1526 acc=99.3% 932.2 ex/s eta 0.1 min
8
+ 2048/8100 correct= 2037 acc=99.5% 925.5 ex/s eta 0.1 min
9
+ 2560/8100 correct= 2545 acc=99.4% 967.9 ex/s eta 0.1 min
10
+ 3072/8100 correct= 3055 acc=99.4% 958.0 ex/s eta 0.1 min
11
+ 3584/8100 correct= 3562 acc=99.4% 983.3 ex/s eta 0.1 min
12
+ 4096/8100 correct= 4073 acc=99.4% 972.9 ex/s eta 0.1 min
13
+ 4608/8100 correct= 4582 acc=99.4% 990.4 ex/s eta 0.1 min
14
+ 5120/8100 correct= 5093 acc=99.5% 1004.3 ex/s eta 0.0 min
15
+ 5632/8100 correct= 5602 acc=99.5% 1016.6 ex/s eta 0.0 min
16
+ 6144/8100 correct= 6113 acc=99.5% 1027.4 ex/s eta 0.0 min
17
+ 6656/8100 correct= 6624 acc=99.5% 1035.8 ex/s eta 0.0 min
18
+ 7168/8100 correct= 7136 acc=99.6% 1042.8 ex/s eta 0.0 min
19
+ 7680/8100 correct= 7648 acc=99.6% 1049.1 ex/s eta 0.0 min
20
+ 8100/8100 correct= 8067 acc=99.6% 1052.6 ex/s eta 0.0 min
21
+
22
+ =======================================================
23
+ Tested: 8,100
24
+ Correct: 8,067 (99.6%)
25
+ Correct saved: data/qwen25_math_1p5b_rank_ladder_r1_correct_addition.json
26
+ =======================================================
circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_1_direct_rank_shard0.log ADDED
@@ -0,0 +1,68 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ [records] results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_1/position_adacs positions=hundreds,tens,ones
2
+ records=1,500 skipped=0
3
+ [model] results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_1/lora_r1_beta005_merged device=cuda dtype=bfloat16 batch=256
4
+
5
+ topk_100 mlp= 2193 exact=8.93%
6
+ rel_0.05 mlp= 3102 exact=17.20%
7
+ topk_500 mlp= 13350 exact=50.33%
8
+ rel_0.01 mlp= 25436 exact=38.27%
9
+ topk_2000 mlp= 52581 exact=30.87%
10
+ rel_0.001 mlp= 224993 exact=99.80%
11
+ [rank] todo=28 shard=0/4
12
+ drop__topk_500__layer_00 mlp= 13255 exact=5.27%
13
+ delta=-45.07% per1k=-474.386%
14
+ drop__topk_500__layer_04 mlp= 13178 exact=20.13%
15
+ delta=-30.20% per1k=-175.581%
16
+ drop__topk_500__layer_08 mlp= 13233 exact=49.93%
17
+ delta=-0.40% per1k=-3.419%
18
+ drop__topk_500__layer_12 mlp= 13235 exact=52.47%
19
+ delta=+2.13% per1k=+18.551%
20
+ drop__topk_500__layer_16 mlp= 13244 exact=51.00%
21
+ delta=+0.67% per1k=+6.289%
22
+ drop__topk_500__layer_20 mlp= 12925 exact=36.60%
23
+ delta=-13.73% per1k=-32.314%
24
+ drop__topk_500__layer_24 mlp= 10446 exact=39.93%
25
+ delta=-10.40% per1k=-3.581%
26
+ add__rel_0.01_minus_topk_500__layer_00 mlp= 13417 exact=50.40%
27
+ delta=+0.07% per1k=+0.995%
28
+ add__rel_0.01_minus_topk_500__layer_04 mlp= 13501 exact=49.73%
29
+ delta=-0.60% per1k=-3.974%
30
+ add__rel_0.01_minus_topk_500__layer_08 mlp= 13536 exact=43.33%
31
+ delta=-7.00% per1k=-37.634%
32
+ add__rel_0.01_minus_topk_500__layer_12 mlp= 13553 exact=48.80%
33
+ delta=-1.53% per1k=-7.553%
34
+ add__rel_0.01_minus_topk_500__layer_16 mlp= 13589 exact=52.87%
35
+ delta=+2.53% per1k=+10.600%
36
+ add__rel_0.01_minus_topk_500__layer_20 mlp= 13941 exact=50.00%
37
+ delta=-0.33% per1k=-0.564%
38
+ add__rel_0.01_minus_topk_500__layer_24 mlp= 15137 exact=50.20%
39
+ delta=-0.13% per1k=-0.075%
40
+ add__topk_2000_minus_topk_500__layer_00 mlp= 13602 exact=51.73%
41
+ delta=+1.40% per1k=+5.556%
42
+ add__topk_2000_minus_topk_500__layer_04 mlp= 13772 exact=51.40%
43
+ delta=+1.07% per1k=+2.528%
44
+ add__topk_2000_minus_topk_500__layer_08 mlp= 13920 exact=43.27%
45
+ delta=-7.07% per1k=-12.398%
46
+ add__topk_2000_minus_topk_500__layer_12 mlp= 13879 exact=46.73%
47
+ delta=-3.60% per1k=-6.805%
48
+ add__topk_2000_minus_topk_500__layer_16 mlp= 14054 exact=53.13%
49
+ delta=+2.80% per1k=+3.977%
50
+ add__topk_2000_minus_topk_500__layer_20 mlp= 15662 exact=52.07%
51
+ delta=+1.73% per1k=+0.750%
52
+ add__topk_2000_minus_topk_500__layer_24 mlp= 18072 exact=49.53%
53
+ delta=-0.80% per1k=-0.169%
54
+ add__rel_0.001_minus_topk_500__layer_00 mlp= 19323 exact=52.80%
55
+ delta=+2.47% per1k=+0.413%
56
+ add__rel_0.001_minus_topk_500__layer_04 mlp= 21073 exact=55.73%
57
+ delta=+5.40% per1k=+0.699%
58
+ add__rel_0.001_minus_topk_500__layer_08 mlp= 22118 exact=57.07%
59
+ delta=+6.73% per1k=+0.768%
60
+ add__rel_0.001_minus_topk_500__layer_12 mlp= 21918 exact=59.13%
61
+ delta=+8.80% per1k=+1.027%
62
+ add__rel_0.001_minus_topk_500__layer_16 mlp= 22005 exact=54.93%
63
+ delta=+4.60% per1k=+0.531%
64
+ add__rel_0.001_minus_topk_500__layer_20 mlp= 21880 exact=52.80%
65
+ delta=+2.47% per1k=+0.289%
66
+ add__rel_0.001_minus_topk_500__layer_24 mlp= 19406 exact=49.33%
67
+ delta=-1.00% per1k=-0.165%
68
+ [done] rank-only -> results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_1/direct_group_rank_shard0.json
circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_1_direct_rank_shard1.log ADDED
@@ -0,0 +1,68 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ [records] results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_1/position_adacs positions=hundreds,tens,ones
2
+ records=1,500 skipped=0
3
+ [model] results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_1/lora_r1_beta005_merged device=cuda dtype=bfloat16 batch=256
4
+
5
+ topk_100 mlp= 2193 exact=8.93%
6
+ rel_0.05 mlp= 3102 exact=17.20%
7
+ topk_500 mlp= 13350 exact=50.33%
8
+ rel_0.01 mlp= 25436 exact=38.27%
9
+ topk_2000 mlp= 52581 exact=30.87%
10
+ rel_0.001 mlp= 224993 exact=99.80%
11
+ [rank] todo=28 shard=1/4
12
+ drop__topk_500__layer_01 mlp= 13275 exact=38.13%
13
+ delta=-12.20% per1k=-162.667%
14
+ drop__topk_500__layer_05 mlp= 13243 exact=56.93%
15
+ delta=+6.60% per1k=+61.682%
16
+ drop__topk_500__layer_09 mlp= 13243 exact=56.13%
17
+ delta=+5.80% per1k=+54.206%
18
+ drop__topk_500__layer_13 mlp= 13249 exact=51.47%
19
+ delta=+1.13% per1k=+11.221%
20
+ drop__topk_500__layer_17 mlp= 13240 exact=52.73%
21
+ delta=+2.40% per1k=+21.818%
22
+ drop__topk_500__layer_21 mlp= 12750 exact=15.33%
23
+ delta=-35.00% per1k=-58.333%
24
+ drop__topk_500__layer_25 mlp= 10248 exact=39.80%
25
+ delta=-10.53% per1k=-3.396%
26
+ add__rel_0.01_minus_topk_500__layer_01 mlp= 13403 exact=50.00%
27
+ delta=-0.33% per1k=-6.289%
28
+ add__rel_0.01_minus_topk_500__layer_05 mlp= 13442 exact=48.00%
29
+ delta=-2.33% per1k=-25.362%
30
+ add__rel_0.01_minus_topk_500__layer_09 mlp= 13547 exact=48.27%
31
+ delta=-2.07% per1k=-10.491%
32
+ add__rel_0.01_minus_topk_500__layer_13 mlp= 13560 exact=49.73%
33
+ delta=-0.60% per1k=-2.857%
34
+ add__rel_0.01_minus_topk_500__layer_17 mlp= 13590 exact=46.80%
35
+ delta=-3.53% per1k=-14.722%
36
+ add__rel_0.01_minus_topk_500__layer_21 mlp= 14014 exact=51.13%
37
+ delta=+0.80% per1k=+1.205%
38
+ add__rel_0.01_minus_topk_500__layer_25 mlp= 15274 exact=49.93%
39
+ delta=-0.40% per1k=-0.208%
40
+ add__topk_2000_minus_topk_500__layer_01 mlp= 13511 exact=50.07%
41
+ delta=-0.27% per1k=-1.656%
42
+ add__topk_2000_minus_topk_500__layer_05 mlp= 13653 exact=49.13%
43
+ delta=-1.20% per1k=-3.960%
44
+ add__topk_2000_minus_topk_500__layer_09 mlp= 13964 exact=48.13%
45
+ delta=-2.20% per1k=-3.583%
46
+ add__topk_2000_minus_topk_500__layer_13 mlp= 13958 exact=51.27%
47
+ delta=+0.93% per1k=+1.535%
48
+ add__topk_2000_minus_topk_500__layer_17 mlp= 14113 exact=45.67%
49
+ delta=-4.67% per1k=-6.116%
50
+ add__topk_2000_minus_topk_500__layer_21 mlp= 16313 exact=49.87%
51
+ delta=-0.47% per1k=-0.157%
52
+ add__topk_2000_minus_topk_500__layer_25 mlp= 17933 exact=50.00%
53
+ delta=-0.33% per1k=-0.073%
54
+ add__rel_0.001_minus_topk_500__layer_01 mlp= 15618 exact=51.87%
55
+ delta=+1.53% per1k=+0.676%
56
+ add__rel_0.001_minus_topk_500__layer_05 mlp= 19936 exact=53.13%
57
+ delta=+2.80% per1k=+0.425%
58
+ add__rel_0.001_minus_topk_500__layer_09 mlp= 22101 exact=54.40%
59
+ delta=+4.07% per1k=+0.465%
60
+ add__rel_0.001_minus_topk_500__layer_13 mlp= 21879 exact=57.00%
61
+ delta=+6.67% per1k=+0.782%
62
+ add__rel_0.001_minus_topk_500__layer_17 mlp= 22038 exact=50.47%
63
+ delta=+0.13% per1k=+0.015%
64
+ add__rel_0.001_minus_topk_500__layer_21 mlp= 21707 exact=51.80%
65
+ delta=+1.47% per1k=+0.176%
66
+ add__rel_0.001_minus_topk_500__layer_25 mlp= 19207 exact=49.87%
67
+ delta=-0.47% per1k=-0.080%
68
+ [done] rank-only -> results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_1/direct_group_rank_shard1.json
circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_1_direct_rank_shard2.log ADDED
@@ -0,0 +1,68 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ [records] results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_1/position_adacs positions=hundreds,tens,ones
2
+ records=1,500 skipped=0
3
+ [model] results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_1/lora_r1_beta005_merged device=cuda dtype=bfloat16 batch=256
4
+
5
+ topk_100 mlp= 2193 exact=8.93%
6
+ rel_0.05 mlp= 3102 exact=17.20%
7
+ topk_500 mlp= 13350 exact=50.33%
8
+ rel_0.01 mlp= 25436 exact=38.27%
9
+ topk_2000 mlp= 52581 exact=30.87%
10
+ rel_0.001 mlp= 224993 exact=99.80%
11
+ [rank] todo=28 shard=2/4
12
+ drop__topk_500__layer_02 mlp= 13261 exact=36.00%
13
+ delta=-14.33% per1k=-161.049%
14
+ drop__topk_500__layer_06 mlp= 13216 exact=50.07%
15
+ delta=-0.27% per1k=-1.990%
16
+ drop__topk_500__layer_10 mlp= 13257 exact=53.93%
17
+ delta=+3.60% per1k=+38.710%
18
+ drop__topk_500__layer_14 mlp= 13224 exact=52.93%
19
+ delta=+2.60% per1k=+20.635%
20
+ drop__topk_500__layer_18 mlp= 13189 exact=42.27%
21
+ delta=-8.07% per1k=-50.104%
22
+ drop__topk_500__layer_22 mlp= 12377 exact=4.93%
23
+ delta=-45.40% per1k=-46.660%
24
+ drop__topk_500__layer_26 mlp= 12386 exact=47.73%
25
+ delta=-2.60% per1k=-2.697%
26
+ add__rel_0.01_minus_topk_500__layer_02 mlp= 13424 exact=53.00%
27
+ delta=+2.67% per1k=+36.036%
28
+ add__rel_0.01_minus_topk_500__layer_06 mlp= 13538 exact=50.87%
29
+ delta=+0.53% per1k=+2.837%
30
+ add__rel_0.01_minus_topk_500__layer_10 mlp= 13524 exact=50.27%
31
+ delta=-0.07% per1k=-0.383%
32
+ add__rel_0.01_minus_topk_500__layer_14 mlp= 13557 exact=46.87%
33
+ delta=-3.47% per1k=-16.747%
34
+ add__rel_0.01_minus_topk_500__layer_18 mlp= 13650 exact=49.07%
35
+ delta=-1.27% per1k=-4.222%
36
+ add__rel_0.01_minus_topk_500__layer_22 mlp= 14294 exact=49.87%
37
+ delta=-0.47% per1k=-0.494%
38
+ add__rel_0.01_minus_topk_500__layer_26 mlp= 14307 exact=49.80%
39
+ delta=-0.53% per1k=-0.557%
40
+ add__topk_2000_minus_topk_500__layer_02 mlp= 13594 exact=55.27%
41
+ delta=+4.93% per1k=+20.219%
42
+ add__topk_2000_minus_topk_500__layer_06 mlp= 13901 exact=50.27%
43
+ delta=-0.07% per1k=-0.121%
44
+ add__topk_2000_minus_topk_500__layer_10 mlp= 13912 exact=47.87%
45
+ delta=-2.47% per1k=-4.389%
46
+ add__topk_2000_minus_topk_500__layer_14 mlp= 13978 exact=47.00%
47
+ delta=-3.33% per1k=-5.308%
48
+ add__topk_2000_minus_topk_500__layer_18 mlp= 14322 exact=46.93%
49
+ delta=-3.40% per1k=-3.498%
50
+ add__topk_2000_minus_topk_500__layer_22 mlp= 17597 exact=50.80%
51
+ delta=+0.47% per1k=+0.110%
52
+ add__topk_2000_minus_topk_500__layer_26 mlp= 16603 exact=50.00%
53
+ delta=-0.33% per1k=-0.102%
54
+ add__rel_0.001_minus_topk_500__layer_02 mlp= 17491 exact=60.27%
55
+ delta=+9.93% per1k=+2.399%
56
+ add__rel_0.001_minus_topk_500__layer_06 mlp= 22085 exact=56.00%
57
+ delta=+5.67% per1k=+0.649%
58
+ add__rel_0.001_minus_topk_500__layer_10 mlp= 22079 exact=58.67%
59
+ delta=+8.33% per1k=+0.955%
60
+ add__rel_0.001_minus_topk_500__layer_14 mlp= 21887 exact=52.47%
61
+ delta=+2.13% per1k=+0.250%
62
+ add__rel_0.001_minus_topk_500__layer_18 mlp= 21950 exact=51.60%
63
+ delta=+1.27% per1k=+0.147%
64
+ add__rel_0.001_minus_topk_500__layer_22 mlp= 21336 exact=51.87%
65
+ delta=+1.53% per1k=+0.192%
66
+ add__rel_0.001_minus_topk_500__layer_26 mlp= 21341 exact=49.60%
67
+ delta=-0.73% per1k=-0.092%
68
+ [done] rank-only -> results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_1/direct_group_rank_shard2.json
circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_1_direct_rank_shard3.log ADDED
@@ -0,0 +1,68 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ [records] results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_1/position_adacs positions=hundreds,tens,ones
2
+ records=1,500 skipped=0
3
+ [model] results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_1/lora_r1_beta005_merged device=cuda dtype=bfloat16 batch=256
4
+
5
+ topk_100 mlp= 2193 exact=8.93%
6
+ rel_0.05 mlp= 3102 exact=17.20%
7
+ topk_500 mlp= 13350 exact=50.33%
8
+ rel_0.01 mlp= 25436 exact=38.27%
9
+ topk_2000 mlp= 52581 exact=30.87%
10
+ rel_0.001 mlp= 224993 exact=99.80%
11
+ [rank] todo=28 shard=3/4
12
+ drop__topk_500__layer_03 mlp= 13293 exact=39.87%
13
+ delta=-10.47% per1k=-183.626%
14
+ drop__topk_500__layer_07 mlp= 13227 exact=47.07%
15
+ delta=-3.27% per1k=-26.558%
16
+ drop__topk_500__layer_11 mlp= 13261 exact=49.40%
17
+ delta=-0.93% per1k=-10.487%
18
+ drop__topk_500__layer_15 mlp= 13263 exact=51.20%
19
+ delta=+0.87% per1k=+9.962%
20
+ drop__topk_500__layer_19 mlp= 13017 exact=26.20%
21
+ delta=-24.13% per1k=-72.472%
22
+ drop__topk_500__layer_23 mlp= 11834 exact=7.07%
23
+ delta=-43.27% per1k=-28.540%
24
+ drop__topk_500__layer_27 mlp= 12881 exact=47.33%
25
+ delta=-3.00% per1k=-6.397%
26
+ add__rel_0.01_minus_topk_500__layer_03 mlp= 13414 exact=49.53%
27
+ delta=-0.80% per1k=-12.500%
28
+ add__rel_0.01_minus_topk_500__layer_07 mlp= 13542 exact=48.73%
29
+ delta=-1.60% per1k=-8.333%
30
+ add__rel_0.01_minus_topk_500__layer_11 mlp= 13519 exact=48.13%
31
+ delta=-2.20% per1k=-13.018%
32
+ add__rel_0.01_minus_topk_500__layer_15 mlp= 13534 exact=50.67%
33
+ delta=+0.33% per1k=+1.812%
34
+ add__rel_0.01_minus_topk_500__layer_19 mlp= 13774 exact=49.40%
35
+ delta=-0.93% per1k=-2.201%
36
+ add__rel_0.01_minus_topk_500__layer_23 mlp= 14581 exact=49.93%
37
+ delta=-0.40% per1k=-0.325%
38
+ add__rel_0.01_minus_topk_500__layer_27 mlp= 13741 exact=50.00%
39
+ delta=-0.33% per1k=-0.853%
40
+ add__topk_2000_minus_topk_500__layer_03 mlp= 13511 exact=50.20%
41
+ delta=-0.13% per1k=-0.828%
42
+ add__topk_2000_minus_topk_500__layer_07 mlp= 13951 exact=47.73%
43
+ delta=-2.60% per1k=-4.326%
44
+ add__topk_2000_minus_topk_500__layer_11 mlp= 13865 exact=46.87%
45
+ delta=-3.47% per1k=-6.731%
46
+ add__topk_2000_minus_topk_500__layer_15 mlp= 13930 exact=51.33%
47
+ delta=+1.00% per1k=+1.724%
48
+ add__topk_2000_minus_topk_500__layer_19 mlp= 14754 exact=50.20%
49
+ delta=-0.13% per1k=-0.095%
50
+ add__topk_2000_minus_topk_500__layer_23 mlp= 17984 exact=51.20%
51
+ delta=+0.87% per1k=+0.187%
52
+ add__topk_2000_minus_topk_500__layer_27 mlp= 14723 exact=50.20%
53
+ delta=-0.13% per1k=-0.097%
54
+ add__rel_0.001_minus_topk_500__layer_03 mlp= 17233 exact=51.87%
55
+ delta=+1.53% per1k=+0.395%
56
+ add__rel_0.001_minus_topk_500__layer_07 mlp= 22088 exact=56.20%
57
+ delta=+5.87% per1k=+0.671%
58
+ add__rel_0.001_minus_topk_500__layer_11 mlp= 21990 exact=56.13%
59
+ delta=+5.80% per1k=+0.671%
60
+ add__rel_0.001_minus_topk_500__layer_15 mlp= 21789 exact=52.93%
61
+ delta=+2.60% per1k=+0.308%
62
+ add__rel_0.001_minus_topk_500__layer_19 mlp= 21909 exact=54.40%
63
+ delta=+4.07% per1k=+0.475%
64
+ add__rel_0.001_minus_topk_500__layer_23 mlp= 20794 exact=50.60%
65
+ delta=+0.27% per1k=+0.036%
66
+ add__rel_0.001_minus_topk_500__layer_27 mlp= 21262 exact=50.27%
67
+ delta=-0.07% per1k=-0.008%
68
+ [done] rank-only -> results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_1/direct_group_rank_shard3.json
circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_1_fresh_build.log ADDED
@@ -0,0 +1,7 @@
 
 
 
 
 
 
 
 
1
+ [dataset] data/qwen25_math_1p5b_rank_ladder_r1_correct_addition.json
2
+ [pairing] loading tokenizer for prompt-length matching
3
+ enriched examples=8,067
4
+ hundreds pairs= 3150 buckets= 79
5
+ tens pairs= 4000 buckets= 20
6
+ ones pairs= 4000 buckets= 18
7
+ [done] build-only artifacts -> results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_1/fresh_pairs_seed123
circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_1_fresh_target90_eval.log ADDED
@@ -0,0 +1,6 @@
 
 
 
 
 
 
 
1
+ [records] results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_1/fresh_pairs_seed123
2
+ records=1,500 skipped=0
3
+ [model] results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_1/lora_r1_beta005_merged device=cuda dtype=bfloat16 batch=256
4
+
5
+ target90 mlp= 171936 exact=88.13%
6
+ [done] results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_1/fresh_target90_eval.json
circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_1_fresh_target90_generation.log ADDED
@@ -0,0 +1,4 @@
 
 
 
 
 
1
+ [records] 1,500 skipped=0
2
+
3
+ target90 mlp= 171936 gen_exact=88.53%
4
+ [done] results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_1/fresh_target90_generation.json
circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_1_merge.log ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ Warning: You are sending unauthenticated requests to the HF Hub. Please set a HF_TOKEN to enable higher rate limits and faster downloads.
2
+ [model] Qwen/Qwen2.5-Math-1.5B adapter=results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_1/lora_r1_beta005/adapter device=cuda
3
+
4
+ scale=0.45 kl=0.678664 acc=99.69% correct=8075/8100
5
+ scale=0.55 kl=0.781586 acc=99.78% correct=8082/8100
6
+ scale=0.75 kl=0.932692 acc=99.86% correct=8089/8100
7
+ scale=1 kl=1.04953 acc=99.93% correct=8094/8100
8
+
9
+ [done] selected scale=0.45 rule=lowest_kl_at_or_above_min_accuracy merged=results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_1/lora_r1_beta005_merged
circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_1_position_trace.log ADDED
@@ -0,0 +1,64 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ [worker gpu=0 positions=hundreds]
2
+ [dataset] data/qwen25_math_1p5b_rank_ladder_r1_correct_addition.json
3
+ [pairing] loading tokenizer for prompt-length matching
4
+ enriched examples=8,067
5
+ hundreds pairs= 3150 buckets= 79
6
+ [model] results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_1/lora_r1_beta005_merged
7
+ device=cuda dtype=bfloat16 batch_size=256
8
+
9
+ Loading weights: 0%| | 0/338 [00:00<?, ?it/s]
10
+ Loading weights: 100%|██████████| 338/338 [00:00<00:00, 5988.95it/s]
11
+ layers=28 d_ffn=8960 heads=12
12
+ [hundreds] attr_pairs=2650 test_pairs=500
13
+ hundreds positive mlp= 248711 zero=100.00% cf=100.00%
14
+ hundreds rel_0.001 mlp= 179236 zero=99.80% cf=99.20%
15
+ hundreds rel_0.01 mlp= 10305 zero=56.60% cf=65.20%
16
+ hundreds rel_0.05 mlp= 1261 zero=49.00% cf=2.80%
17
+ hundreds topk_100 mlp= 950 zero=23.20% cf=0.00%
18
+ hundreds topk_500 mlp= 5750 zero=63.60% cf=56.60%
19
+ hundreds topk_2000 mlp= 25273 zero=81.40% cf=81.80%
20
+ [done] artifacts -> results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_1/position_adacs/_workers/gpu0
21
+
22
+ [worker gpu=1 positions=tens]
23
+ [dataset] data/qwen25_math_1p5b_rank_ladder_r1_correct_addition.json
24
+ [pairing] loading tokenizer for prompt-length matching
25
+ enriched examples=8,067
26
+ tens pairs= 4000 buckets= 20
27
+ [model] results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_1/lora_r1_beta005_merged
28
+ device=cuda dtype=bfloat16 batch_size=256
29
+
30
+ Loading weights: 0%| | 0/338 [00:00<?, ?it/s]
31
+ Loading weights: 100%|██████████| 338/338 [00:00<00:00, 7085.22it/s]
32
+ layers=28 d_ffn=8960 heads=12
33
+ [tens] attr_pairs=3500 test_pairs=500
34
+ tens positive mlp= 250588 zero=99.40% cf=100.00%
35
+ tens rel_0.001 mlp= 208497 zero=99.80% cf=100.00%
36
+ tens rel_0.01 mlp= 18204 zero=13.60% cf=69.00%
37
+ tens rel_0.05 mlp= 2266 zero=6.60% cf=33.60%
38
+ tens topk_100 mlp= 1648 zero=8.80% cf=14.00%
39
+ tens topk_500 mlp= 9775 zero=12.00% cf=78.80%
40
+ tens topk_2000 mlp= 41217 zero=8.40% cf=55.80%
41
+ [done] artifacts -> results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_1/position_adacs/_workers/gpu1
42
+
43
+ [worker gpu=2 positions=ones]
44
+ [dataset] data/qwen25_math_1p5b_rank_ladder_r1_correct_addition.json
45
+ [pairing] loading tokenizer for prompt-length matching
46
+ enriched examples=8,067
47
+ ones pairs= 4000 buckets= 18
48
+ [model] results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_1/lora_r1_beta005_merged
49
+ device=cuda dtype=bfloat16 batch_size=256
50
+
51
+ Loading weights: 0%| | 0/338 [00:00<?, ?it/s]
52
+ Loading weights: 100%|██████████| 338/338 [00:00<00:00, 7076.98it/s]
53
+ layers=28 d_ffn=8960 heads=12
54
+ [ones] attr_pairs=3500 test_pairs=500
55
+ ones positive mlp= 250847 zero=100.00% cf=100.00%
56
+ ones rel_0.001 mlp= 221809 zero=100.00% cf=100.00%
57
+ ones rel_0.01 mlp= 17674 zero=22.80% cf=71.40%
58
+ ones rel_0.05 mlp= 2191 zero=9.60% cf=44.20%
59
+ ones topk_100 mlp= 1521 zero=8.20% cf=24.40%
60
+ ones topk_500 mlp= 8175 zero=22.20% cf=75.40%
61
+ ones topk_2000 mlp= 39213 zero=18.60% cf=66.00%
62
+ [done] artifacts -> results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_1/position_adacs/_workers/gpu2
63
+
64
+ [done] merged multi-gpu artifacts -> results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_1/position_adacs
circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_1_prep_stage.log ADDED
@@ -0,0 +1,5 @@
 
 
 
 
 
 
1
+ [rank 1] prep start on gpu 0
2
+ [rank 1] train cached
3
+ [rank 1] merge cached
4
+ [rank 1] correct dataset cached
5
+ [rank 1] prep done
circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_1_target90_search.log ADDED
@@ -0,0 +1,370 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+
2
+ topk_100 mlp= 2193 exact=8.93%
3
+ rel_0.05 mlp= 3102 exact=17.20%
4
+ topk_500 mlp= 13350 exact=50.33%
5
+ rel_0.01 mlp= 25436 exact=38.27%
6
+ topk_2000 mlp= 52581 exact=30.87%
7
+ rel_0.001 mlp= 224993 exact=99.80%
8
+ add_step_1_add__rel_0.001_minus_topk_500__layer_02 mlp= 17491 exact=60.27%
9
+ add_step_1_add__rel_0.001_minus_topk_500__layer_12 mlp= 21918 exact=59.13%
10
+ add_step_1_add__rel_0.001_minus_topk_500__layer_10 mlp= 22079 exact=58.67%
11
+ add_step_1_add__rel_0.001_minus_topk_500__layer_08 mlp= 22118 exact=57.07%
12
+ add_step_1_add__rel_0.001_minus_topk_500__layer_13 mlp= 21879 exact=57.00%
13
+ add_step_1_add__rel_0.001_minus_topk_500__layer_07 mlp= 22088 exact=56.20%
14
+ add_step_1_add__rel_0.001_minus_topk_500__layer_11 mlp= 21990 exact=56.13%
15
+ add_step_1_add__rel_0.001_minus_topk_500__layer_06 mlp= 22085 exact=56.00%
16
+ add_step_1_add__rel_0.001_minus_topk_500__layer_04 mlp= 21073 exact=55.73%
17
+ add_step_1_add__topk_2000_minus_topk_500__layer_02 mlp= 13594 exact=55.27%
18
+ add_step_1_add__rel_0.001_minus_topk_500__layer_16 mlp= 22005 exact=54.93%
19
+ add_step_1_add__rel_0.001_minus_topk_500__layer_19 mlp= 21909 exact=54.40%
20
+ add_step_1_add__rel_0.001_minus_topk_500__layer_09 mlp= 22101 exact=54.40%
21
+ add_step_1_add__topk_2000_minus_topk_500__layer_16 mlp= 14054 exact=53.13%
22
+ add_step_1_add__rel_0.001_minus_topk_500__layer_05 mlp= 19936 exact=53.13%
23
+ add_step_1_add__rel_0.01_minus_topk_500__layer_02 mlp= 13424 exact=53.00%
24
+ add_step_1_add__rel_0.001_minus_topk_500__layer_15 mlp= 21789 exact=52.93%
25
+ add_step_1_add__rel_0.01_minus_topk_500__layer_16 mlp= 13589 exact=52.87%
26
+ add_step_1_add__rel_0.001_minus_topk_500__layer_00 mlp= 19323 exact=52.80%
27
+ add_step_1_add__rel_0.001_minus_topk_500__layer_20 mlp= 21880 exact=52.80%
28
+ add_step_1_add__rel_0.001_minus_topk_500__layer_14 mlp= 21887 exact=52.47%
29
+ add_step_1_add__topk_2000_minus_topk_500__layer_20 mlp= 15662 exact=52.07%
30
+ add_step_1_add__rel_0.001_minus_topk_500__layer_01 mlp= 15618 exact=51.87%
31
+ add_step_1_add__rel_0.001_minus_topk_500__layer_03 mlp= 17233 exact=51.87%
32
+ add_step_1_add__rel_0.001_minus_topk_500__layer_22 mlp= 21336 exact=51.87%
33
+ add_step_1_add__rel_0.001_minus_topk_500__layer_21 mlp= 21707 exact=51.80%
34
+ add_step_1_add__topk_2000_minus_topk_500__layer_00 mlp= 13602 exact=51.73%
35
+ add_step_1_add__rel_0.001_minus_topk_500__layer_18 mlp= 21950 exact=51.60%
36
+ add_step_1_add__topk_2000_minus_topk_500__layer_04 mlp= 13772 exact=51.40%
37
+ add_step_1_add__topk_2000_minus_topk_500__layer_15 mlp= 13930 exact=51.33%
38
+ add_step_1_add__topk_2000_minus_topk_500__layer_13 mlp= 13958 exact=51.27%
39
+ add_step_1_add__topk_2000_minus_topk_500__layer_23 mlp= 17984 exact=51.20%
40
+ add_step_1_add__rel_0.01_minus_topk_500__layer_21 mlp= 14014 exact=51.13%
41
+ add_step_1_add__rel_0.01_minus_topk_500__layer_06 mlp= 13538 exact=50.87%
42
+ add_step_1_add__topk_2000_minus_topk_500__layer_22 mlp= 17597 exact=50.80%
43
+ add_step_1_add__rel_0.01_minus_topk_500__layer_15 mlp= 13534 exact=50.67%
44
+ add_step_1_add__rel_0.001_minus_topk_500__layer_23 mlp= 20794 exact=50.60%
45
+ add_step_1_add__rel_0.001_minus_topk_500__layer_17 mlp= 22038 exact=50.47%
46
+ add_step_1_add__rel_0.01_minus_topk_500__layer_00 mlp= 13417 exact=50.40%
47
+ add_step_1_add__rel_0.001_minus_topk_500__layer_27 mlp= 21262 exact=50.27%
48
+ add_step_1_add__topk_2000_minus_topk_500__layer_06 mlp= 13901 exact=50.27%
49
+ add_step_1_add__rel_0.01_minus_topk_500__layer_10 mlp= 13524 exact=50.27%
50
+ add_step_1_add__rel_0.01_minus_topk_500__layer_24 mlp= 15137 exact=50.20%
51
+ add_step_1_add__topk_2000_minus_topk_500__layer_19 mlp= 14754 exact=50.20%
52
+ add_step_1_add__topk_2000_minus_topk_500__layer_27 mlp= 14723 exact=50.20%
53
+ add_step_1_add__topk_2000_minus_topk_500__layer_03 mlp= 13511 exact=50.20%
54
+ add_step_1_add__topk_2000_minus_topk_500__layer_01 mlp= 13511 exact=50.07%
55
+ add_step_1_add__topk_2000_minus_topk_500__layer_25 mlp= 17933 exact=50.00%
56
+ add_step_1_accepted mlp= 17491 exact=60.27%
57
+ add_step_2_add__rel_0.001_minus_topk_500__layer_12 mlp= 26059 exact=67.47%
58
+ add_step_2_add__rel_0.001_minus_topk_500__layer_10 mlp= 26220 exact=69.53%
59
+ add_step_2_add__rel_0.001_minus_topk_500__layer_08 mlp= 26259 exact=66.53%
60
+ add_step_2_add__rel_0.001_minus_topk_500__layer_13 mlp= 26020 exact=66.27%
61
+ add_step_2_add__rel_0.001_minus_topk_500__layer_07 mlp= 26229 exact=63.73%
62
+ add_step_2_add__rel_0.001_minus_topk_500__layer_11 mlp= 26131 exact=64.80%
63
+ add_step_2_add__rel_0.001_minus_topk_500__layer_06 mlp= 26226 exact=64.73%
64
+ add_step_2_add__rel_0.001_minus_topk_500__layer_04 mlp= 25214 exact=64.47%
65
+ add_step_2_add__rel_0.001_minus_topk_500__layer_16 mlp= 26146 exact=62.27%
66
+ add_step_2_add__rel_0.001_minus_topk_500__layer_19 mlp= 26050 exact=64.33%
67
+ add_step_2_add__rel_0.001_minus_topk_500__layer_09 mlp= 26242 exact=64.40%
68
+ add_step_2_add__topk_2000_minus_topk_500__layer_16 mlp= 18195 exact=61.60%
69
+ add_step_2_add__rel_0.001_minus_topk_500__layer_05 mlp= 24077 exact=62.27%
70
+ add_step_2_add__rel_0.001_minus_topk_500__layer_15 mlp= 25930 exact=62.67%
71
+ add_step_2_add__rel_0.01_minus_topk_500__layer_16 mlp= 17730 exact=62.67%
72
+ add_step_2_add__rel_0.001_minus_topk_500__layer_00 mlp= 23464 exact=62.53%
73
+ add_step_2_add__rel_0.001_minus_topk_500__layer_20 mlp= 26021 exact=62.60%
74
+ add_step_2_add__rel_0.001_minus_topk_500__layer_14 mlp= 26028 exact=65.27%
75
+ add_step_2_add__topk_2000_minus_topk_500__layer_20 mlp= 19803 exact=61.53%
76
+ add_step_2_add__rel_0.001_minus_topk_500__layer_01 mlp= 19759 exact=59.53%
77
+ add_step_2_add__rel_0.001_minus_topk_500__layer_03 mlp= 21374 exact=58.73%
78
+ add_step_2_add__rel_0.001_minus_topk_500__layer_22 mlp= 25477 exact=60.93%
79
+ add_step_2_add__rel_0.001_minus_topk_500__layer_21 mlp= 25848 exact=62.27%
80
+ add_step_2_add__topk_2000_minus_topk_500__layer_00 mlp= 17743 exact=60.27%
81
+ add_step_2_add__rel_0.001_minus_topk_500__layer_18 mlp= 26091 exact=61.13%
82
+ add_step_2_add__topk_2000_minus_topk_500__layer_04 mlp= 17913 exact=61.20%
83
+ add_step_2_add__topk_2000_minus_topk_500__layer_15 mlp= 18071 exact=61.07%
84
+ add_step_2_add__topk_2000_minus_topk_500__layer_13 mlp= 18099 exact=60.27%
85
+ add_step_2_add__topk_2000_minus_topk_500__layer_23 mlp= 22125 exact=61.07%
86
+ add_step_2_add__rel_0.01_minus_topk_500__layer_21 mlp= 18155 exact=60.87%
87
+ add_step_2_add__rel_0.01_minus_topk_500__layer_06 mlp= 17679 exact=60.07%
88
+ add_step_2_add__topk_2000_minus_topk_500__layer_22 mlp= 21738 exact=60.00%
89
+ add_step_2_add__rel_0.01_minus_topk_500__layer_15 mlp= 17675 exact=60.20%
90
+ add_step_2_add__rel_0.001_minus_topk_500__layer_23 mlp= 24935 exact=60.60%
91
+ add_step_2_add__rel_0.001_minus_topk_500__layer_17 mlp= 26179 exact=61.80%
92
+ add_step_2_add__rel_0.01_minus_topk_500__layer_00 mlp= 17558 exact=59.73%
93
+ add_step_2_add__rel_0.001_minus_topk_500__layer_27 mlp= 25403 exact=60.20%
94
+ add_step_2_add__topk_2000_minus_topk_500__layer_06 mlp= 18042 exact=59.60%
95
+ add_step_2_add__rel_0.01_minus_topk_500__layer_10 mlp= 17665 exact=60.47%
96
+ add_step_2_add__rel_0.01_minus_topk_500__layer_24 mlp= 19278 exact=60.53%
97
+ add_step_2_add__topk_2000_minus_topk_500__layer_19 mlp= 18895 exact=59.80%
98
+ add_step_2_add__topk_2000_minus_topk_500__layer_27 mlp= 18864 exact=60.33%
99
+ add_step_2_add__topk_2000_minus_topk_500__layer_03 mlp= 17652 exact=58.27%
100
+ add_step_2_add__topk_2000_minus_topk_500__layer_01 mlp= 17652 exact=59.87%
101
+ add_step_2_add__topk_2000_minus_topk_500__layer_25 mlp= 22074 exact=59.93%
102
+ add_step_2_accepted mlp= 26220 exact=69.53%
103
+ add_step_3_add__rel_0.001_minus_topk_500__layer_12 mlp= 34788 exact=77.67%
104
+ add_step_3_add__rel_0.001_minus_topk_500__layer_08 mlp= 34988 exact=73.93%
105
+ add_step_3_add__rel_0.001_minus_topk_500__layer_13 mlp= 34749 exact=78.60%
106
+ add_step_3_add__rel_0.001_minus_topk_500__layer_07 mlp= 34958 exact=70.07%
107
+ add_step_3_add__rel_0.001_minus_topk_500__layer_11 mlp= 34860 exact=74.40%
108
+ add_step_3_add__rel_0.001_minus_topk_500__layer_06 mlp= 34955 exact=73.80%
109
+ add_step_3_add__rel_0.001_minus_topk_500__layer_04 mlp= 33943 exact=74.73%
110
+ add_step_3_add__rel_0.001_minus_topk_500__layer_16 mlp= 34875 exact=71.80%
111
+ add_step_3_add__rel_0.001_minus_topk_500__layer_19 mlp= 34779 exact=74.40%
112
+ add_step_3_add__rel_0.001_minus_topk_500__layer_09 mlp= 34971 exact=71.73%
113
+ add_step_3_add__topk_2000_minus_topk_500__layer_16 mlp= 26924 exact=70.80%
114
+ add_step_3_add__rel_0.001_minus_topk_500__layer_05 mlp= 32806 exact=71.80%
115
+ add_step_3_add__rel_0.001_minus_topk_500__layer_15 mlp= 34659 exact=78.00%
116
+ add_step_3_add__rel_0.01_minus_topk_500__layer_16 mlp= 26459 exact=70.87%
117
+ add_step_3_add__rel_0.001_minus_topk_500__layer_00 mlp= 32193 exact=72.53%
118
+ add_step_3_add__rel_0.001_minus_topk_500__layer_20 mlp= 34750 exact=71.20%
119
+ add_step_3_add__rel_0.001_minus_topk_500__layer_14 mlp= 34757 exact=73.27%
120
+ add_step_3_add__topk_2000_minus_topk_500__layer_20 mlp= 28532 exact=70.27%
121
+ add_step_3_add__rel_0.001_minus_topk_500__layer_01 mlp= 28488 exact=70.60%
122
+ add_step_3_add__rel_0.001_minus_topk_500__layer_03 mlp= 30103 exact=70.67%
123
+ add_step_3_add__rel_0.001_minus_topk_500__layer_22 mlp= 34206 exact=71.60%
124
+ add_step_3_add__rel_0.001_minus_topk_500__layer_21 mlp= 34577 exact=72.53%
125
+ add_step_3_add__topk_2000_minus_topk_500__layer_00 mlp= 26472 exact=70.47%
126
+ add_step_3_add__rel_0.001_minus_topk_500__layer_18 mlp= 34820 exact=73.00%
127
+ add_step_3_add__topk_2000_minus_topk_500__layer_04 mlp= 26642 exact=70.13%
128
+ add_step_3_add__topk_2000_minus_topk_500__layer_15 mlp= 26800 exact=70.53%
129
+ add_step_3_add__topk_2000_minus_topk_500__layer_13 mlp= 26828 exact=70.47%
130
+ add_step_3_add__topk_2000_minus_topk_500__layer_23 mlp= 30854 exact=70.00%
131
+ add_step_3_add__rel_0.01_minus_topk_500__layer_21 mlp= 26884 exact=71.33%
132
+ add_step_3_add__rel_0.01_minus_topk_500__layer_06 mlp= 26408 exact=69.00%
133
+ add_step_3_add__topk_2000_minus_topk_500__layer_22 mlp= 30467 exact=69.93%
134
+ add_step_3_add__rel_0.01_minus_topk_500__layer_15 mlp= 26404 exact=69.53%
135
+ add_step_3_add__rel_0.001_minus_topk_500__layer_23 mlp= 33664 exact=70.33%
136
+ add_step_3_add__rel_0.001_minus_topk_500__layer_17 mlp= 34908 exact=74.53%
137
+ add_step_3_add__rel_0.01_minus_topk_500__layer_00 mlp= 26287 exact=69.20%
138
+ add_step_3_add__rel_0.001_minus_topk_500__layer_27 mlp= 34132 exact=69.73%
139
+ add_step_3_add__topk_2000_minus_topk_500__layer_06 mlp= 26771 exact=68.07%
140
+ add_step_3_add__rel_0.01_minus_topk_500__layer_24 mlp= 28007 exact=69.07%
141
+ add_step_3_add__topk_2000_minus_topk_500__layer_19 mlp= 27624 exact=69.73%
142
+ add_step_3_add__topk_2000_minus_topk_500__layer_27 mlp= 27593 exact=69.53%
143
+ add_step_3_add__topk_2000_minus_topk_500__layer_03 mlp= 26381 exact=68.47%
144
+ add_step_3_add__topk_2000_minus_topk_500__layer_01 mlp= 26381 exact=70.00%
145
+ add_step_3_add__topk_2000_minus_topk_500__layer_25 mlp= 30803 exact=69.87%
146
+ add_step_3_accepted mlp= 34749 exact=78.60%
147
+ add_step_4_add__rel_0.001_minus_topk_500__layer_12 mlp= 43317 exact=84.00%
148
+ add_step_4_add__rel_0.001_minus_topk_500__layer_08 mlp= 43517 exact=79.73%
149
+ add_step_4_add__rel_0.001_minus_topk_500__layer_07 mlp= 43487 exact=79.93%
150
+ add_step_4_add__rel_0.001_minus_topk_500__layer_11 mlp= 43389 exact=79.53%
151
+ add_step_4_add__rel_0.001_minus_topk_500__layer_06 mlp= 43484 exact=82.60%
152
+ add_step_4_add__rel_0.001_minus_topk_500__layer_04 mlp= 42472 exact=80.80%
153
+ add_step_4_add__rel_0.001_minus_topk_500__layer_16 mlp= 43404 exact=80.13%
154
+ add_step_4_add__rel_0.001_minus_topk_500__layer_19 mlp= 43308 exact=80.07%
155
+ add_step_4_add__rel_0.001_minus_topk_500__layer_09 mlp= 43500 exact=78.07%
156
+ add_step_4_add__topk_2000_minus_topk_500__layer_16 mlp= 35453 exact=78.73%
157
+ add_step_4_add__rel_0.001_minus_topk_500__layer_05 mlp= 41335 exact=80.87%
158
+ add_step_4_add__rel_0.001_minus_topk_500__layer_15 mlp= 43188 exact=83.67%
159
+ add_step_4_add__rel_0.01_minus_topk_500__layer_16 mlp= 34988 exact=79.00%
160
+ add_step_4_add__rel_0.001_minus_topk_500__layer_00 mlp= 40722 exact=81.33%
161
+ add_step_4_add__rel_0.001_minus_topk_500__layer_20 mlp= 43279 exact=79.80%
162
+ add_step_4_add__rel_0.001_minus_topk_500__layer_14 mlp= 43286 exact=81.20%
163
+ add_step_4_add__topk_2000_minus_topk_500__layer_20 mlp= 37061 exact=78.33%
164
+ add_step_4_add__rel_0.001_minus_topk_500__layer_01 mlp= 37017 exact=80.33%
165
+ add_step_4_add__rel_0.001_minus_topk_500__layer_03 mlp= 38632 exact=78.27%
166
+ add_step_4_add__rel_0.001_minus_topk_500__layer_22 mlp= 42735 exact=79.87%
167
+ add_step_4_add__rel_0.001_minus_topk_500__layer_21 mlp= 43106 exact=79.27%
168
+ add_step_4_add__topk_2000_minus_topk_500__layer_00 mlp= 35001 exact=79.33%
169
+ add_step_4_add__rel_0.001_minus_topk_500__layer_18 mlp= 43349 exact=81.27%
170
+ add_step_4_add__topk_2000_minus_topk_500__layer_04 mlp= 35171 exact=80.07%
171
+ add_step_4_add__topk_2000_minus_topk_500__layer_15 mlp= 35329 exact=79.00%
172
+ add_step_4_add__topk_2000_minus_topk_500__layer_23 mlp= 39383 exact=79.13%
173
+ add_step_4_add__rel_0.01_minus_topk_500__layer_21 mlp= 35413 exact=78.53%
174
+ add_step_4_add__rel_0.01_minus_topk_500__layer_06 mlp= 34937 exact=77.33%
175
+ add_step_4_add__topk_2000_minus_topk_500__layer_22 mlp= 38996 exact=79.33%
176
+ add_step_4_add__rel_0.01_minus_topk_500__layer_15 mlp= 34933 exact=77.47%
177
+ add_step_4_add__rel_0.001_minus_topk_500__layer_23 mlp= 42193 exact=78.93%
178
+ add_step_4_add__rel_0.001_minus_topk_500__layer_17 mlp= 43437 exact=82.33%
179
+ add_step_4_add__rel_0.01_minus_topk_500__layer_00 mlp= 34816 exact=78.40%
180
+ add_step_4_add__rel_0.001_minus_topk_500__layer_27 mlp= 42661 exact=78.60%
181
+ add_step_4_add__topk_2000_minus_topk_500__layer_06 mlp= 35300 exact=77.67%
182
+ add_step_4_add__rel_0.01_minus_topk_500__layer_24 mlp= 36536 exact=78.47%
183
+ add_step_4_add__topk_2000_minus_topk_500__layer_19 mlp= 36153 exact=77.73%
184
+ add_step_4_add__topk_2000_minus_topk_500__layer_27 mlp= 36122 exact=78.67%
185
+ add_step_4_add__topk_2000_minus_topk_500__layer_03 mlp= 34910 exact=78.13%
186
+ add_step_4_add__topk_2000_minus_topk_500__layer_01 mlp= 34910 exact=79.73%
187
+ add_step_4_add__topk_2000_minus_topk_500__layer_25 mlp= 39332 exact=78.53%
188
+ add_step_4_accepted mlp= 43317 exact=84.00%
189
+ baseline_rel_0.001_compress_1_drop_00 mlp= 218925 exact=22.33%
190
+ baseline_rel_0.001_compress_1_drop_01 mlp= 222650 exact=72.40%
191
+ baseline_rel_0.001_compress_1_drop_02 mlp= 220763 exact=95.13%
192
+ baseline_rel_0.001_compress_1_drop_03 mlp= 221053 exact=94.40%
193
+ baseline_rel_0.001_compress_1_drop_04 mlp= 217098 exact=89.27%
194
+ baseline_rel_0.001_compress_1_drop_05 mlp= 218300 exact=98.80%
195
+ baseline_rel_0.001_compress_1_drop_06 mlp= 216124 exact=98.73%
196
+ baseline_rel_0.001_compress_1_drop_07 mlp= 216132 exact=99.60%
197
+ baseline_rel_0.001_compress_1_drop_08 mlp= 216108 exact=99.73%
198
+ baseline_rel_0.001_compress_1_drop_09 mlp= 216135 exact=99.73%
199
+ baseline_rel_0.001_compress_1_drop_10 mlp= 216171 exact=99.40%
200
+ baseline_rel_0.001_compress_1_drop_11 mlp= 216264 exact=98.93%
201
+ baseline_rel_0.001_compress_1_drop_12 mlp= 216310 exact=98.33%
202
+ baseline_rel_0.001_compress_1_drop_13 mlp= 216363 exact=99.47%
203
+ baseline_rel_0.001_compress_1_drop_14 mlp= 216330 exact=99.60%
204
+ baseline_rel_0.001_compress_1_drop_15 mlp= 216467 exact=99.60%
205
+ baseline_rel_0.001_compress_1_drop_16 mlp= 216232 exact=99.87%
206
+ baseline_rel_0.001_compress_1_drop_17 mlp= 216195 exact=99.87%
207
+ baseline_rel_0.001_compress_1_drop_18 mlp= 216232 exact=96.93%
208
+ baseline_rel_0.001_compress_1_drop_19 mlp= 216101 exact=96.40%
209
+ baseline_rel_0.001_compress_1_drop_20 mlp= 216038 exact=96.73%
210
+ baseline_rel_0.001_compress_1_drop_21 mlp= 216036 exact=77.07%
211
+ baseline_rel_0.001_compress_1_drop_22 mlp= 216034 exact=10.73%
212
+ baseline_rel_0.001_compress_1_drop_23 mlp= 216033 exact=14.47%
213
+ baseline_rel_0.001_compress_1_drop_24 mlp= 216033 exact=93.67%
214
+ baseline_rel_0.001_compress_1_drop_25 mlp= 216034 exact=86.87%
215
+ baseline_rel_0.001_compress_1_drop_26 mlp= 216038 exact=99.67%
216
+ baseline_rel_0.001_compress_1_drop_27 mlp= 216612 exact=99.87%
217
+ baseline_rel_0.001_compress_1_accepted mlp= 216033 exact=93.67%
218
+ baseline_rel_0.001_compress_2_drop_00 mlp= 209965 exact=19.33%
219
+ baseline_rel_0.001_compress_2_drop_01 mlp= 213690 exact=69.53%
220
+ baseline_rel_0.001_compress_2_drop_02 mlp= 211803 exact=86.07%
221
+ baseline_rel_0.001_compress_2_drop_03 mlp= 212093 exact=86.13%
222
+ baseline_rel_0.001_compress_2_drop_04 mlp= 208138 exact=76.07%
223
+ baseline_rel_0.001_compress_2_drop_05 mlp= 209340 exact=93.20%
224
+ baseline_rel_0.001_compress_2_drop_06 mlp= 207164 exact=91.80%
225
+ baseline_rel_0.001_compress_2_drop_07 mlp= 207172 exact=92.20%
226
+ baseline_rel_0.001_compress_2_drop_08 mlp= 207148 exact=93.53%
227
+ baseline_rel_0.001_compress_2_drop_09 mlp= 207175 exact=93.33%
228
+ baseline_rel_0.001_compress_2_drop_10 mlp= 207211 exact=93.20%
229
+ baseline_rel_0.001_compress_2_drop_11 mlp= 207304 exact=92.00%
230
+ baseline_rel_0.001_compress_2_drop_12 mlp= 207350 exact=90.67%
231
+ baseline_rel_0.001_compress_2_drop_13 mlp= 207403 exact=93.33%
232
+ baseline_rel_0.001_compress_2_drop_14 mlp= 207370 exact=93.33%
233
+ baseline_rel_0.001_compress_2_drop_15 mlp= 207507 exact=94.27%
234
+ baseline_rel_0.001_compress_2_drop_16 mlp= 207272 exact=93.93%
235
+ baseline_rel_0.001_compress_2_drop_17 mlp= 207235 exact=93.60%
236
+ baseline_rel_0.001_compress_2_drop_18 mlp= 207272 exact=88.33%
237
+ baseline_rel_0.001_compress_2_drop_19 mlp= 207141 exact=83.47%
238
+ baseline_rel_0.001_compress_2_drop_20 mlp= 207078 exact=85.27%
239
+ baseline_rel_0.001_compress_2_drop_21 mlp= 207076 exact=67.20%
240
+ baseline_rel_0.001_compress_2_drop_22 mlp= 207074 exact=10.07%
241
+ baseline_rel_0.001_compress_2_drop_23 mlp= 207073 exact=7.13%
242
+ baseline_rel_0.001_compress_2_drop_25 mlp= 207074 exact=54.73%
243
+ baseline_rel_0.001_compress_2_drop_26 mlp= 207078 exact=88.27%
244
+ baseline_rel_0.001_compress_2_drop_27 mlp= 207652 exact=91.60%
245
+ baseline_rel_0.001_compress_2_accepted mlp= 207148 exact=93.53%
246
+ baseline_rel_0.001_compress_3_drop_00 mlp= 201080 exact=16.93%
247
+ baseline_rel_0.001_compress_3_drop_01 mlp= 204805 exact=68.33%
248
+ baseline_rel_0.001_compress_3_drop_02 mlp= 202918 exact=83.80%
249
+ baseline_rel_0.001_compress_3_drop_03 mlp= 203208 exact=85.73%
250
+ baseline_rel_0.001_compress_3_drop_04 mlp= 199253 exact=70.53%
251
+ baseline_rel_0.001_compress_3_drop_05 mlp= 200455 exact=92.80%
252
+ baseline_rel_0.001_compress_3_drop_06 mlp= 198279 exact=90.93%
253
+ baseline_rel_0.001_compress_3_drop_07 mlp= 198287 exact=90.00%
254
+ baseline_rel_0.001_compress_3_drop_09 mlp= 198290 exact=93.07%
255
+ baseline_rel_0.001_compress_3_drop_10 mlp= 198326 exact=93.73%
256
+ baseline_rel_0.001_compress_3_drop_11 mlp= 198419 exact=91.53%
257
+ baseline_rel_0.001_compress_3_drop_12 mlp= 198465 exact=90.87%
258
+ baseline_rel_0.001_compress_3_drop_13 mlp= 198518 exact=93.53%
259
+ baseline_rel_0.001_compress_3_drop_14 mlp= 198485 exact=92.93%
260
+ baseline_rel_0.001_compress_3_drop_15 mlp= 198622 exact=94.00%
261
+ baseline_rel_0.001_compress_3_drop_16 mlp= 198387 exact=94.07%
262
+ baseline_rel_0.001_compress_3_drop_17 mlp= 198350 exact=93.40%
263
+ baseline_rel_0.001_compress_3_drop_18 mlp= 198387 exact=87.13%
264
+ baseline_rel_0.001_compress_3_drop_19 mlp= 198256 exact=81.53%
265
+ baseline_rel_0.001_compress_3_drop_20 mlp= 198193 exact=84.27%
266
+ baseline_rel_0.001_compress_3_drop_21 mlp= 198191 exact=64.53%
267
+ baseline_rel_0.001_compress_3_drop_22 mlp= 198189 exact=9.60%
268
+ baseline_rel_0.001_compress_3_drop_23 mlp= 198188 exact=6.87%
269
+ baseline_rel_0.001_compress_3_drop_25 mlp= 198189 exact=54.87%
270
+ baseline_rel_0.001_compress_3_drop_26 mlp= 198193 exact=87.00%
271
+ baseline_rel_0.001_compress_3_drop_27 mlp= 198767 exact=91.00%
272
+ baseline_rel_0.001_compress_3_accepted mlp= 198279 exact=90.93%
273
+ baseline_rel_0.001_compress_4_drop_00 mlp= 192211 exact=8.93%
274
+ baseline_rel_0.001_compress_4_drop_01 mlp= 195936 exact=59.00%
275
+ baseline_rel_0.001_compress_4_drop_02 mlp= 194049 exact=70.67%
276
+ baseline_rel_0.001_compress_4_drop_03 mlp= 194339 exact=74.87%
277
+ baseline_rel_0.001_compress_4_drop_04 mlp= 190384 exact=48.47%
278
+ baseline_rel_0.001_compress_4_drop_05 mlp= 191586 exact=89.33%
279
+ baseline_rel_0.001_compress_4_drop_07 mlp= 189418 exact=84.40%
280
+ baseline_rel_0.001_compress_4_drop_09 mlp= 189421 exact=90.60%
281
+ baseline_rel_0.001_compress_4_drop_10 mlp= 189457 exact=90.87%
282
+ baseline_rel_0.001_compress_4_drop_11 mlp= 189550 exact=85.33%
283
+ baseline_rel_0.001_compress_4_drop_12 mlp= 189596 exact=85.87%
284
+ baseline_rel_0.001_compress_4_drop_13 mlp= 189649 exact=90.13%
285
+ baseline_rel_0.001_compress_4_drop_14 mlp= 189616 exact=91.20%
286
+ baseline_rel_0.001_compress_4_drop_15 mlp= 189753 exact=90.67%
287
+ baseline_rel_0.001_compress_4_drop_16 mlp= 189518 exact=91.67%
288
+ baseline_rel_0.001_compress_4_drop_17 mlp= 189481 exact=90.33%
289
+ baseline_rel_0.001_compress_4_drop_18 mlp= 189518 exact=79.60%
290
+ baseline_rel_0.001_compress_4_drop_19 mlp= 189387 exact=69.40%
291
+ baseline_rel_0.001_compress_4_drop_20 mlp= 189324 exact=81.27%
292
+ baseline_rel_0.001_compress_4_drop_21 mlp= 189322 exact=48.20%
293
+ baseline_rel_0.001_compress_4_drop_22 mlp= 189320 exact=9.13%
294
+ baseline_rel_0.001_compress_4_drop_23 mlp= 189319 exact=6.13%
295
+ baseline_rel_0.001_compress_4_drop_25 mlp= 189320 exact=54.00%
296
+ baseline_rel_0.001_compress_4_drop_26 mlp= 189324 exact=83.27%
297
+ baseline_rel_0.001_compress_4_drop_27 mlp= 189898 exact=88.60%
298
+ baseline_rel_0.001_compress_4_accepted mlp= 189421 exact=90.60%
299
+ baseline_rel_0.001_compress_5_drop_00 mlp= 183353 exact=8.47%
300
+ baseline_rel_0.001_compress_5_drop_01 mlp= 187078 exact=60.67%
301
+ baseline_rel_0.001_compress_5_drop_02 mlp= 185191 exact=74.60%
302
+ baseline_rel_0.001_compress_5_drop_03 mlp= 185481 exact=75.47%
303
+ baseline_rel_0.001_compress_5_drop_04 mlp= 181526 exact=47.13%
304
+ baseline_rel_0.001_compress_5_drop_05 mlp= 182728 exact=88.67%
305
+ baseline_rel_0.001_compress_5_drop_07 mlp= 180560 exact=84.27%
306
+ baseline_rel_0.001_compress_5_drop_10 mlp= 180599 exact=90.40%
307
+ baseline_rel_0.001_compress_5_drop_11 mlp= 180692 exact=84.00%
308
+ baseline_rel_0.001_compress_5_drop_12 mlp= 180738 exact=83.87%
309
+ baseline_rel_0.001_compress_5_drop_13 mlp= 180791 exact=89.93%
310
+ baseline_rel_0.001_compress_5_drop_14 mlp= 180758 exact=90.80%
311
+ baseline_rel_0.001_compress_5_drop_15 mlp= 180895 exact=90.20%
312
+ baseline_rel_0.001_compress_5_drop_16 mlp= 180660 exact=90.87%
313
+ baseline_rel_0.001_compress_5_drop_17 mlp= 180623 exact=90.60%
314
+ baseline_rel_0.001_compress_5_drop_18 mlp= 180660 exact=79.27%
315
+ baseline_rel_0.001_compress_5_drop_19 mlp= 180529 exact=68.80%
316
+ baseline_rel_0.001_compress_5_drop_20 mlp= 180466 exact=80.80%
317
+ baseline_rel_0.001_compress_5_drop_21 mlp= 180464 exact=45.87%
318
+ baseline_rel_0.001_compress_5_drop_22 mlp= 180462 exact=8.87%
319
+ baseline_rel_0.001_compress_5_drop_23 mlp= 180461 exact=6.47%
320
+ baseline_rel_0.001_compress_5_drop_25 mlp= 180462 exact=54.13%
321
+ baseline_rel_0.001_compress_5_drop_26 mlp= 180466 exact=83.47%
322
+ baseline_rel_0.001_compress_5_drop_27 mlp= 181040 exact=88.73%
323
+ baseline_rel_0.001_compress_5_accepted mlp= 180599 exact=90.40%
324
+ baseline_rel_0.001_compress_6_drop_00 mlp= 174531 exact=7.80%
325
+ baseline_rel_0.001_compress_6_drop_01 mlp= 178256 exact=60.33%
326
+ baseline_rel_0.001_compress_6_drop_02 mlp= 176369 exact=72.20%
327
+ baseline_rel_0.001_compress_6_drop_03 mlp= 176659 exact=74.20%
328
+ baseline_rel_0.001_compress_6_drop_04 mlp= 172704 exact=43.20%
329
+ baseline_rel_0.001_compress_6_drop_05 mlp= 173906 exact=87.87%
330
+ baseline_rel_0.001_compress_6_drop_07 mlp= 171738 exact=81.80%
331
+ baseline_rel_0.001_compress_6_drop_11 mlp= 171870 exact=82.07%
332
+ baseline_rel_0.001_compress_6_drop_12 mlp= 171916 exact=80.73%
333
+ baseline_rel_0.001_compress_6_drop_13 mlp= 171969 exact=89.87%
334
+ baseline_rel_0.001_compress_6_drop_14 mlp= 171936 exact=90.00%
335
+ baseline_rel_0.001_compress_6_drop_15 mlp= 172073 exact=89.67%
336
+ baseline_rel_0.001_compress_6_drop_16 mlp= 171838 exact=89.93%
337
+ baseline_rel_0.001_compress_6_drop_17 mlp= 171801 exact=89.47%
338
+ baseline_rel_0.001_compress_6_drop_18 mlp= 171838 exact=77.67%
339
+ baseline_rel_0.001_compress_6_drop_19 mlp= 171707 exact=66.33%
340
+ baseline_rel_0.001_compress_6_drop_20 mlp= 171644 exact=79.27%
341
+ baseline_rel_0.001_compress_6_drop_21 mlp= 171642 exact=42.67%
342
+ baseline_rel_0.001_compress_6_drop_22 mlp= 171640 exact=8.33%
343
+ baseline_rel_0.001_compress_6_drop_23 mlp= 171639 exact=6.67%
344
+ baseline_rel_0.001_compress_6_drop_25 mlp= 171640 exact=53.27%
345
+ baseline_rel_0.001_compress_6_drop_26 mlp= 171644 exact=83.00%
346
+ baseline_rel_0.001_compress_6_drop_27 mlp= 172218 exact=88.40%
347
+ baseline_rel_0.001_compress_6_accepted mlp= 171936 exact=90.00%
348
+ baseline_rel_0.001_compress_7_drop_00 mlp= 165868 exact=7.80%
349
+ baseline_rel_0.001_compress_7_drop_01 mlp= 169593 exact=60.20%
350
+ baseline_rel_0.001_compress_7_drop_02 mlp= 167706 exact=71.13%
351
+ baseline_rel_0.001_compress_7_drop_03 mlp= 167996 exact=74.40%
352
+ baseline_rel_0.001_compress_7_drop_04 mlp= 164041 exact=42.27%
353
+ baseline_rel_0.001_compress_7_drop_05 mlp= 165243 exact=88.27%
354
+ baseline_rel_0.001_compress_7_drop_07 mlp= 163075 exact=81.60%
355
+ baseline_rel_0.001_compress_7_drop_11 mlp= 163207 exact=80.87%
356
+ baseline_rel_0.001_compress_7_drop_12 mlp= 163253 exact=81.07%
357
+ baseline_rel_0.001_compress_7_drop_13 mlp= 163306 exact=87.20%
358
+ baseline_rel_0.001_compress_7_drop_15 mlp= 163410 exact=87.20%
359
+ baseline_rel_0.001_compress_7_drop_16 mlp= 163175 exact=89.07%
360
+ baseline_rel_0.001_compress_7_drop_17 mlp= 163138 exact=87.93%
361
+ baseline_rel_0.001_compress_7_drop_18 mlp= 163175 exact=76.80%
362
+ baseline_rel_0.001_compress_7_drop_19 mlp= 163044 exact=63.07%
363
+ baseline_rel_0.001_compress_7_drop_20 mlp= 162981 exact=77.13%
364
+ baseline_rel_0.001_compress_7_drop_21 mlp= 162979 exact=39.93%
365
+ baseline_rel_0.001_compress_7_drop_22 mlp= 162977 exact=7.67%
366
+ baseline_rel_0.001_compress_7_drop_23 mlp= 162976 exact=6.33%
367
+ baseline_rel_0.001_compress_7_drop_25 mlp= 162977 exact=52.93%
368
+ baseline_rel_0.001_compress_7_drop_26 mlp= 162981 exact=82.87%
369
+ baseline_rel_0.001_compress_7_drop_27 mlp= 163555 exact=88.07%
370
+ [done] results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_1/target90_search.json
circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_1_train.log ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ [model] Qwen/Qwen2.5-Math-1.5B device=cuda dtype=bfloat16
2
+
3
+ Warning: You are sending unauthenticated requests to the HF Hub. Please set a HF_TOKEN to enable higher rate limits and faster downloads.
4
+ trainable params: 1,154,048 || all params: 1,544,868,352 || trainable%: 0.0747
5
+ {"step": 1, "loss": 2.296875, "ce": 2.296875, "kl": -0.000545501708984375, "lr": 1.3333333333333333e-05, "elapsed_s": 0.7464790344238281, "compact_exact": {"n": 8100, "correct": 7590, "accuracy": 0.937037037037037, "wrong_sample": [{"prompt": "11 + 13 =", "answer": "24", "pred": "35"}, {"prompt": "11 + 21 =", "answer": "32", "pred": "132"}, {"prompt": "11 + 31 =", "answer": "42", "pred": "142"}, {"prompt": "11 + 41 =", "answer": "52", "pred": "152"}, {"prompt": "11 + 47 =", "answer": "58", "pred": "158"}, {"prompt": "11 + 48 =", "answer": "59", "pred": "159"}, {"prompt": "11 + 49 =", "answer": "60", "pred": "160"}, {"prompt": "11 + 50 =", "answer": "61", "pred": "161"}, {"prompt": "11 + 51 =", "answer": "62", "pred": "162"}, {"prompt": "11 + 53 =", "answer": "64", "pred": "164"}, {"prompt": "11 + 57 =", "answer": "68", "pred": "168"}, {"prompt": "11 + 58 =", "answer": "69", "pred": "169"}, {"prompt": "11 + 59 =", "answer": "70", "pred": "170"}, {"prompt": "11 + 61 =", "answer": "72", "pred": "172"}, {"prompt": "11 + 62 =", "answer": "73", "pred": "173"}, {"prompt": "11 + 67 =", "answer": "78", "pred": "178"}, {"prompt": "11 + 68 =", "answer": "79", "pred": "179"}, {"prompt": "11 + 69 =", "answer": "80", "pred": "180"}, {"prompt": "11 + 70 =", "answer": "81", "pred": "181"}, {"prompt": "11 + 71 =", "answer": "82", "pred": "182"}, {"prompt": "11 + 75 =", "answer": "86", "pred": "186"}, {"prompt": "11 + 76 =", "answer": "87", "pred": "187"}, {"prompt": "11 + 78 =", "answer": "89", "pred": "189"}, {"prompt": "11 + 79 =", "answer": "90", "pred": "190"}, {"prompt": "11 + 80 =", "answer": "91", "pred": "191"}]}}
6
+ {"step": 100, "loss": 0.3946804470486111, "ce": 0.3804256554805871, "kl": 0.2851865556504991, "lr": 0.00015922352526649803, "elapsed_s": 68.56942892074585, "compact_exact": {"n": 8100, "correct": 8083, "accuracy": 0.9979012345679013, "wrong_sample": [{"prompt": "55 + 25 =", "answer": "80", "pred": "75"}, {"prompt": "55 + 35 =", "answer": "90", "pred": "80"}, {"prompt": "62 + 18 =", "answer": "80", "pred": "78"}, {"prompt": "63 + 17 =", "answer": "80", "pred": "70"}, {"prompt": "66 + 14 =", "answer": "80", "pred": "78"}, {"prompt": "66 + 24 =", "answer": "90", "pred": "80"}, {"prompt": "66 + 25 =", "answer": "91", "pred": "81"}, {"prompt": "66 + 34 =", "answer": "100", "pred": "90"}, {"prompt": "66 + 64 =", "answer": "130", "pred": "120"}, {"prompt": "67 + 32 =", "answer": "99", "pred": "100"}, {"prompt": "69 + 12 =", "answer": "81", "pred": "811"}, {"prompt": "69 + 15 =", "answer": "84", "pred": "814"}, {"prompt": "76 + 14 =", "answer": "90", "pred": "80"}, {"prompt": "77 + 73 =", "answer": "150", "pred": "140"}, {"prompt": "87 + 22 =", "answer": "109", "pred": "110"}, {"prompt": "98 + 11 =", "answer": "109", "pred": "110"}, {"prompt": "99 + 81 =", "answer": "180", "pred": "170"}]}}
7
+ {"step": 200, "loss": 0.0204315185546875, "ce": 0.004192657470703125, "kl": 0.324755859375, "lr": 5.484666416891109e-05, "elapsed_s": 122.76283836364746, "compact_exact": {"n": 8100, "correct": 8093, "accuracy": 0.9991358024691358, "wrong_sample": [{"prompt": "62 + 18 =", "answer": "80", "pred": "78"}, {"prompt": "63 + 17 =", "answer": "80", "pred": "70"}, {"prompt": "66 + 64 =", "answer": "130", "pred": "120"}, {"prompt": "76 + 14 =", "answer": "90", "pred": "80"}, {"prompt": "77 + 73 =", "answer": "150", "pred": "140"}, {"prompt": "88 + 73 =", "answer": "161", "pred": "151"}, {"prompt": "99 + 81 =", "answer": "180", "pred": "170"}]}}
8
+ {"step": 300, "loss": 0.020775146484375, "ce": 0.004177169799804687, "kl": 0.33209716796875, "lr": 0.0, "elapsed_s": 176.21899390220642, "compact_exact": {"n": 8100, "correct": 8095, "accuracy": 0.9993827160493827, "wrong_sample": [{"prompt": "62 + 18 =", "answer": "80", "pred": "78"}, {"prompt": "63 + 17 =", "answer": "80", "pred": "70"}, {"prompt": "77 + 73 =", "answer": "150", "pred": "140"}, {"prompt": "88 + 73 =", "answer": "161", "pred": "151"}, {"prompt": "99 + 81 =", "answer": "180", "pred": "170"}]}}
9
+ [done] adapter=results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_1/lora_r1_beta005/adapter accuracy=99.9383%
circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_2_correct_dataset.log ADDED
@@ -0,0 +1,26 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ Device: cuda dtype: bfloat16
2
+ Loading results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_2/lora_r2_beta005_merged ...
3
+
4
+ Pool: 8,100 unique (a, b) pairs in [10, 99] shard=1/1
5
+ 512/8100 correct= 510 acc=99.6% 383.1 ex/s eta 0.3 min
6
+ 1024/8100 correct= 1020 acc=99.6% 539.6 ex/s eta 0.2 min
7
+ 1536/8100 correct= 1530 acc=99.6% 297.5 ex/s eta 0.4 min
8
+ 2048/8100 correct= 2040 acc=99.6% 365.7 ex/s eta 0.3 min
9
+ 2560/8100 correct= 2551 acc=99.6% 404.8 ex/s eta 0.2 min
10
+ 3072/8100 correct= 3062 acc=99.7% 454.0 ex/s eta 0.2 min
11
+ 3584/8100 correct= 3573 acc=99.7% 497.5 ex/s eta 0.2 min
12
+ 4096/8100 correct= 4084 acc=99.7% 526.8 ex/s eta 0.1 min
13
+ 4608/8100 correct= 4596 acc=99.7% 551.7 ex/s eta 0.1 min
14
+ 5120/8100 correct= 5107 acc=99.7% 582.0 ex/s eta 0.1 min
15
+ 5632/8100 correct= 5619 acc=99.8% 608.8 ex/s eta 0.1 min
16
+ 6144/8100 correct= 6127 acc=99.7% 633.4 ex/s eta 0.1 min
17
+ 6656/8100 correct= 6638 acc=99.7% 647.4 ex/s eta 0.0 min
18
+ 7168/8100 correct= 7149 acc=99.7% 659.6 ex/s eta 0.0 min
19
+ 7680/8100 correct= 7659 acc=99.7% 678.4 ex/s eta 0.0 min
20
+ 8100/8100 correct= 8078 acc=99.7% 685.0 ex/s eta 0.0 min
21
+
22
+ =======================================================
23
+ Tested: 8,100
24
+ Correct: 8,078 (99.7%)
25
+ Correct saved: data/qwen25_math_1p5b_rank_ladder_r2_correct_addition.json
26
+ =======================================================
circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_2_direct_rank_shard0.log ADDED
@@ -0,0 +1,68 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ [records] results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_2/position_adacs positions=hundreds,tens,ones
2
+ records=1,500 skipped=0
3
+ [model] results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_2/lora_r2_beta005_merged device=cuda dtype=bfloat16 batch=256
4
+
5
+ topk_100 mlp= 2025 exact=7.07%
6
+ rel_0.05 mlp= 2965 exact=12.40%
7
+ topk_500 mlp= 12798 exact=51.33%
8
+ rel_0.01 mlp= 24067 exact=43.93%
9
+ topk_2000 mlp= 51432 exact=34.00%
10
+ rel_0.001 mlp= 224312 exact=99.93%
11
+ [rank] todo=28 shard=0/4
12
+ drop__topk_500__layer_00 mlp= 12700 exact=5.53%
13
+ delta=-45.80% per1k=-467.347%
14
+ drop__topk_500__layer_04 mlp= 12613 exact=20.33%
15
+ delta=-31.00% per1k=-167.568%
16
+ drop__topk_500__layer_08 mlp= 12677 exact=53.00%
17
+ delta=+1.67% per1k=+13.774%
18
+ drop__topk_500__layer_12 mlp= 12673 exact=53.00%
19
+ delta=+1.67% per1k=+13.333%
20
+ drop__topk_500__layer_16 mlp= 12680 exact=50.73%
21
+ delta=-0.60% per1k=-5.085%
22
+ drop__topk_500__layer_20 mlp= 12395 exact=37.53%
23
+ delta=-13.80% per1k=-34.243%
24
+ drop__topk_500__layer_24 mlp= 10138 exact=37.73%
25
+ delta=-13.60% per1k=-5.113%
26
+ add__rel_0.01_minus_topk_500__layer_00 mlp= 12870 exact=50.80%
27
+ delta=-0.53% per1k=-7.407%
28
+ add__rel_0.01_minus_topk_500__layer_04 mlp= 12939 exact=50.87%
29
+ delta=-0.47% per1k=-3.310%
30
+ add__rel_0.01_minus_topk_500__layer_08 mlp= 12967 exact=49.87%
31
+ delta=-1.47% per1k=-8.679%
32
+ add__rel_0.01_minus_topk_500__layer_12 mlp= 12981 exact=49.67%
33
+ delta=-1.67% per1k=-9.107%
34
+ add__rel_0.01_minus_topk_500__layer_16 mlp= 13013 exact=50.27%
35
+ delta=-1.07% per1k=-4.961%
36
+ add__rel_0.01_minus_topk_500__layer_20 mlp= 13301 exact=52.00%
37
+ delta=+0.67% per1k=+1.325%
38
+ add__rel_0.01_minus_topk_500__layer_24 mlp= 14498 exact=51.47%
39
+ delta=+0.13% per1k=+0.078%
40
+ add__topk_2000_minus_topk_500__layer_00 mlp= 13063 exact=51.33%
41
+ delta=+0.00% per1k=+0.000%
42
+ add__topk_2000_minus_topk_500__layer_04 mlp= 13230 exact=49.80%
43
+ delta=-1.53% per1k=-3.549%
44
+ add__topk_2000_minus_topk_500__layer_08 mlp= 13368 exact=49.67%
45
+ delta=-1.67% per1k=-2.924%
46
+ add__topk_2000_minus_topk_500__layer_12 mlp= 13347 exact=50.27%
47
+ delta=-1.07% per1k=-1.943%
48
+ add__topk_2000_minus_topk_500__layer_16 mlp= 13477 exact=50.27%
49
+ delta=-1.07% per1k=-1.571%
50
+ add__topk_2000_minus_topk_500__layer_20 mlp= 15012 exact=51.73%
51
+ delta=+0.40% per1k=+0.181%
52
+ add__topk_2000_minus_topk_500__layer_24 mlp= 17516 exact=51.40%
53
+ delta=+0.07% per1k=+0.014%
54
+ add__rel_0.001_minus_topk_500__layer_00 mlp= 18726 exact=54.07%
55
+ delta=+2.73% per1k=+0.461%
56
+ add__rel_0.001_minus_topk_500__layer_04 mlp= 20537 exact=60.67%
57
+ delta=+9.33% per1k=+1.206%
58
+ add__rel_0.001_minus_topk_500__layer_08 mlp= 21549 exact=62.87%
59
+ delta=+11.53% per1k=+1.318%
60
+ add__rel_0.001_minus_topk_500__layer_12 mlp= 21339 exact=55.60%
61
+ delta=+4.27% per1k=+0.500%
62
+ add__rel_0.001_minus_topk_500__layer_16 mlp= 21383 exact=55.07%
63
+ delta=+3.73% per1k=+0.435%
64
+ add__rel_0.001_minus_topk_500__layer_20 mlp= 21350 exact=54.53%
65
+ delta=+3.20% per1k=+0.374%
66
+ add__rel_0.001_minus_topk_500__layer_24 mlp= 19098 exact=51.47%
67
+ delta=+0.13% per1k=+0.021%
68
+ [done] rank-only -> results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_2/direct_group_rank_shard0.json
circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_2_direct_rank_shard1.log ADDED
@@ -0,0 +1,68 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ [records] results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_2/position_adacs positions=hundreds,tens,ones
2
+ records=1,500 skipped=0
3
+ [model] results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_2/lora_r2_beta005_merged device=cuda dtype=bfloat16 batch=256
4
+
5
+ topk_100 mlp= 2025 exact=7.07%
6
+ rel_0.05 mlp= 2965 exact=12.40%
7
+ topk_500 mlp= 12798 exact=51.33%
8
+ rel_0.01 mlp= 24067 exact=43.93%
9
+ topk_2000 mlp= 51432 exact=34.00%
10
+ rel_0.001 mlp= 224312 exact=99.93%
11
+ [rank] todo=28 shard=1/4
12
+ drop__topk_500__layer_01 mlp= 12716 exact=41.27%
13
+ delta=-10.07% per1k=-122.764%
14
+ drop__topk_500__layer_05 mlp= 12683 exact=56.07%
15
+ delta=+4.73% per1k=+41.159%
16
+ drop__topk_500__layer_09 mlp= 12691 exact=54.33%
17
+ delta=+3.00% per1k=+28.037%
18
+ drop__topk_500__layer_13 mlp= 12706 exact=51.73%
19
+ delta=+0.40% per1k=+4.348%
20
+ drop__topk_500__layer_17 mlp= 12680 exact=50.47%
21
+ delta=-0.87% per1k=-7.345%
22
+ drop__topk_500__layer_21 mlp= 12250 exact=16.00%
23
+ delta=-35.33% per1k=-64.477%
24
+ drop__topk_500__layer_25 mlp= 9845 exact=38.47%
25
+ delta=-12.87% per1k=-4.357%
26
+ add__rel_0.01_minus_topk_500__layer_01 mlp= 12848 exact=51.53%
27
+ delta=+0.20% per1k=+4.000%
28
+ add__rel_0.01_minus_topk_500__layer_05 mlp= 12885 exact=51.27%
29
+ delta=-0.07% per1k=-0.766%
30
+ add__rel_0.01_minus_topk_500__layer_09 mlp= 12989 exact=49.73%
31
+ delta=-1.60% per1k=-8.377%
32
+ add__rel_0.01_minus_topk_500__layer_13 mlp= 13003 exact=51.13%
33
+ delta=-0.20% per1k=-0.976%
34
+ add__rel_0.01_minus_topk_500__layer_17 mlp= 13004 exact=50.07%
35
+ delta=-1.27% per1k=-6.149%
36
+ add__rel_0.01_minus_topk_500__layer_21 mlp= 13436 exact=52.40%
37
+ delta=+1.07% per1k=+1.672%
38
+ add__rel_0.01_minus_topk_500__layer_25 mlp= 14568 exact=51.00%
39
+ delta=-0.33% per1k=-0.188%
40
+ add__topk_2000_minus_topk_500__layer_01 mlp= 12967 exact=51.47%
41
+ delta=+0.13% per1k=+0.789%
42
+ add__topk_2000_minus_topk_500__layer_05 mlp= 13117 exact=51.33%
43
+ delta=+0.00% per1k=+0.000%
44
+ add__topk_2000_minus_topk_500__layer_09 mlp= 13419 exact=48.73%
45
+ delta=-2.60% per1k=-4.187%
46
+ add__topk_2000_minus_topk_500__layer_13 mlp= 13394 exact=51.60%
47
+ delta=+0.27% per1k=+0.447%
48
+ add__topk_2000_minus_topk_500__layer_17 mlp= 13530 exact=49.60%
49
+ delta=-1.73% per1k=-2.368%
50
+ add__topk_2000_minus_topk_500__layer_21 mlp= 15542 exact=52.80%
51
+ delta=+1.47% per1k=+0.534%
52
+ add__topk_2000_minus_topk_500__layer_25 mlp= 17385 exact=51.27%
53
+ delta=-0.07% per1k=-0.015%
54
+ add__rel_0.001_minus_topk_500__layer_01 mlp= 14976 exact=52.80%
55
+ delta=+1.47% per1k=+0.673%
56
+ add__rel_0.001_minus_topk_500__layer_05 mlp= 19350 exact=56.40%
57
+ delta=+5.07% per1k=+0.773%
58
+ add__rel_0.001_minus_topk_500__layer_09 mlp= 21532 exact=56.47%
59
+ delta=+5.13% per1k=+0.588%
60
+ add__rel_0.001_minus_topk_500__layer_13 mlp= 21290 exact=57.13%
61
+ delta=+5.80% per1k=+0.683%
62
+ add__rel_0.001_minus_topk_500__layer_17 mlp= 21458 exact=55.87%
63
+ delta=+4.53% per1k=+0.523%
64
+ add__rel_0.001_minus_topk_500__layer_21 mlp= 21205 exact=54.73%
65
+ delta=+3.40% per1k=+0.404%
66
+ add__rel_0.001_minus_topk_500__layer_25 mlp= 18804 exact=51.47%
67
+ delta=+0.13% per1k=+0.022%
68
+ [done] rank-only -> results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_2/direct_group_rank_shard1.json
circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_2_direct_rank_shard2.log ADDED
@@ -0,0 +1,68 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ [records] results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_2/position_adacs positions=hundreds,tens,ones
2
+ records=1,500 skipped=0
3
+ [model] results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_2/lora_r2_beta005_merged device=cuda dtype=bfloat16 batch=256
4
+
5
+ topk_100 mlp= 2025 exact=7.07%
6
+ rel_0.05 mlp= 2965 exact=12.40%
7
+ topk_500 mlp= 12798 exact=51.33%
8
+ rel_0.01 mlp= 24067 exact=43.93%
9
+ topk_2000 mlp= 51432 exact=34.00%
10
+ rel_0.001 mlp= 224312 exact=99.93%
11
+ [rank] todo=28 shard=2/4
12
+ drop__topk_500__layer_02 mlp= 12709 exact=38.20%
13
+ delta=-13.13% per1k=-147.566%
14
+ drop__topk_500__layer_06 mlp= 12669 exact=46.33%
15
+ delta=-5.00% per1k=-38.760%
16
+ drop__topk_500__layer_10 mlp= 12706 exact=54.47%
17
+ delta=+3.13% per1k=+34.058%
18
+ drop__topk_500__layer_14 mlp= 12673 exact=55.60%
19
+ delta=+4.27% per1k=+34.133%
20
+ drop__topk_500__layer_18 mlp= 12642 exact=41.93%
21
+ delta=-9.40% per1k=-60.256%
22
+ drop__topk_500__layer_22 mlp= 11865 exact=6.00%
23
+ delta=-45.33% per1k=-48.589%
24
+ drop__topk_500__layer_26 mlp= 11797 exact=50.13%
25
+ delta=-1.20% per1k=-1.199%
26
+ add__rel_0.01_minus_topk_500__layer_02 mlp= 12866 exact=52.67%
27
+ delta=+1.33% per1k=+19.608%
28
+ add__rel_0.01_minus_topk_500__layer_06 mlp= 12974 exact=50.80%
29
+ delta=-0.53% per1k=-3.030%
30
+ add__rel_0.01_minus_topk_500__layer_10 mlp= 12971 exact=50.93%
31
+ delta=-0.40% per1k=-2.312%
32
+ add__rel_0.01_minus_topk_500__layer_14 mlp= 12972 exact=47.93%
33
+ delta=-3.40% per1k=-19.540%
34
+ add__rel_0.01_minus_topk_500__layer_18 mlp= 13063 exact=52.53%
35
+ delta=+1.20% per1k=+4.528%
36
+ add__rel_0.01_minus_topk_500__layer_22 mlp= 13633 exact=52.20%
37
+ delta=+0.87% per1k=+1.038%
38
+ add__rel_0.01_minus_topk_500__layer_26 mlp= 13701 exact=51.20%
39
+ delta=-0.13% per1k=-0.148%
40
+ add__topk_2000_minus_topk_500__layer_02 mlp= 13049 exact=54.00%
41
+ delta=+2.67% per1k=+10.624%
42
+ add__topk_2000_minus_topk_500__layer_06 mlp= 13338 exact=51.13%
43
+ delta=-0.20% per1k=-0.370%
44
+ add__topk_2000_minus_topk_500__layer_10 mlp= 13365 exact=49.73%
45
+ delta=-1.60% per1k=-2.822%
46
+ add__topk_2000_minus_topk_500__layer_14 mlp= 13422 exact=46.53%
47
+ delta=-4.80% per1k=-7.692%
48
+ add__topk_2000_minus_topk_500__layer_18 mlp= 13747 exact=54.13%
49
+ delta=+2.80% per1k=+2.950%
50
+ add__topk_2000_minus_topk_500__layer_22 mlp= 16881 exact=52.13%
51
+ delta=+0.80% per1k=+0.196%
52
+ add__topk_2000_minus_topk_500__layer_26 mlp= 16025 exact=50.80%
53
+ delta=-0.53% per1k=-0.165%
54
+ add__rel_0.001_minus_topk_500__layer_02 mlp= 16855 exact=60.33%
55
+ delta=+9.00% per1k=+2.218%
56
+ add__rel_0.001_minus_topk_500__layer_06 mlp= 21548 exact=58.27%
57
+ delta=+6.93% per1k=+0.792%
58
+ add__rel_0.001_minus_topk_500__layer_10 mlp= 21530 exact=59.47%
59
+ delta=+8.13% per1k=+0.931%
60
+ add__rel_0.001_minus_topk_500__layer_14 mlp= 21290 exact=52.20%
61
+ delta=+0.87% per1k=+0.102%
62
+ add__rel_0.001_minus_topk_500__layer_18 mlp= 21371 exact=52.73%
63
+ delta=+1.40% per1k=+0.163%
64
+ add__rel_0.001_minus_topk_500__layer_22 mlp= 20824 exact=53.67%
65
+ delta=+2.33% per1k=+0.291%
66
+ add__rel_0.001_minus_topk_500__layer_26 mlp= 20752 exact=50.73%
67
+ delta=-0.60% per1k=-0.075%
68
+ [done] rank-only -> results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_2/direct_group_rank_shard2.json
circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_2_direct_rank_shard3.log ADDED
@@ -0,0 +1,68 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ [records] results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_2/position_adacs positions=hundreds,tens,ones
2
+ records=1,500 skipped=0
3
+ [model] results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_2/lora_r2_beta005_merged device=cuda dtype=bfloat16 batch=256
4
+
5
+ topk_100 mlp= 2025 exact=7.07%
6
+ rel_0.05 mlp= 2965 exact=12.40%
7
+ topk_500 mlp= 12798 exact=51.33%
8
+ rel_0.01 mlp= 24067 exact=43.93%
9
+ topk_2000 mlp= 51432 exact=34.00%
10
+ rel_0.001 mlp= 224312 exact=99.93%
11
+ [rank] todo=28 shard=3/4
12
+ drop__topk_500__layer_03 mlp= 12738 exact=39.33%
13
+ delta=-12.00% per1k=-200.000%
14
+ drop__topk_500__layer_07 mlp= 12673 exact=48.60%
15
+ delta=-2.73% per1k=-21.867%
16
+ drop__topk_500__layer_11 mlp= 12698 exact=45.07%
17
+ delta=-6.27% per1k=-62.667%
18
+ drop__topk_500__layer_15 mlp= 12704 exact=50.53%
19
+ delta=-0.80% per1k=-8.511%
20
+ drop__topk_500__layer_19 mlp= 12472 exact=27.27%
21
+ delta=-24.07% per1k=-73.824%
22
+ drop__topk_500__layer_23 mlp= 11389 exact=9.07%
23
+ delta=-42.27% per1k=-29.998%
24
+ drop__topk_500__layer_27 mlp= 12364 exact=47.93%
25
+ delta=-3.40% per1k=-7.834%
26
+ add__rel_0.01_minus_topk_500__layer_03 mlp= 12863 exact=50.40%
27
+ delta=-0.93% per1k=-14.359%
28
+ add__rel_0.01_minus_topk_500__layer_07 mlp= 12997 exact=50.67%
29
+ delta=-0.67% per1k=-3.350%
30
+ add__rel_0.01_minus_topk_500__layer_11 mlp= 12968 exact=50.00%
31
+ delta=-1.33% per1k=-7.843%
32
+ add__rel_0.01_minus_topk_500__layer_15 mlp= 12979 exact=52.40%
33
+ delta=+1.07% per1k=+5.893%
34
+ add__rel_0.01_minus_topk_500__layer_19 mlp= 13189 exact=52.07%
35
+ delta=+0.73% per1k=+1.876%
36
+ add__rel_0.01_minus_topk_500__layer_23 mlp= 13958 exact=51.13%
37
+ delta=-0.20% per1k=-0.172%
38
+ add__rel_0.01_minus_topk_500__layer_27 mlp= 13181 exact=51.27%
39
+ delta=-0.07% per1k=-0.174%
40
+ add__topk_2000_minus_topk_500__layer_03 mlp= 12961 exact=51.07%
41
+ delta=-0.27% per1k=-1.636%
42
+ add__topk_2000_minus_topk_500__layer_07 mlp= 13426 exact=49.13%
43
+ delta=-2.20% per1k=-3.503%
44
+ add__topk_2000_minus_topk_500__layer_11 mlp= 13319 exact=48.13%
45
+ delta=-3.20% per1k=-6.142%
46
+ add__topk_2000_minus_topk_500__layer_15 mlp= 13372 exact=52.40%
47
+ delta=+1.07% per1k=+1.858%
48
+ add__topk_2000_minus_topk_500__layer_19 mlp= 14153 exact=50.20%
49
+ delta=-1.13% per1k=-0.836%
50
+ add__topk_2000_minus_topk_500__layer_23 mlp= 17311 exact=52.27%
51
+ delta=+0.93% per1k=+0.207%
52
+ add__topk_2000_minus_topk_500__layer_27 mlp= 14242 exact=51.33%
53
+ delta=+0.00% per1k=+0.000%
54
+ add__rel_0.001_minus_topk_500__layer_03 mlp= 16571 exact=53.60%
55
+ delta=+2.27% per1k=+0.601%
56
+ add__rel_0.001_minus_topk_500__layer_07 mlp= 21507 exact=57.67%
57
+ delta=+6.33% per1k=+0.727%
58
+ add__rel_0.001_minus_topk_500__layer_11 mlp= 21417 exact=56.87%
59
+ delta=+5.53% per1k=+0.642%
60
+ add__rel_0.001_minus_topk_500__layer_15 mlp= 21178 exact=55.00%
61
+ delta=+3.67% per1k=+0.438%
62
+ add__rel_0.001_minus_topk_500__layer_19 mlp= 21351 exact=58.53%
63
+ delta=+7.20% per1k=+0.842%
64
+ add__rel_0.001_minus_topk_500__layer_23 mlp= 20349 exact=52.07%
65
+ delta=+0.73% per1k=+0.097%
66
+ add__rel_0.001_minus_topk_500__layer_27 mlp= 20718 exact=51.53%
67
+ delta=+0.20% per1k=+0.025%
68
+ [done] rank-only -> results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_2/direct_group_rank_shard3.json
circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_2_fresh_build.log ADDED
@@ -0,0 +1,7 @@
 
 
 
 
 
 
 
 
1
+ [dataset] data/qwen25_math_1p5b_rank_ladder_r2_correct_addition.json
2
+ [pairing] loading tokenizer for prompt-length matching
3
+ enriched examples=8,078
4
+ hundreds pairs= 3156 buckets= 79
5
+ tens pairs= 4000 buckets= 20
6
+ ones pairs= 4000 buckets= 18
7
+ [done] build-only artifacts -> results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_2/fresh_pairs_seed123
circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_2_fresh_target90_eval.log ADDED
@@ -0,0 +1,6 @@
 
 
 
 
 
 
 
1
+ [records] results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_2/fresh_pairs_seed123
2
+ records=1,500 skipped=0
3
+ [model] results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_2/lora_r2_beta005_merged device=cuda dtype=bfloat16 batch=256
4
+
5
+ target90 mlp= 171233 exact=88.47%
6
+ [done] results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_2/fresh_target90_eval.json
circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_2_fresh_target90_generation.log ADDED
@@ -0,0 +1,4 @@
 
 
 
 
 
1
+ [records] 1,500 skipped=0
2
+
3
+ target90 mlp= 171233 gen_exact=88.73%
4
+ [done] results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_2/fresh_target90_generation.json
circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_2_merge.log ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ Warning: You are sending unauthenticated requests to the HF Hub. Please set a HF_TOKEN to enable higher rate limits and faster downloads.
2
+ [model] Qwen/Qwen2.5-Math-1.5B adapter=results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_2/lora_r2_beta005/adapter device=cuda
3
+
4
+ scale=0.45 kl=0.768908 acc=99.32% correct=8045/8100
5
+ scale=0.55 kl=0.874731 acc=99.96% correct=8097/8100
6
+ scale=0.75 kl=0.981092 acc=99.99% correct=8099/8100
7
+ scale=1 kl=1.09497 acc=100.00% correct=8100/8100
8
+
9
+ [done] selected scale=0.55 rule=lowest_kl_at_or_above_min_accuracy merged=results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_2/lora_r2_beta005_merged
circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_2_position_trace.log ADDED
@@ -0,0 +1,64 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ [worker gpu=0 positions=hundreds]
2
+ [dataset] data/qwen25_math_1p5b_rank_ladder_r2_correct_addition.json
3
+ [pairing] loading tokenizer for prompt-length matching
4
+ enriched examples=8,078
5
+ hundreds pairs= 3156 buckets= 79
6
+ [model] results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_2/lora_r2_beta005_merged
7
+ device=cuda dtype=bfloat16 batch_size=256
8
+
9
+ Loading weights: 0%| | 0/338 [00:00<?, ?it/s]
10
+ Loading weights: 100%|██████████| 338/338 [00:00<00:00, 6336.49it/s]
11
+ layers=28 d_ffn=8960 heads=12
12
+ [hundreds] attr_pairs=2656 test_pairs=500
13
+ hundreds positive mlp= 247901 zero=100.00% cf=99.80%
14
+ hundreds rel_0.001 mlp= 187956 zero=99.80% cf=99.20%
15
+ hundreds rel_0.01 mlp= 10928 zero=47.20% cf=88.20%
16
+ hundreds rel_0.05 mlp= 1311 zero=33.40% cf=2.80%
17
+ hundreds topk_100 mlp= 870 zero=3.00% cf=2.00%
18
+ hundreds topk_500 mlp= 5601 zero=49.20% cf=46.60%
19
+ hundreds topk_2000 mlp= 24762 zero=90.20% cf=75.80%
20
+ [done] artifacts -> results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_2/position_adacs/_workers/gpu0
21
+
22
+ [worker gpu=1 positions=tens]
23
+ [dataset] data/qwen25_math_1p5b_rank_ladder_r2_correct_addition.json
24
+ [pairing] loading tokenizer for prompt-length matching
25
+ enriched examples=8,078
26
+ tens pairs= 4000 buckets= 20
27
+ [model] results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_2/lora_r2_beta005_merged
28
+ device=cuda dtype=bfloat16 batch_size=256
29
+
30
+ Loading weights: 0%| | 0/338 [00:00<?, ?it/s]
31
+ Loading weights: 100%|██████████| 338/338 [00:00<00:00, 7371.86it/s]
32
+ layers=28 d_ffn=8960 heads=12
33
+ [tens] attr_pairs=3500 test_pairs=500
34
+ tens positive mlp= 250283 zero=99.80% cf=99.80%
35
+ tens rel_0.001 mlp= 208667 zero=100.00% cf=100.00%
36
+ tens rel_0.01 mlp= 17578 zero=13.20% cf=69.00%
37
+ tens rel_0.05 mlp= 2195 zero=15.40% cf=28.40%
38
+ tens topk_100 mlp= 1541 zero=13.40% cf=11.80%
39
+ tens topk_500 mlp= 9524 zero=11.80% cf=75.80%
40
+ tens topk_2000 mlp= 40538 zero=14.40% cf=55.20%
41
+ [done] artifacts -> results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_2/position_adacs/_workers/gpu1
42
+
43
+ [worker gpu=2 positions=ones]
44
+ [dataset] data/qwen25_math_1p5b_rank_ladder_r2_correct_addition.json
45
+ [pairing] loading tokenizer for prompt-length matching
46
+ enriched examples=8,078
47
+ ones pairs= 4000 buckets= 18
48
+ [model] results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_2/lora_r2_beta005_merged
49
+ device=cuda dtype=bfloat16 batch_size=256
50
+
51
+ Loading weights: 0%| | 0/338 [00:00<?, ?it/s]
52
+ Loading weights: 100%|██████████| 338/338 [00:00<00:00, 6705.68it/s]
53
+ layers=28 d_ffn=8960 heads=12
54
+ [ones] attr_pairs=3500 test_pairs=500
55
+ ones positive mlp= 250280 zero=100.00% cf=100.00%
56
+ ones rel_0.001 mlp= 219808 zero=100.00% cf=100.00%
57
+ ones rel_0.01 mlp= 15401 zero=18.60% cf=75.40%
58
+ ones rel_0.05 mlp= 1968 zero=13.80% cf=28.80%
59
+ ones topk_100 mlp= 1437 zero=8.20% cf=25.00%
60
+ ones topk_500 mlp= 7528 zero=17.40% cf=82.00%
61
+ ones topk_2000 mlp= 37223 zero=23.00% cf=78.20%
62
+ [done] artifacts -> results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_2/position_adacs/_workers/gpu2
63
+
64
+ [done] merged multi-gpu artifacts -> results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_2/position_adacs
circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_2_prep_stage.log ADDED
@@ -0,0 +1,5 @@
 
 
 
 
 
 
1
+ [rank 2] prep start on gpu 3
2
+ [rank 2] train cached
3
+ [rank 2] merge cached
4
+ [rank 2] correct dataset cached
5
+ [rank 2] prep done
circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_2_target90_search.log ADDED
@@ -0,0 +1,377 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+
2
+ topk_100 mlp= 2025 exact=7.07%
3
+ rel_0.05 mlp= 2965 exact=12.40%
4
+ topk_500 mlp= 12798 exact=51.33%
5
+ rel_0.01 mlp= 24067 exact=43.93%
6
+ topk_2000 mlp= 51432 exact=34.00%
7
+ rel_0.001 mlp= 224312 exact=99.93%
8
+ add_step_1_add__rel_0.001_minus_topk_500__layer_08 mlp= 21549 exact=62.87%
9
+ add_step_1_add__rel_0.001_minus_topk_500__layer_04 mlp= 20537 exact=60.67%
10
+ add_step_1_add__rel_0.001_minus_topk_500__layer_02 mlp= 16855 exact=60.33%
11
+ add_step_1_add__rel_0.001_minus_topk_500__layer_10 mlp= 21530 exact=59.47%
12
+ add_step_1_add__rel_0.001_minus_topk_500__layer_19 mlp= 21351 exact=58.53%
13
+ add_step_1_add__rel_0.001_minus_topk_500__layer_06 mlp= 21548 exact=58.27%
14
+ add_step_1_add__rel_0.001_minus_topk_500__layer_07 mlp= 21507 exact=57.67%
15
+ add_step_1_add__rel_0.001_minus_topk_500__layer_13 mlp= 21290 exact=57.13%
16
+ add_step_1_add__rel_0.001_minus_topk_500__layer_11 mlp= 21417 exact=56.87%
17
+ add_step_1_add__rel_0.001_minus_topk_500__layer_09 mlp= 21532 exact=56.47%
18
+ add_step_1_add__rel_0.001_minus_topk_500__layer_05 mlp= 19350 exact=56.40%
19
+ add_step_1_add__rel_0.001_minus_topk_500__layer_17 mlp= 21458 exact=55.87%
20
+ add_step_1_add__rel_0.001_minus_topk_500__layer_12 mlp= 21339 exact=55.60%
21
+ add_step_1_add__rel_0.001_minus_topk_500__layer_16 mlp= 21383 exact=55.07%
22
+ add_step_1_add__rel_0.001_minus_topk_500__layer_15 mlp= 21178 exact=55.00%
23
+ add_step_1_add__rel_0.001_minus_topk_500__layer_21 mlp= 21205 exact=54.73%
24
+ add_step_1_add__rel_0.001_minus_topk_500__layer_20 mlp= 21350 exact=54.53%
25
+ add_step_1_add__topk_2000_minus_topk_500__layer_18 mlp= 13747 exact=54.13%
26
+ add_step_1_add__rel_0.001_minus_topk_500__layer_00 mlp= 18726 exact=54.07%
27
+ add_step_1_add__topk_2000_minus_topk_500__layer_02 mlp= 13049 exact=54.00%
28
+ add_step_1_add__rel_0.001_minus_topk_500__layer_22 mlp= 20824 exact=53.67%
29
+ add_step_1_add__rel_0.001_minus_topk_500__layer_03 mlp= 16571 exact=53.60%
30
+ add_step_1_add__rel_0.001_minus_topk_500__layer_01 mlp= 14976 exact=52.80%
31
+ add_step_1_add__topk_2000_minus_topk_500__layer_21 mlp= 15542 exact=52.80%
32
+ add_step_1_add__rel_0.001_minus_topk_500__layer_18 mlp= 21371 exact=52.73%
33
+ add_step_1_add__rel_0.01_minus_topk_500__layer_02 mlp= 12866 exact=52.67%
34
+ add_step_1_add__rel_0.01_minus_topk_500__layer_18 mlp= 13063 exact=52.53%
35
+ add_step_1_add__rel_0.01_minus_topk_500__layer_15 mlp= 12979 exact=52.40%
36
+ add_step_1_add__topk_2000_minus_topk_500__layer_15 mlp= 13372 exact=52.40%
37
+ add_step_1_add__rel_0.01_minus_topk_500__layer_21 mlp= 13436 exact=52.40%
38
+ add_step_1_add__topk_2000_minus_topk_500__layer_23 mlp= 17311 exact=52.27%
39
+ add_step_1_add__rel_0.01_minus_topk_500__layer_22 mlp= 13633 exact=52.20%
40
+ add_step_1_add__rel_0.001_minus_topk_500__layer_14 mlp= 21290 exact=52.20%
41
+ add_step_1_add__topk_2000_minus_topk_500__layer_22 mlp= 16881 exact=52.13%
42
+ add_step_1_add__rel_0.01_minus_topk_500__layer_19 mlp= 13189 exact=52.07%
43
+ add_step_1_add__rel_0.001_minus_topk_500__layer_23 mlp= 20349 exact=52.07%
44
+ add_step_1_add__rel_0.01_minus_topk_500__layer_20 mlp= 13301 exact=52.00%
45
+ add_step_1_add__topk_2000_minus_topk_500__layer_20 mlp= 15012 exact=51.73%
46
+ add_step_1_add__topk_2000_minus_topk_500__layer_13 mlp= 13394 exact=51.60%
47
+ add_step_1_add__rel_0.01_minus_topk_500__layer_01 mlp= 12848 exact=51.53%
48
+ add_step_1_add__rel_0.001_minus_topk_500__layer_27 mlp= 20718 exact=51.53%
49
+ add_step_1_add__topk_2000_minus_topk_500__layer_01 mlp= 12967 exact=51.47%
50
+ add_step_1_add__rel_0.01_minus_topk_500__layer_24 mlp= 14498 exact=51.47%
51
+ add_step_1_add__rel_0.001_minus_topk_500__layer_25 mlp= 18804 exact=51.47%
52
+ add_step_1_add__rel_0.001_minus_topk_500__layer_24 mlp= 19098 exact=51.47%
53
+ add_step_1_add__topk_2000_minus_topk_500__layer_24 mlp= 17516 exact=51.40%
54
+ add_step_1_add__topk_2000_minus_topk_500__layer_00 mlp= 13063 exact=51.33%
55
+ add_step_1_add__topk_2000_minus_topk_500__layer_05 mlp= 13117 exact=51.33%
56
+ add_step_1_accepted mlp= 21549 exact=62.87%
57
+ add_step_2_add__rel_0.001_minus_topk_500__layer_04 mlp= 29288 exact=72.13%
58
+ add_step_2_add__rel_0.001_minus_topk_500__layer_02 mlp= 25606 exact=71.47%
59
+ add_step_2_add__rel_0.001_minus_topk_500__layer_10 mlp= 30281 exact=70.27%
60
+ add_step_2_add__rel_0.001_minus_topk_500__layer_19 mlp= 30102 exact=73.93%
61
+ add_step_2_add__rel_0.001_minus_topk_500__layer_06 mlp= 30299 exact=67.13%
62
+ add_step_2_add__rel_0.001_minus_topk_500__layer_07 mlp= 30258 exact=67.47%
63
+ add_step_2_add__rel_0.001_minus_topk_500__layer_13 mlp= 30041 exact=73.13%
64
+ add_step_2_add__rel_0.001_minus_topk_500__layer_11 mlp= 30168 exact=70.60%
65
+ add_step_2_add__rel_0.001_minus_topk_500__layer_09 mlp= 30283 exact=69.80%
66
+ add_step_2_add__rel_0.001_minus_topk_500__layer_05 mlp= 28101 exact=68.47%
67
+ add_step_2_add__rel_0.001_minus_topk_500__layer_17 mlp= 30209 exact=68.67%
68
+ add_step_2_add__rel_0.001_minus_topk_500__layer_12 mlp= 30090 exact=73.60%
69
+ add_step_2_add__rel_0.001_minus_topk_500__layer_16 mlp= 30134 exact=68.33%
70
+ add_step_2_add__rel_0.001_minus_topk_500__layer_15 mlp= 29929 exact=71.93%
71
+ add_step_2_add__rel_0.001_minus_topk_500__layer_21 mlp= 29956 exact=66.40%
72
+ add_step_2_add__rel_0.001_minus_topk_500__layer_20 mlp= 30101 exact=67.60%
73
+ add_step_2_add__topk_2000_minus_topk_500__layer_18 mlp= 22498 exact=66.20%
74
+ add_step_2_add__rel_0.001_minus_topk_500__layer_00 mlp= 27477 exact=68.13%
75
+ add_step_2_add__topk_2000_minus_topk_500__layer_02 mlp= 21800 exact=66.13%
76
+ add_step_2_add__rel_0.001_minus_topk_500__layer_22 mlp= 29575 exact=65.00%
77
+ add_step_2_add__rel_0.001_minus_topk_500__layer_03 mlp= 25322 exact=64.67%
78
+ add_step_2_add__rel_0.001_minus_topk_500__layer_01 mlp= 23727 exact=64.67%
79
+ add_step_2_add__topk_2000_minus_topk_500__layer_21 mlp= 24293 exact=65.73%
80
+ add_step_2_add__rel_0.001_minus_topk_500__layer_18 mlp= 30122 exact=66.33%
81
+ add_step_2_add__rel_0.01_minus_topk_500__layer_02 mlp= 21617 exact=65.27%
82
+ add_step_2_add__rel_0.01_minus_topk_500__layer_18 mlp= 21814 exact=64.07%
83
+ add_step_2_add__rel_0.01_minus_topk_500__layer_15 mlp= 21730 exact=64.20%
84
+ add_step_2_add__topk_2000_minus_topk_500__layer_15 mlp= 22123 exact=64.33%
85
+ add_step_2_add__rel_0.01_minus_topk_500__layer_21 mlp= 22187 exact=64.20%
86
+ add_step_2_add__topk_2000_minus_topk_500__layer_23 mlp= 26062 exact=63.47%
87
+ add_step_2_add__rel_0.01_minus_topk_500__layer_22 mlp= 22384 exact=63.80%
88
+ add_step_2_add__rel_0.001_minus_topk_500__layer_14 mlp= 30041 exact=64.80%
89
+ add_step_2_add__topk_2000_minus_topk_500__layer_22 mlp= 25632 exact=64.67%
90
+ add_step_2_add__rel_0.01_minus_topk_500__layer_19 mlp= 21940 exact=65.73%
91
+ add_step_2_add__rel_0.001_minus_topk_500__layer_23 mlp= 29100 exact=63.93%
92
+ add_step_2_add__rel_0.01_minus_topk_500__layer_20 mlp= 22052 exact=64.87%
93
+ add_step_2_add__topk_2000_minus_topk_500__layer_20 mlp= 23763 exact=64.27%
94
+ add_step_2_add__topk_2000_minus_topk_500__layer_13 mlp= 22145 exact=64.33%
95
+ add_step_2_add__rel_0.01_minus_topk_500__layer_01 mlp= 21599 exact=62.60%
96
+ add_step_2_add__rel_0.001_minus_topk_500__layer_27 mlp= 29469 exact=62.93%
97
+ add_step_2_add__topk_2000_minus_topk_500__layer_01 mlp= 21718 exact=63.40%
98
+ add_step_2_add__rel_0.01_minus_topk_500__layer_24 mlp= 23249 exact=62.73%
99
+ add_step_2_add__rel_0.001_minus_topk_500__layer_25 mlp= 27555 exact=63.00%
100
+ add_step_2_add__rel_0.001_minus_topk_500__layer_24 mlp= 27849 exact=63.27%
101
+ add_step_2_add__topk_2000_minus_topk_500__layer_24 mlp= 26267 exact=63.13%
102
+ add_step_2_add__topk_2000_minus_topk_500__layer_00 mlp= 21814 exact=64.00%
103
+ add_step_2_add__topk_2000_minus_topk_500__layer_05 mlp= 21868 exact=62.20%
104
+ add_step_2_accepted mlp= 30102 exact=73.93%
105
+ add_step_3_add__rel_0.001_minus_topk_500__layer_04 mlp= 37841 exact=82.20%
106
+ add_step_3_add__rel_0.001_minus_topk_500__layer_02 mlp= 34159 exact=81.47%
107
+ add_step_3_add__rel_0.001_minus_topk_500__layer_10 mlp= 38834 exact=79.87%
108
+ add_step_3_add__rel_0.001_minus_topk_500__layer_06 mlp= 38852 exact=78.73%
109
+ add_step_3_add__rel_0.001_minus_topk_500__layer_07 mlp= 38811 exact=76.53%
110
+ add_step_3_add__rel_0.001_minus_topk_500__layer_13 mlp= 38594 exact=81.40%
111
+ add_step_3_add__rel_0.001_minus_topk_500__layer_11 mlp= 38721 exact=79.27%
112
+ add_step_3_add__rel_0.001_minus_topk_500__layer_09 mlp= 38836 exact=80.60%
113
+ add_step_3_add__rel_0.001_minus_topk_500__layer_05 mlp= 36654 exact=78.60%
114
+ add_step_3_add__rel_0.001_minus_topk_500__layer_17 mlp= 38762 exact=77.20%
115
+ add_step_3_add__rel_0.001_minus_topk_500__layer_12 mlp= 38643 exact=80.67%
116
+ add_step_3_add__rel_0.001_minus_topk_500__layer_16 mlp= 38687 exact=77.87%
117
+ add_step_3_add__rel_0.001_minus_topk_500__layer_15 mlp= 38482 exact=77.87%
118
+ add_step_3_add__rel_0.001_minus_topk_500__layer_21 mlp= 38509 exact=76.60%
119
+ add_step_3_add__rel_0.001_minus_topk_500__layer_20 mlp= 38654 exact=76.67%
120
+ add_step_3_add__topk_2000_minus_topk_500__layer_18 mlp= 31051 exact=73.47%
121
+ add_step_3_add__rel_0.001_minus_topk_500__layer_00 mlp= 36030 exact=76.33%
122
+ add_step_3_add__topk_2000_minus_topk_500__layer_02 mlp= 30353 exact=77.27%
123
+ add_step_3_add__rel_0.001_minus_topk_500__layer_22 mlp= 38128 exact=75.27%
124
+ add_step_3_add__rel_0.001_minus_topk_500__layer_03 mlp= 33875 exact=77.87%
125
+ add_step_3_add__rel_0.001_minus_topk_500__layer_01 mlp= 32280 exact=75.80%
126
+ add_step_3_add__topk_2000_minus_topk_500__layer_21 mlp= 32846 exact=75.93%
127
+ add_step_3_add__rel_0.001_minus_topk_500__layer_18 mlp= 38675 exact=74.93%
128
+ add_step_3_add__rel_0.01_minus_topk_500__layer_02 mlp= 30170 exact=75.47%
129
+ add_step_3_add__rel_0.01_minus_topk_500__layer_18 mlp= 30367 exact=73.60%
130
+ add_step_3_add__rel_0.01_minus_topk_500__layer_15 mlp= 30283 exact=72.33%
131
+ add_step_3_add__topk_2000_minus_topk_500__layer_15 mlp= 30676 exact=72.80%
132
+ add_step_3_add__rel_0.01_minus_topk_500__layer_21 mlp= 30740 exact=75.53%
133
+ add_step_3_add__topk_2000_minus_topk_500__layer_23 mlp= 34615 exact=73.53%
134
+ add_step_3_add__rel_0.01_minus_topk_500__layer_22 mlp= 30937 exact=74.20%
135
+ add_step_3_add__rel_0.001_minus_topk_500__layer_14 mlp= 38594 exact=74.73%
136
+ add_step_3_add__topk_2000_minus_topk_500__layer_22 mlp= 34185 exact=75.00%
137
+ add_step_3_add__rel_0.001_minus_topk_500__layer_23 mlp= 37653 exact=74.00%
138
+ add_step_3_add__rel_0.01_minus_topk_500__layer_20 mlp= 30605 exact=74.73%
139
+ add_step_3_add__topk_2000_minus_topk_500__layer_20 mlp= 32316 exact=74.07%
140
+ add_step_3_add__topk_2000_minus_topk_500__layer_13 mlp= 30698 exact=75.27%
141
+ add_step_3_add__rel_0.01_minus_topk_500__layer_01 mlp= 30152 exact=73.73%
142
+ add_step_3_add__rel_0.001_minus_topk_500__layer_27 mlp= 38022 exact=74.27%
143
+ add_step_3_add__topk_2000_minus_topk_500__layer_01 mlp= 30271 exact=74.40%
144
+ add_step_3_add__rel_0.01_minus_topk_500__layer_24 mlp= 31802 exact=73.73%
145
+ add_step_3_add__rel_0.001_minus_topk_500__layer_25 mlp= 36108 exact=73.40%
146
+ add_step_3_add__rel_0.001_minus_topk_500__layer_24 mlp= 36402 exact=73.53%
147
+ add_step_3_add__topk_2000_minus_topk_500__layer_24 mlp= 34820 exact=73.87%
148
+ add_step_3_add__topk_2000_minus_topk_500__layer_00 mlp= 30367 exact=74.40%
149
+ add_step_3_add__topk_2000_minus_topk_500__layer_05 mlp= 30421 exact=73.80%
150
+ add_step_3_accepted mlp= 37841 exact=82.20%
151
+ add_step_4_add__rel_0.001_minus_topk_500__layer_02 mlp= 41898 exact=87.00%
152
+ add_step_4_add__rel_0.001_minus_topk_500__layer_10 mlp= 46573 exact=86.67%
153
+ add_step_4_add__rel_0.001_minus_topk_500__layer_06 mlp= 46591 exact=82.13%
154
+ add_step_4_add__rel_0.001_minus_topk_500__layer_07 mlp= 46550 exact=83.53%
155
+ add_step_4_add__rel_0.001_minus_topk_500__layer_13 mlp= 46333 exact=87.53%
156
+ add_step_4_add__rel_0.001_minus_topk_500__layer_11 mlp= 46460 exact=86.80%
157
+ add_step_4_add__rel_0.001_minus_topk_500__layer_09 mlp= 46575 exact=86.20%
158
+ add_step_4_add__rel_0.001_minus_topk_500__layer_05 mlp= 44393 exact=85.40%
159
+ add_step_4_add__rel_0.001_minus_topk_500__layer_17 mlp= 46501 exact=83.93%
160
+ add_step_4_add__rel_0.001_minus_topk_500__layer_12 mlp= 46382 exact=88.20%
161
+ add_step_4_add__rel_0.001_minus_topk_500__layer_16 mlp= 46426 exact=85.53%
162
+ add_step_4_add__rel_0.001_minus_topk_500__layer_15 mlp= 46221 exact=85.40%
163
+ add_step_4_add__rel_0.001_minus_topk_500__layer_21 mlp= 46248 exact=84.67%
164
+ add_step_4_add__rel_0.001_minus_topk_500__layer_20 mlp= 46393 exact=85.20%
165
+ add_step_4_add__topk_2000_minus_topk_500__layer_18 mlp= 38790 exact=82.00%
166
+ add_step_4_add__rel_0.001_minus_topk_500__layer_00 mlp= 43769 exact=82.13%
167
+ add_step_4_add__topk_2000_minus_topk_500__layer_02 mlp= 38092 exact=84.07%
168
+ add_step_4_add__rel_0.001_minus_topk_500__layer_22 mlp= 45867 exact=83.40%
169
+ add_step_4_add__rel_0.001_minus_topk_500__layer_03 mlp= 41614 exact=85.20%
170
+ add_step_4_add__rel_0.001_minus_topk_500__layer_01 mlp= 40019 exact=83.07%
171
+ add_step_4_add__topk_2000_minus_topk_500__layer_21 mlp= 40585 exact=83.93%
172
+ add_step_4_add__rel_0.001_minus_topk_500__layer_18 mlp= 46414 exact=83.73%
173
+ add_step_4_add__rel_0.01_minus_topk_500__layer_02 mlp= 37909 exact=82.87%
174
+ add_step_4_add__rel_0.01_minus_topk_500__layer_18 mlp= 38106 exact=82.27%
175
+ add_step_4_add__rel_0.01_minus_topk_500__layer_15 mlp= 38022 exact=81.73%
176
+ add_step_4_add__topk_2000_minus_topk_500__layer_15 mlp= 38415 exact=81.93%
177
+ add_step_4_add__rel_0.01_minus_topk_500__layer_21 mlp= 38479 exact=82.40%
178
+ add_step_4_add__topk_2000_minus_topk_500__layer_23 mlp= 42354 exact=81.93%
179
+ add_step_4_add__rel_0.01_minus_topk_500__layer_22 mlp= 38676 exact=82.87%
180
+ add_step_4_add__rel_0.001_minus_topk_500__layer_14 mlp= 46333 exact=82.13%
181
+ add_step_4_add__topk_2000_minus_topk_500__layer_22 mlp= 41924 exact=83.13%
182
+ add_step_4_add__rel_0.001_minus_topk_500__layer_23 mlp= 45392 exact=82.47%
183
+ add_step_4_add__rel_0.01_minus_topk_500__layer_20 mlp= 38344 exact=83.00%
184
+ add_step_4_add__topk_2000_minus_topk_500__layer_20 mlp= 40055 exact=83.27%
185
+ add_step_4_add__topk_2000_minus_topk_500__layer_13 mlp= 38437 exact=81.47%
186
+ add_step_4_add__rel_0.01_minus_topk_500__layer_01 mlp= 37891 exact=81.60%
187
+ add_step_4_add__rel_0.001_minus_topk_500__layer_27 mlp= 45761 exact=82.27%
188
+ add_step_4_add__topk_2000_minus_topk_500__layer_01 mlp= 38010 exact=82.40%
189
+ add_step_4_add__rel_0.01_minus_topk_500__layer_24 mlp= 39541 exact=82.33%
190
+ add_step_4_add__rel_0.001_minus_topk_500__layer_25 mlp= 43847 exact=81.87%
191
+ add_step_4_add__rel_0.001_minus_topk_500__layer_24 mlp= 44141 exact=82.33%
192
+ add_step_4_add__topk_2000_minus_topk_500__layer_24 mlp= 42559 exact=82.27%
193
+ add_step_4_add__topk_2000_minus_topk_500__layer_00 mlp= 38106 exact=82.27%
194
+ add_step_4_add__topk_2000_minus_topk_500__layer_05 mlp= 38160 exact=81.80%
195
+ add_step_4_accepted mlp= 46382 exact=88.20%
196
+ baseline_rel_0.001_compress_1_drop_00 mlp= 218286 exact=21.67%
197
+ baseline_rel_0.001_compress_1_drop_01 mlp= 222052 exact=70.53%
198
+ baseline_rel_0.001_compress_1_drop_02 mlp= 220166 exact=96.93%
199
+ baseline_rel_0.001_compress_1_drop_03 mlp= 220479 exact=93.73%
200
+ baseline_rel_0.001_compress_1_drop_04 mlp= 216388 exact=88.87%
201
+ baseline_rel_0.001_compress_1_drop_05 mlp= 217645 exact=99.33%
202
+ baseline_rel_0.001_compress_1_drop_06 mlp= 215433 exact=99.47%
203
+ baseline_rel_0.001_compress_1_drop_07 mlp= 215478 exact=99.80%
204
+ baseline_rel_0.001_compress_1_drop_08 mlp= 215440 exact=100.00%
205
+ baseline_rel_0.001_compress_1_drop_09 mlp= 215471 exact=100.00%
206
+ baseline_rel_0.001_compress_1_drop_10 mlp= 215488 exact=99.80%
207
+ baseline_rel_0.001_compress_1_drop_11 mlp= 215593 exact=98.87%
208
+ baseline_rel_0.001_compress_1_drop_12 mlp= 215646 exact=98.60%
209
+ baseline_rel_0.001_compress_1_drop_13 mlp= 215728 exact=99.87%
210
+ baseline_rel_0.001_compress_1_drop_14 mlp= 215695 exact=99.93%
211
+ baseline_rel_0.001_compress_1_drop_15 mlp= 215838 exact=99.87%
212
+ baseline_rel_0.001_compress_1_drop_16 mlp= 215609 exact=100.00%
213
+ baseline_rel_0.001_compress_1_drop_17 mlp= 215534 exact=99.93%
214
+ baseline_rel_0.001_compress_1_drop_18 mlp= 215583 exact=99.20%
215
+ baseline_rel_0.001_compress_1_drop_19 mlp= 215433 exact=98.13%
216
+ baseline_rel_0.001_compress_1_drop_20 mlp= 215357 exact=96.47%
217
+ baseline_rel_0.001_compress_1_drop_21 mlp= 215357 exact=75.13%
218
+ baseline_rel_0.001_compress_1_drop_22 mlp= 215353 exact=11.40%
219
+ baseline_rel_0.001_compress_1_drop_23 mlp= 215352 exact=15.60%
220
+ baseline_rel_0.001_compress_1_drop_24 mlp= 215352 exact=93.93%
221
+ baseline_rel_0.001_compress_1_drop_25 mlp= 215353 exact=85.00%
222
+ baseline_rel_0.001_compress_1_drop_26 mlp= 215357 exact=99.60%
223
+ baseline_rel_0.001_compress_1_drop_27 mlp= 215958 exact=99.87%
224
+ baseline_rel_0.001_compress_1_accepted mlp= 215352 exact=93.93%
225
+ baseline_rel_0.001_compress_2_drop_00 mlp= 209326 exact=17.93%
226
+ baseline_rel_0.001_compress_2_drop_01 mlp= 213092 exact=67.60%
227
+ baseline_rel_0.001_compress_2_drop_02 mlp= 211206 exact=88.80%
228
+ baseline_rel_0.001_compress_2_drop_03 mlp= 211519 exact=85.80%
229
+ baseline_rel_0.001_compress_2_drop_04 mlp= 207428 exact=77.07%
230
+ baseline_rel_0.001_compress_2_drop_05 mlp= 208685 exact=91.73%
231
+ baseline_rel_0.001_compress_2_drop_06 mlp= 206473 exact=93.13%
232
+ baseline_rel_0.001_compress_2_drop_07 mlp= 206518 exact=93.07%
233
+ baseline_rel_0.001_compress_2_drop_08 mlp= 206480 exact=93.93%
234
+ baseline_rel_0.001_compress_2_drop_09 mlp= 206511 exact=94.60%
235
+ baseline_rel_0.001_compress_2_drop_10 mlp= 206528 exact=93.87%
236
+ baseline_rel_0.001_compress_2_drop_11 mlp= 206633 exact=90.40%
237
+ baseline_rel_0.001_compress_2_drop_12 mlp= 206686 exact=90.60%
238
+ baseline_rel_0.001_compress_2_drop_13 mlp= 206768 exact=94.40%
239
+ baseline_rel_0.001_compress_2_drop_14 mlp= 206735 exact=93.93%
240
+ baseline_rel_0.001_compress_2_drop_15 mlp= 206878 exact=93.93%
241
+ baseline_rel_0.001_compress_2_drop_16 mlp= 206649 exact=93.87%
242
+ baseline_rel_0.001_compress_2_drop_17 mlp= 206574 exact=93.73%
243
+ baseline_rel_0.001_compress_2_drop_18 mlp= 206623 exact=90.87%
244
+ baseline_rel_0.001_compress_2_drop_19 mlp= 206473 exact=84.60%
245
+ baseline_rel_0.001_compress_2_drop_20 mlp= 206397 exact=86.27%
246
+ baseline_rel_0.001_compress_2_drop_21 mlp= 206397 exact=67.20%
247
+ baseline_rel_0.001_compress_2_drop_22 mlp= 206393 exact=9.47%
248
+ baseline_rel_0.001_compress_2_drop_23 mlp= 206392 exact=7.73%
249
+ baseline_rel_0.001_compress_2_drop_25 mlp= 206393 exact=54.93%
250
+ baseline_rel_0.001_compress_2_drop_26 mlp= 206397 exact=85.00%
251
+ baseline_rel_0.001_compress_2_drop_27 mlp= 206998 exact=90.33%
252
+ baseline_rel_0.001_compress_2_accepted mlp= 206473 exact=93.13%
253
+ baseline_rel_0.001_compress_3_drop_00 mlp= 200447 exact=9.47%
254
+ baseline_rel_0.001_compress_3_drop_01 mlp= 204213 exact=60.67%
255
+ baseline_rel_0.001_compress_3_drop_02 mlp= 202327 exact=81.40%
256
+ baseline_rel_0.001_compress_3_drop_03 mlp= 202640 exact=79.07%
257
+ baseline_rel_0.001_compress_3_drop_04 mlp= 198549 exact=55.93%
258
+ baseline_rel_0.001_compress_3_drop_05 mlp= 199806 exact=91.40%
259
+ baseline_rel_0.001_compress_3_drop_07 mlp= 197639 exact=89.00%
260
+ baseline_rel_0.001_compress_3_drop_08 mlp= 197601 exact=91.67%
261
+ baseline_rel_0.001_compress_3_drop_09 mlp= 197632 exact=92.80%
262
+ baseline_rel_0.001_compress_3_drop_10 mlp= 197649 exact=93.80%
263
+ baseline_rel_0.001_compress_3_drop_11 mlp= 197754 exact=87.60%
264
+ baseline_rel_0.001_compress_3_drop_12 mlp= 197807 exact=88.53%
265
+ baseline_rel_0.001_compress_3_drop_13 mlp= 197889 exact=92.87%
266
+ baseline_rel_0.001_compress_3_drop_14 mlp= 197856 exact=92.33%
267
+ baseline_rel_0.001_compress_3_drop_15 mlp= 197999 exact=93.27%
268
+ baseline_rel_0.001_compress_3_drop_16 mlp= 197770 exact=92.87%
269
+ baseline_rel_0.001_compress_3_drop_17 mlp= 197695 exact=92.20%
270
+ baseline_rel_0.001_compress_3_drop_18 mlp= 197744 exact=87.40%
271
+ baseline_rel_0.001_compress_3_drop_19 mlp= 197594 exact=77.07%
272
+ baseline_rel_0.001_compress_3_drop_20 mlp= 197518 exact=83.67%
273
+ baseline_rel_0.001_compress_3_drop_21 mlp= 197518 exact=54.53%
274
+ baseline_rel_0.001_compress_3_drop_22 mlp= 197514 exact=8.47%
275
+ baseline_rel_0.001_compress_3_drop_23 mlp= 197513 exact=7.27%
276
+ baseline_rel_0.001_compress_3_drop_25 mlp= 197514 exact=54.13%
277
+ baseline_rel_0.001_compress_3_drop_26 mlp= 197518 exact=82.00%
278
+ baseline_rel_0.001_compress_3_drop_27 mlp= 198119 exact=90.07%
279
+ baseline_rel_0.001_compress_3_accepted mlp= 197601 exact=91.67%
280
+ baseline_rel_0.001_compress_4_drop_00 mlp= 191575 exact=8.27%
281
+ baseline_rel_0.001_compress_4_drop_01 mlp= 195341 exact=59.33%
282
+ baseline_rel_0.001_compress_4_drop_02 mlp= 193455 exact=79.00%
283
+ baseline_rel_0.001_compress_4_drop_03 mlp= 193768 exact=75.73%
284
+ baseline_rel_0.001_compress_4_drop_04 mlp= 189677 exact=48.73%
285
+ baseline_rel_0.001_compress_4_drop_05 mlp= 190934 exact=89.93%
286
+ baseline_rel_0.001_compress_4_drop_07 mlp= 188767 exact=86.27%
287
+ baseline_rel_0.001_compress_4_drop_09 mlp= 188760 exact=90.87%
288
+ baseline_rel_0.001_compress_4_drop_10 mlp= 188777 exact=91.93%
289
+ baseline_rel_0.001_compress_4_drop_11 mlp= 188882 exact=85.80%
290
+ baseline_rel_0.001_compress_4_drop_12 mlp= 188935 exact=86.87%
291
+ baseline_rel_0.001_compress_4_drop_13 mlp= 189017 exact=91.40%
292
+ baseline_rel_0.001_compress_4_drop_14 mlp= 188984 exact=90.60%
293
+ baseline_rel_0.001_compress_4_drop_15 mlp= 189127 exact=91.33%
294
+ baseline_rel_0.001_compress_4_drop_16 mlp= 188898 exact=91.00%
295
+ baseline_rel_0.001_compress_4_drop_17 mlp= 188823 exact=90.93%
296
+ baseline_rel_0.001_compress_4_drop_18 mlp= 188872 exact=84.13%
297
+ baseline_rel_0.001_compress_4_drop_19 mlp= 188722 exact=72.07%
298
+ baseline_rel_0.001_compress_4_drop_20 mlp= 188646 exact=82.67%
299
+ baseline_rel_0.001_compress_4_drop_21 mlp= 188646 exact=51.07%
300
+ baseline_rel_0.001_compress_4_drop_22 mlp= 188642 exact=8.00%
301
+ baseline_rel_0.001_compress_4_drop_23 mlp= 188641 exact=7.00%
302
+ baseline_rel_0.001_compress_4_drop_25 mlp= 188642 exact=53.47%
303
+ baseline_rel_0.001_compress_4_drop_26 mlp= 188646 exact=79.53%
304
+ baseline_rel_0.001_compress_4_drop_27 mlp= 189247 exact=88.53%
305
+ baseline_rel_0.001_compress_4_accepted mlp= 188760 exact=90.87%
306
+ baseline_rel_0.001_compress_5_drop_00 mlp= 182734 exact=8.07%
307
+ baseline_rel_0.001_compress_5_drop_01 mlp= 186500 exact=58.47%
308
+ baseline_rel_0.001_compress_5_drop_02 mlp= 184614 exact=77.73%
309
+ baseline_rel_0.001_compress_5_drop_03 mlp= 184927 exact=75.53%
310
+ baseline_rel_0.001_compress_5_drop_04 mlp= 180836 exact=46.00%
311
+ baseline_rel_0.001_compress_5_drop_05 mlp= 182093 exact=88.33%
312
+ baseline_rel_0.001_compress_5_drop_07 mlp= 179926 exact=85.60%
313
+ baseline_rel_0.001_compress_5_drop_10 mlp= 179936 exact=90.27%
314
+ baseline_rel_0.001_compress_5_drop_11 mlp= 180041 exact=84.40%
315
+ baseline_rel_0.001_compress_5_drop_12 mlp= 180094 exact=84.93%
316
+ baseline_rel_0.001_compress_5_drop_13 mlp= 180176 exact=90.00%
317
+ baseline_rel_0.001_compress_5_drop_14 mlp= 180143 exact=89.60%
318
+ baseline_rel_0.001_compress_5_drop_15 mlp= 180286 exact=90.80%
319
+ baseline_rel_0.001_compress_5_drop_16 mlp= 180057 exact=90.67%
320
+ baseline_rel_0.001_compress_5_drop_17 mlp= 179982 exact=90.20%
321
+ baseline_rel_0.001_compress_5_drop_18 mlp= 180031 exact=83.80%
322
+ baseline_rel_0.001_compress_5_drop_19 mlp= 179881 exact=71.20%
323
+ baseline_rel_0.001_compress_5_drop_20 mlp= 179805 exact=81.53%
324
+ baseline_rel_0.001_compress_5_drop_21 mlp= 179805 exact=47.67%
325
+ baseline_rel_0.001_compress_5_drop_22 mlp= 179801 exact=8.40%
326
+ baseline_rel_0.001_compress_5_drop_23 mlp= 179800 exact=6.73%
327
+ baseline_rel_0.001_compress_5_drop_25 mlp= 179801 exact=53.93%
328
+ baseline_rel_0.001_compress_5_drop_26 mlp= 179805 exact=79.67%
329
+ baseline_rel_0.001_compress_5_drop_27 mlp= 180406 exact=87.93%
330
+ baseline_rel_0.001_compress_5_accepted mlp= 179936 exact=90.27%
331
+ baseline_rel_0.001_compress_6_drop_00 mlp= 173910 exact=7.13%
332
+ baseline_rel_0.001_compress_6_drop_01 mlp= 177676 exact=59.13%
333
+ baseline_rel_0.001_compress_6_drop_02 mlp= 175790 exact=76.60%
334
+ baseline_rel_0.001_compress_6_drop_03 mlp= 176103 exact=74.13%
335
+ baseline_rel_0.001_compress_6_drop_04 mlp= 172012 exact=43.60%
336
+ baseline_rel_0.001_compress_6_drop_05 mlp= 173269 exact=87.67%
337
+ baseline_rel_0.001_compress_6_drop_07 mlp= 171102 exact=84.13%
338
+ baseline_rel_0.001_compress_6_drop_11 mlp= 171217 exact=84.20%
339
+ baseline_rel_0.001_compress_6_drop_12 mlp= 171270 exact=84.73%
340
+ baseline_rel_0.001_compress_6_drop_13 mlp= 171352 exact=90.60%
341
+ baseline_rel_0.001_compress_6_drop_14 mlp= 171319 exact=89.20%
342
+ baseline_rel_0.001_compress_6_drop_15 mlp= 171462 exact=89.87%
343
+ baseline_rel_0.001_compress_6_drop_16 mlp= 171233 exact=90.00%
344
+ baseline_rel_0.001_compress_6_drop_17 mlp= 171158 exact=89.60%
345
+ baseline_rel_0.001_compress_6_drop_18 mlp= 171207 exact=82.33%
346
+ baseline_rel_0.001_compress_6_drop_19 mlp= 171057 exact=69.07%
347
+ baseline_rel_0.001_compress_6_drop_20 mlp= 170981 exact=79.07%
348
+ baseline_rel_0.001_compress_6_drop_21 mlp= 170981 exact=43.87%
349
+ baseline_rel_0.001_compress_6_drop_22 mlp= 170977 exact=8.40%
350
+ baseline_rel_0.001_compress_6_drop_23 mlp= 170976 exact=6.73%
351
+ baseline_rel_0.001_compress_6_drop_25 mlp= 170977 exact=53.87%
352
+ baseline_rel_0.001_compress_6_drop_26 mlp= 170981 exact=79.60%
353
+ baseline_rel_0.001_compress_6_drop_27 mlp= 171582 exact=87.67%
354
+ baseline_rel_0.001_compress_6_accepted mlp= 171233 exact=90.00%
355
+ baseline_rel_0.001_compress_7_drop_00 mlp= 165207 exact=6.93%
356
+ baseline_rel_0.001_compress_7_drop_01 mlp= 168973 exact=59.13%
357
+ baseline_rel_0.001_compress_7_drop_02 mlp= 167087 exact=74.13%
358
+ baseline_rel_0.001_compress_7_drop_03 mlp= 167400 exact=72.80%
359
+ baseline_rel_0.001_compress_7_drop_04 mlp= 163309 exact=40.40%
360
+ baseline_rel_0.001_compress_7_drop_05 mlp= 164566 exact=86.73%
361
+ baseline_rel_0.001_compress_7_drop_07 mlp= 162399 exact=82.13%
362
+ baseline_rel_0.001_compress_7_drop_11 mlp= 162514 exact=82.53%
363
+ baseline_rel_0.001_compress_7_drop_12 mlp= 162567 exact=81.80%
364
+ baseline_rel_0.001_compress_7_drop_13 mlp= 162649 exact=89.47%
365
+ baseline_rel_0.001_compress_7_drop_14 mlp= 162616 exact=88.67%
366
+ baseline_rel_0.001_compress_7_drop_15 mlp= 162759 exact=89.33%
367
+ baseline_rel_0.001_compress_7_drop_17 mlp= 162455 exact=88.20%
368
+ baseline_rel_0.001_compress_7_drop_18 mlp= 162504 exact=78.60%
369
+ baseline_rel_0.001_compress_7_drop_19 mlp= 162354 exact=62.60%
370
+ baseline_rel_0.001_compress_7_drop_20 mlp= 162278 exact=76.87%
371
+ baseline_rel_0.001_compress_7_drop_21 mlp= 162278 exact=39.87%
372
+ baseline_rel_0.001_compress_7_drop_22 mlp= 162274 exact=7.80%
373
+ baseline_rel_0.001_compress_7_drop_23 mlp= 162273 exact=6.40%
374
+ baseline_rel_0.001_compress_7_drop_25 mlp= 162274 exact=54.07%
375
+ baseline_rel_0.001_compress_7_drop_26 mlp= 162278 exact=78.00%
376
+ baseline_rel_0.001_compress_7_drop_27 mlp= 162879 exact=87.40%
377
+ [done] results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_2/target90_search.json
circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_2_train.log ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ Warning: You are sending unauthenticated requests to the HF Hub. Please set a HF_TOKEN to enable higher rate limits and faster downloads.
2
+ [model] Qwen/Qwen2.5-Math-1.5B device=cuda dtype=bfloat16
3
+
4
+ trainable params: 2,308,096 || all params: 1,546,022,400 || trainable%: 0.1493
5
+ {"step": 1, "loss": 1.9453125, "ce": 1.9453125, "kl": -2.872943878173828e-05, "lr": 1.3333333333333333e-05, "elapsed_s": 0.7714500427246094, "compact_exact": {"n": 8100, "correct": 7590, "accuracy": 0.937037037037037, "wrong_sample": [{"prompt": "11 + 13 =", "answer": "24", "pred": "35"}, {"prompt": "11 + 21 =", "answer": "32", "pred": "132"}, {"prompt": "11 + 31 =", "answer": "42", "pred": "142"}, {"prompt": "11 + 41 =", "answer": "52", "pred": "152"}, {"prompt": "11 + 47 =", "answer": "58", "pred": "158"}, {"prompt": "11 + 48 =", "answer": "59", "pred": "159"}, {"prompt": "11 + 49 =", "answer": "60", "pred": "160"}, {"prompt": "11 + 50 =", "answer": "61", "pred": "161"}, {"prompt": "11 + 51 =", "answer": "62", "pred": "162"}, {"prompt": "11 + 53 =", "answer": "64", "pred": "164"}, {"prompt": "11 + 57 =", "answer": "68", "pred": "168"}, {"prompt": "11 + 58 =", "answer": "69", "pred": "169"}, {"prompt": "11 + 59 =", "answer": "70", "pred": "170"}, {"prompt": "11 + 61 =", "answer": "72", "pred": "172"}, {"prompt": "11 + 62 =", "answer": "73", "pred": "173"}, {"prompt": "11 + 67 =", "answer": "78", "pred": "178"}, {"prompt": "11 + 68 =", "answer": "79", "pred": "179"}, {"prompt": "11 + 69 =", "answer": "80", "pred": "180"}, {"prompt": "11 + 70 =", "answer": "81", "pred": "181"}, {"prompt": "11 + 71 =", "answer": "82", "pred": "182"}, {"prompt": "11 + 75 =", "answer": "86", "pred": "186"}, {"prompt": "11 + 76 =", "answer": "87", "pred": "187"}, {"prompt": "11 + 78 =", "answer": "89", "pred": "189"}, {"prompt": "11 + 79 =", "answer": "90", "pred": "190"}, {"prompt": "11 + 80 =", "answer": "91", "pred": "191"}]}}
6
+ {"step": 100, "loss": 0.2996332958491162, "ce": 0.28403449781013257, "kl": 0.31230572016552244, "lr": 0.00015922352526649803, "elapsed_s": 69.5799949169159, "compact_exact": {"n": 8100, "correct": 8083, "accuracy": 0.9979012345679013, "wrong_sample": [{"prompt": "20 + 24 =", "answer": "44", "pred": "40"}, {"prompt": "21 + 24 =", "answer": "45", "pred": "42"}, {"prompt": "22 + 48 =", "answer": "70", "pred": "60"}, {"prompt": "23 + 37 =", "answer": "60", "pred": "50"}, {"prompt": "33 + 14 =", "answer": "47", "pred": "177"}, {"prompt": "63 + 17 =", "answer": "80", "pred": "70"}, {"prompt": "77 + 10 =", "answer": "87", "pred": "187"}, {"prompt": "77 + 12 =", "answer": "89", "pred": "189"}, {"prompt": "78 + 11 =", "answer": "89", "pred": "90"}, {"prompt": "79 + 10 =", "answer": "89", "pred": "189"}, {"prompt": "87 + 12 =", "answer": "99", "pred": "100"}, {"prompt": "87 + 22 =", "answer": "109", "pred": "110"}, {"prompt": "88 + 10 =", "answer": "98", "pred": "188"}, {"prompt": "88 + 12 =", "answer": "100", "pred": "200"}, {"prompt": "89 + 10 =", "answer": "99", "pred": "100"}, {"prompt": "89 + 20 =", "answer": "109", "pred": "110"}, {"prompt": "98 + 11 =", "answer": "109", "pred": "110"}]}}
7
+ {"step": 200, "loss": 0.0214703369140625, "ce": 0.004865188598632813, "kl": 0.33212890625, "lr": 5.484666416891109e-05, "elapsed_s": 125.57691788673401, "compact_exact": {"n": 8100, "correct": 8094, "accuracy": 0.9992592592592593, "wrong_sample": [{"prompt": "22 + 28 =", "answer": "50", "pred": "40"}, {"prompt": "22 + 48 =", "answer": "70", "pred": "69"}, {"prompt": "28 + 42 =", "answer": "70", "pred": "60"}, {"prompt": "62 + 18 =", "answer": "80", "pred": "70"}, {"prompt": "63 + 17 =", "answer": "80", "pred": "70"}, {"prompt": "89 + 10 =", "answer": "99", "pred": "100"}]}}
8
+ {"step": 300, "loss": 0.02060302734375, "ce": 0.004363555908203125, "kl": 0.32482421875, "lr": 0.0, "elapsed_s": 181.03875494003296, "compact_exact": {"n": 8100, "correct": 8099, "accuracy": 0.9998765432098765, "wrong_sample": [{"prompt": "63 + 17 =", "answer": "80", "pred": "70"}]}}
9
+ [done] adapter=results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_2/lora_r2_beta005/adapter accuracy=99.9877%
circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_4_correct_dataset.log ADDED
@@ -0,0 +1,26 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ Device: cuda dtype: bfloat16
2
+ Loading results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_4/lora_r4_beta005_merged ...
3
+
4
+ Pool: 8,100 unique (a, b) pairs in [10, 99] shard=1/1
5
+ 512/8100 correct= 512 acc=100.0% 585.3 ex/s eta 0.2 min
6
+ 1024/8100 correct= 1024 acc=100.0% 776.5 ex/s eta 0.2 min
7
+ 1536/8100 correct= 1534 acc=99.9% 871.0 ex/s eta 0.1 min
8
+ 2048/8100 correct= 2046 acc=99.9% 927.8 ex/s eta 0.1 min
9
+ 2560/8100 correct= 2558 acc=99.9% 965.9 ex/s eta 0.1 min
10
+ 3072/8100 correct= 3069 acc=99.9% 992.2 ex/s eta 0.1 min
11
+ 3584/8100 correct= 3581 acc=99.9% 1012.8 ex/s eta 0.1 min
12
+ 4096/8100 correct= 4093 acc=99.9% 1028.2 ex/s eta 0.1 min
13
+ 4608/8100 correct= 4605 acc=99.9% 1039.7 ex/s eta 0.1 min
14
+ 5120/8100 correct= 5117 acc=99.9% 1048.9 ex/s eta 0.0 min
15
+ 5632/8100 correct= 5629 acc=99.9% 1055.9 ex/s eta 0.0 min
16
+ 6144/8100 correct= 6141 acc=100.0% 1060.3 ex/s eta 0.0 min
17
+ 6656/8100 correct= 6653 acc=100.0% 1064.8 ex/s eta 0.0 min
18
+ 7168/8100 correct= 7165 acc=100.0% 1069.4 ex/s eta 0.0 min
19
+ 7680/8100 correct= 7677 acc=100.0% 1072.1 ex/s eta 0.0 min
20
+ 8100/8100 correct= 8097 acc=100.0% 1073.1 ex/s eta 0.0 min
21
+
22
+ =======================================================
23
+ Tested: 8,100
24
+ Correct: 8,097 (100.0%)
25
+ Correct saved: data/qwen25_math_1p5b_rank_ladder_r4_correct_addition.json
26
+ =======================================================
circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_4_direct_rank_shard0.log ADDED
@@ -0,0 +1,68 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ [records] results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_4/position_adacs positions=hundreds,tens,ones
2
+ records=1,500 skipped=0
3
+ [model] results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_4/lora_r4_beta005_merged device=cuda dtype=bfloat16 batch=256
4
+
5
+ topk_100 mlp= 2109 exact=9.53%
6
+ rel_0.05 mlp= 2971 exact=13.20%
7
+ topk_500 mlp= 13017 exact=50.20%
8
+ rel_0.01 mlp= 23872 exact=42.87%
9
+ topk_2000 mlp= 52135 exact=33.13%
10
+ rel_0.001 mlp= 223337 exact=100.00%
11
+ [rank] todo=28 shard=0/4
12
+ drop__topk_500__layer_00 mlp= 12919 exact=5.13%
13
+ delta=-45.07% per1k=-459.864%
14
+ drop__topk_500__layer_04 mlp= 12839 exact=16.53%
15
+ delta=-33.67% per1k=-189.139%
16
+ drop__topk_500__layer_08 mlp= 12900 exact=54.53%
17
+ delta=+4.33% per1k=+37.037%
18
+ drop__topk_500__layer_12 mlp= 12894 exact=50.93%
19
+ delta=+0.73% per1k=+5.962%
20
+ drop__topk_500__layer_16 mlp= 12906 exact=51.80%
21
+ delta=+1.60% per1k=+14.414%
22
+ drop__topk_500__layer_20 mlp= 12612 exact=37.60%
23
+ delta=-12.60% per1k=-31.111%
24
+ drop__topk_500__layer_24 mlp= 10286 exact=37.20%
25
+ delta=-13.00% per1k=-4.760%
26
+ add__rel_0.01_minus_topk_500__layer_00 mlp= 13090 exact=50.87%
27
+ delta=+0.67% per1k=+9.132%
28
+ add__rel_0.01_minus_topk_500__layer_04 mlp= 13149 exact=51.20%
29
+ delta=+1.00% per1k=+7.576%
30
+ add__rel_0.01_minus_topk_500__layer_08 mlp= 13163 exact=49.33%
31
+ delta=-0.87% per1k=-5.936%
32
+ add__rel_0.01_minus_topk_500__layer_12 mlp= 13195 exact=50.00%
33
+ delta=-0.20% per1k=-1.124%
34
+ add__rel_0.01_minus_topk_500__layer_16 mlp= 13226 exact=50.40%
35
+ delta=+0.20% per1k=+0.957%
36
+ add__rel_0.01_minus_topk_500__layer_20 mlp= 13529 exact=52.07%
37
+ delta=+1.87% per1k=+3.646%
38
+ add__rel_0.01_minus_topk_500__layer_24 mlp= 14656 exact=50.53%
39
+ delta=+0.33% per1k=+0.203%
40
+ add__topk_2000_minus_topk_500__layer_00 mlp= 13296 exact=52.07%
41
+ delta=+1.87% per1k=+6.691%
42
+ add__topk_2000_minus_topk_500__layer_04 mlp= 13446 exact=51.27%
43
+ delta=+1.07% per1k=+2.486%
44
+ add__topk_2000_minus_topk_500__layer_08 mlp= 13564 exact=48.80%
45
+ delta=-1.40% per1k=-2.559%
46
+ add__topk_2000_minus_topk_500__layer_12 mlp= 13568 exact=51.33%
47
+ delta=+1.13% per1k=+2.057%
48
+ add__topk_2000_minus_topk_500__layer_16 mlp= 13689 exact=50.73%
49
+ delta=+0.53% per1k=+0.794%
50
+ add__topk_2000_minus_topk_500__layer_20 mlp= 15392 exact=52.80%
51
+ delta=+2.60% per1k=+1.095%
52
+ add__topk_2000_minus_topk_500__layer_24 mlp= 17801 exact=49.87%
53
+ delta=-0.33% per1k=-0.070%
54
+ add__rel_0.001_minus_topk_500__layer_00 mlp= 18903 exact=53.20%
55
+ delta=+3.00% per1k=+0.510%
56
+ add__rel_0.001_minus_topk_500__layer_04 mlp= 20650 exact=61.87%
57
+ delta=+11.67% per1k=+1.528%
58
+ add__rel_0.001_minus_topk_500__layer_08 mlp= 21749 exact=60.20%
59
+ delta=+10.00% per1k=+1.145%
60
+ add__rel_0.001_minus_topk_500__layer_12 mlp= 21494 exact=58.47%
61
+ delta=+8.27% per1k=+0.975%
62
+ add__rel_0.001_minus_topk_500__layer_16 mlp= 21582 exact=55.47%
63
+ delta=+5.27% per1k=+0.615%
64
+ add__rel_0.001_minus_topk_500__layer_20 mlp= 21568 exact=55.33%
65
+ delta=+5.13% per1k=+0.600%
66
+ add__rel_0.001_minus_topk_500__layer_24 mlp= 19246 exact=50.13%
67
+ delta=-0.07% per1k=-0.011%
68
+ [done] rank-only -> results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_4/direct_group_rank_shard0.json
circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_4_direct_rank_shard1.log ADDED
@@ -0,0 +1,68 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ [records] results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_4/position_adacs positions=hundreds,tens,ones
2
+ records=1,500 skipped=0
3
+ [model] results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_4/lora_r4_beta005_merged device=cuda dtype=bfloat16 batch=256
4
+
5
+ topk_100 mlp= 2109 exact=9.53%
6
+ rel_0.05 mlp= 2971 exact=13.20%
7
+ topk_500 mlp= 13017 exact=50.20%
8
+ rel_0.01 mlp= 23872 exact=42.87%
9
+ topk_2000 mlp= 52135 exact=33.13%
10
+ rel_0.001 mlp= 223337 exact=100.00%
11
+ [rank] todo=28 shard=1/4
12
+ drop__topk_500__layer_01 mlp= 12935 exact=38.53%
13
+ delta=-11.67% per1k=-142.276%
14
+ drop__topk_500__layer_05 mlp= 12900 exact=56.87%
15
+ delta=+6.67% per1k=+56.980%
16
+ drop__topk_500__layer_09 mlp= 12905 exact=54.07%
17
+ delta=+3.87% per1k=+34.524%
18
+ drop__topk_500__layer_13 mlp= 12903 exact=51.73%
19
+ delta=+1.53% per1k=+13.450%
20
+ drop__topk_500__layer_17 mlp= 12899 exact=50.67%
21
+ delta=+0.47% per1k=+3.955%
22
+ drop__topk_500__layer_21 mlp= 12451 exact=13.40%
23
+ delta=-36.80% per1k=-65.018%
24
+ drop__topk_500__layer_25 mlp= 10027 exact=38.93%
25
+ delta=-11.27% per1k=-3.768%
26
+ add__rel_0.01_minus_topk_500__layer_01 mlp= 13064 exact=49.93%
27
+ delta=-0.27% per1k=-5.674%
28
+ add__rel_0.01_minus_topk_500__layer_05 mlp= 13097 exact=49.53%
29
+ delta=-0.67% per1k=-8.333%
30
+ add__rel_0.01_minus_topk_500__layer_09 mlp= 13172 exact=49.47%
31
+ delta=-0.73% per1k=-4.731%
32
+ add__rel_0.01_minus_topk_500__layer_13 mlp= 13200 exact=50.47%
33
+ delta=+0.27% per1k=+1.457%
34
+ add__rel_0.01_minus_topk_500__layer_17 mlp= 13204 exact=49.00%
35
+ delta=-1.20% per1k=-6.417%
36
+ add__rel_0.01_minus_topk_500__layer_21 mlp= 13665 exact=52.00%
37
+ delta=+1.80% per1k=+2.778%
38
+ add__rel_0.01_minus_topk_500__layer_25 mlp= 14778 exact=50.33%
39
+ delta=+0.13% per1k=+0.076%
40
+ add__topk_2000_minus_topk_500__layer_01 mlp= 13186 exact=50.33%
41
+ delta=+0.13% per1k=+0.789%
42
+ add__topk_2000_minus_topk_500__layer_05 mlp= 13334 exact=50.00%
43
+ delta=-0.20% per1k=-0.631%
44
+ add__topk_2000_minus_topk_500__layer_09 mlp= 13589 exact=49.53%
45
+ delta=-0.67% per1k=-1.166%
46
+ add__topk_2000_minus_topk_500__layer_13 mlp= 13600 exact=50.67%
47
+ delta=+0.47% per1k=+0.800%
48
+ add__topk_2000_minus_topk_500__layer_17 mlp= 13738 exact=49.47%
49
+ delta=-0.73% per1k=-1.017%
50
+ add__topk_2000_minus_topk_500__layer_21 mlp= 15957 exact=52.73%
51
+ delta=+2.53% per1k=+0.862%
52
+ add__topk_2000_minus_topk_500__layer_25 mlp= 17654 exact=51.00%
53
+ delta=+0.80% per1k=+0.173%
54
+ add__rel_0.001_minus_topk_500__layer_01 mlp= 15196 exact=52.00%
55
+ delta=+1.80% per1k=+0.826%
56
+ add__rel_0.001_minus_topk_500__layer_05 mlp= 19473 exact=54.20%
57
+ delta=+4.00% per1k=+0.620%
58
+ add__rel_0.001_minus_topk_500__layer_09 mlp= 21735 exact=57.73%
59
+ delta=+7.53% per1k=+0.864%
60
+ add__rel_0.001_minus_topk_500__layer_13 mlp= 21429 exact=60.07%
61
+ delta=+9.87% per1k=+1.173%
62
+ add__rel_0.001_minus_topk_500__layer_17 mlp= 21643 exact=56.67%
63
+ delta=+6.47% per1k=+0.750%
64
+ add__rel_0.001_minus_topk_500__layer_21 mlp= 21403 exact=54.27%
65
+ delta=+4.07% per1k=+0.485%
66
+ add__rel_0.001_minus_topk_500__layer_25 mlp= 18986 exact=51.47%
67
+ delta=+1.27% per1k=+0.212%
68
+ [done] rank-only -> results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_4/direct_group_rank_shard1.json
circuit-shotting/artifacts/pod_logs/cosmic-matrix-e4/full/qwen25_math_1p5b_2digit_rank_ladder_issue9/logs/rank_4_direct_rank_shard2.log ADDED
@@ -0,0 +1,68 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ [records] results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_4/position_adacs positions=hundreds,tens,ones
2
+ records=1,500 skipped=0
3
+ [model] results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_4/lora_r4_beta005_merged device=cuda dtype=bfloat16 batch=256
4
+
5
+ topk_100 mlp= 2109 exact=9.53%
6
+ rel_0.05 mlp= 2971 exact=13.20%
7
+ topk_500 mlp= 13017 exact=50.20%
8
+ rel_0.01 mlp= 23872 exact=42.87%
9
+ topk_2000 mlp= 52135 exact=33.13%
10
+ rel_0.001 mlp= 223337 exact=100.00%
11
+ [rank] todo=28 shard=2/4
12
+ drop__topk_500__layer_02 mlp= 12929 exact=36.53%
13
+ delta=-13.67% per1k=-155.303%
14
+ drop__topk_500__layer_06 mlp= 12883 exact=45.20%
15
+ delta=-5.00% per1k=-37.313%
16
+ drop__topk_500__layer_10 mlp= 12929 exact=53.80%
17
+ delta=+3.60% per1k=+40.909%
18
+ drop__topk_500__layer_14 mlp= 12884 exact=53.20%
19
+ delta=+3.00% per1k=+22.556%
20
+ drop__topk_500__layer_18 mlp= 12862 exact=40.13%
21
+ delta=-10.07% per1k=-64.946%
22
+ drop__topk_500__layer_22 mlp= 12058 exact=5.27%
23
+ delta=-44.93% per1k=-46.854%
24
+ drop__topk_500__layer_26 mlp= 12045 exact=48.20%
25
+ delta=-2.00% per1k=-2.058%
26
+ add__rel_0.01_minus_topk_500__layer_02 mlp= 13087 exact=51.87%
27
+ delta=+1.67% per1k=+23.810%
28
+ add__rel_0.01_minus_topk_500__layer_06 mlp= 13178 exact=49.87%
29
+ delta=-0.33% per1k=-2.070%
30
+ add__rel_0.01_minus_topk_500__layer_10 mlp= 13172 exact=49.20%
31
+ delta=-1.00% per1k=-6.452%
32
+ add__rel_0.01_minus_topk_500__layer_14 mlp= 13204 exact=47.20%
33
+ delta=-3.00% per1k=-16.043%
34
+ add__rel_0.01_minus_topk_500__layer_18 mlp= 13255 exact=49.33%
35
+ delta=-0.87% per1k=-3.641%
36
+ add__rel_0.01_minus_topk_500__layer_22 mlp= 13889 exact=51.00%
37
+ delta=+0.80% per1k=+0.917%
38
+ add__rel_0.01_minus_topk_500__layer_26 mlp= 13835 exact=50.47%
39
+ delta=+0.27% per1k=+0.326%
40
+ add__topk_2000_minus_topk_500__layer_02 mlp= 13282 exact=55.33%
41
+ delta=+5.13% per1k=+19.371%
42
+ add__topk_2000_minus_topk_500__layer_06 mlp= 13569 exact=49.87%
43
+ delta=-0.33% per1k=-0.604%
44
+ add__topk_2000_minus_topk_500__layer_10 mlp= 13566 exact=47.73%
45
+ delta=-2.47% per1k=-4.493%
46
+ add__topk_2000_minus_topk_500__layer_14 mlp= 13659 exact=46.80%
47
+ delta=-3.40% per1k=-5.296%
48
+ add__topk_2000_minus_topk_500__layer_18 mlp= 13939 exact=50.13%
49
+ delta=-0.07% per1k=-0.072%
50
+ add__topk_2000_minus_topk_500__layer_22 mlp= 17324 exact=51.53%
51
+ delta=+1.33% per1k=+0.310%
52
+ add__topk_2000_minus_topk_500__layer_26 mlp= 16174 exact=50.07%
53
+ delta=-0.13% per1k=-0.042%
54
+ add__rel_0.001_minus_topk_500__layer_02 mlp= 17009 exact=60.67%
55
+ delta=+10.47% per1k=+2.622%
56
+ add__rel_0.001_minus_topk_500__layer_06 mlp= 21754 exact=57.27%
57
+ delta=+7.07% per1k=+0.809%
58
+ add__rel_0.001_minus_topk_500__layer_10 mlp= 21710 exact=59.33%
59
+ delta=+9.13% per1k=+1.051%
60
+ add__rel_0.001_minus_topk_500__layer_14 mlp= 21462 exact=52.47%
61
+ delta=+2.27% per1k=+0.268%
62
+ add__rel_0.001_minus_topk_500__layer_18 mlp= 21536 exact=53.53%
63
+ delta=+3.33% per1k=+0.391%
64
+ add__rel_0.001_minus_topk_500__layer_22 mlp= 21018 exact=51.33%
65
+ delta=+1.13% per1k=+0.142%
66
+ add__rel_0.001_minus_topk_500__layer_26 mlp= 21002 exact=50.20%
67
+ delta=+0.00% per1k=+0.000%
68
+ [done] rank-only -> results/qwen25_math_1p5b_2digit_rank_ladder_issue9/rank_4/direct_group_rank_shard2.json