hbfreed commited on
Commit
a2966b3
·
verified ·
1 Parent(s): 2dcc4fb

Add files using upload-large-folder tool

Browse files
Files changed (50) hide show
  1. healed/correctness_ab.log +31 -0
  2. healed/grid_general_fairness.log +14 -0
  3. healed/healing_breadth.queue.log +47 -0
  4. healed/keep5_math_offpolicy_top128.console.log +324 -0
  5. healed/knee0924_queue.log +43 -0
  6. healed/opd_warm.log +59 -0
  7. healed/opd_warm_fixed.log +186 -0
  8. healed/warm_chain.sh +50 -0
  9. healed/warm_chain2.sh +81 -0
  10. healed/warmup_fixed.log +213 -0
  11. healed/warmup_gate.log +26 -0
  12. healed/warmup_keep50.log +213 -0
  13. pruned/knee0924_keep40_save.log +2 -0
  14. pruned/uniform_keep2575.materialize.log +9 -0
  15. pruned/uniform_keep50.materialize.log +7 -0
  16. quant_ab/bf16.json +13 -0
  17. quant_ab/int8.json +13 -0
  18. quant_ab/nf4.json +13 -0
  19. quant_ab/run_bf16.log +17 -0
  20. quant_ab/run_int8.log +17 -0
  21. quant_ab/run_nf4.log +21 -0
  22. quant_ab/run_w4a16.log +21 -0
  23. quant_ab/w4a16.json +13 -0
  24. qwen35_reap_keep25/chat_template.jinja +154 -0
  25. qwen35_reap_keep25/config.json +2752 -0
  26. qwen35_reap_keep25/configuration_pruned_qwen3_5_moe.py +31 -0
  27. qwen35_reap_keep25/generation_config.json +9 -0
  28. qwen35_reap_keep25/modeling_pruned_qwen3_5_moe.py +122 -0
  29. qwen35_reap_keep25/reap_verify.json +9 -0
  30. qwen35_reap_keep25/tokenizer_config.json +32 -0
  31. qwen35_reap_keep50/chat_template.jinja +154 -0
  32. qwen35_reap_keep50/config.json +5312 -0
  33. qwen35_reap_keep50/configuration_pruned_qwen3_5_moe.py +31 -0
  34. qwen35_reap_keep50/generation_config.json +9 -0
  35. qwen35_reap_keep50/modeling_pruned_qwen3_5_moe.py +122 -0
  36. qwen35_reap_keep50/reap_verify.json +9 -0
  37. qwen35_reap_keep50/tokenizer_config.json +32 -0
  38. redo_policy/score_divergence.json +38 -0
  39. teacher_trajectories/dolci_math_curated.stats.json +29 -0
  40. teacher_trajectories/dolci_math_curated_opd.stats.json +30 -0
  41. teacher_trajectories/dolci_math_nogold_matched.stats.json +10 -0
  42. teacher_trajectories/dolci_math_nogold_matched_top128.score.log +647 -0
  43. teacher_trajectories/gen_state.chunks27-32.json +1 -0
  44. teacher_trajectories/generalgen.log +3 -0
  45. teacher_trajectories/generalgen_A.log +4 -0
  46. teacher_trajectories/generalgen_C.log +4 -0
  47. teacher_trajectories/server_general.log +0 -0
  48. teacher_trajectories/server_general_A.log +41 -0
  49. teacher_trajectories/server_general_C.log +331 -0
  50. teacher_trajectories/top128_score.log +783 -0
healed/correctness_ab.log ADDED
@@ -0,0 +1,31 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 2026-07-17T04:01:37-07:00 === correctness A/B start ===
2
+ 2026-07-17T04:01:37-07:00 scoring top-128 -> outputs/teacher_trajectories/dolci_math_nogold_matched_top128 (GPU 0)
3
+ 2026-07-17T04:21:00-07:00 scoring complete: "total_records": 9918,
4
+ 2026-07-17T04:21:00-07:00 healing glean_keep25_nogold_s1224 on GPU 1
5
+ 2026-07-17T04:21:00-07:00 healing glean_keep25_nogold_s1225 on GPU 2
6
+ 2026-07-17T04:38:30-07:00 === correctness A/B start ===
7
+ 2026-07-17T04:38:30-07:00 scoring complete: "total_records": 9918,
8
+ 2026-07-17T04:38:30-07:00 healing glean_keep25_nogold_s1226 on GPU 2
9
+ 2026-07-17T04:38:30-07:00 healing glean_keep25_nogold_s1224 on GPU 0
10
+ 2026-07-17T04:38:30-07:00 healing glean_keep25_nogold_s1225 on GPU 1
11
+ 2026-07-17T04:41:56-07:00 === correctness A/B start ===
12
+ 2026-07-17T04:41:56-07:00 scoring complete: "total_records": 9918,
13
+ 2026-07-17T04:41:56-07:00 healing glean_keep25_nogold_s1224 on GPU 0
14
+ 2026-07-17T04:41:56-07:00 healing glean_keep25_nogold_s1225 on GPU 1
15
+ 2026-07-17T05:33:14-07:00 eval glean_keep25_nogold_s1224
16
+ 2026-07-17T05:35:05-07:00 eval glean_keep25_nogold_s1225
17
+ 2026-07-17T05:35:37-07:00 glean_keep25_nogold_s1224 done -> 0.41091736163760423
18
+ 2026-07-17T05:35:37-07:00 healing glean_keep25_nogold_s1226 on GPU 0
19
+ 2026-07-17T05:37:25-07:00 glean_keep25_nogold_s1225 done -> 0.4116755117513268
20
+ 2026-07-17T05:37:25-07:00 lane GPU1 done
21
+ 2026-07-17T06:28:01-07:00 eval glean_keep25_nogold_s1226
22
+ 2026-07-17T06:30:25-07:00 glean_keep25_nogold_s1226 done -> 0.42532221379833207
23
+ 2026-07-17T06:30:25-07:00 lane GPU0 done
24
+ 2026-07-17T06:30:25-07:00 === correctness A/B COMPLETE ===
25
+ seed | curated(correct) | nogold(unfiltered) | delta
26
+ 1224 | 0.4131918119787718 | 0.41091736163760423 | -0.2
27
+ 1225 | 0.4245640636846095 | 0.4116755117513268 | -1.3
28
+ 1226 | 0.422289613343442 | 0.42532221379833207 | +0.3
29
+
30
+ MEAN curated=0.4200 nogold=0.4160 delta=-0.40 pts
31
+ Verdict: |delta| small -> correctness barely matters (general column can skip gold-curation).
healed/grid_general_fairness.log ADDED
@@ -0,0 +1,14 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 2026-07-19T14:17:31-07:00 waiting for the general grid to finish...
2
+ 2026-07-19T14:17:31-07:00 GPUs free, starting fairness arms
3
+ 2026-07-19T14:17:31-07:00 HEAL reap_keep50_s1224_lr1e5 on GPU 0 (lr=1e-5 steps=150)
4
+ 2026-07-19T14:20:01-07:00 HEAL reap_keep50_s1224_long500 on GPU 1 (lr=3e-5 steps=500)
5
+ 2026-07-19T14:22:31-07:00 HEAL glean_keep50_s1224_long500 on GPU 2 (lr=3e-5 steps=500)
6
+ 2026-07-19T17:17:10-07:00 EVAL reap_keep50_s1224_lr1e5_step150 on GPU 0 (port 8420)
7
+ 2026-07-19T17:31:13-07:00 arm reap_keep50_s1224_lr1e5 done
8
+ 2026-07-19T21:26:45-07:00 EVAL glean_keep50_s1224_long500_step150 on GPU 2 (port 8422)
9
+ 2026-07-19T21:40:19-07:00 EVAL glean_keep50_s1224_long500_step500 on GPU 2 (port 8422)
10
+ 2026-07-19T21:52:23-07:00 arm glean_keep50_s1224_long500 done
11
+ 2026-07-19T22:48:52-07:00 EVAL reap_keep50_s1224_long500_step150 on GPU 1 (port 8421)
12
+ 2026-07-19T22:58:20-07:00 EVAL reap_keep50_s1224_long500_step500 on GPU 1 (port 8421)
13
+ 2026-07-19T23:07:08-07:00 arm reap_keep50_s1224_long500 done
14
+ 2026-07-19T23:07:08-07:00 ###### FAIRNESS ARMS COMPLETE ######
healed/healing_breadth.queue.log ADDED
@@ -0,0 +1,47 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 2026-07-14T20:27:39-07:00 healing breadth queue started
2
+ 2026-07-14T20:27:39-07:00 starting glean_math_keep25 healing from outputs/pruned/glean-0125inst-math-keep25
3
+ 2026-07-14T20:39:23-07:00 healing breadth queue started
4
+ 2026-07-14T20:39:23-07:00 starting glean_math_keep25 healing from outputs/pruned/glean-0125inst-math-keep25
5
+ 2026-07-14T20:44:27-07:00 healing breadth queue started
6
+ 2026-07-14T20:44:27-07:00 starting glean_math_keep25 healing from outputs/pruned/glean-0125inst-math-keep25
7
+ 2026-07-14T20:45:07-07:00 queue stopped on an error at line 78
8
+ 2026-07-14T20:53:00-07:00 healing breadth queue started
9
+ 2026-07-14T20:53:00-07:00 starting glean_math_keep25 healing from outputs/pruned/glean-0125inst-math-keep25
10
+ 2026-07-14T21:55:23-07:00 glean_math_keep25 healing complete
11
+ 2026-07-14T21:55:24-07:00 evaluating glean_math_keep25 on full chat GSM8K
12
+ {
13
+ "correct": 340,
14
+ "accuracy": 0.2577710386656558,
15
+ "finished": 766,
16
+ "finish_rate": 0.5807429871114481,
17
+ "mean_completion_tokens": 367.81652767247914
18
+ }
19
+ saved item-level results -> outputs/evals/healing_breadth/glean_math_keep25_seed1224.json
20
+ 2026-07-14T21:59:08-07:00 glean_math_keep25 result: 0.2577710386656558
21
+ 2026-07-14T21:59:08-07:00 starting reap_math_keep75 healing from outputs/pruned/reap48-0125inst-math
22
+ 2026-07-14T23:27:48-07:00 reap_math_keep75 healing complete
23
+ 2026-07-14T23:27:48-07:00 evaluating reap_math_keep75 on full chat GSM8K
24
+ {
25
+ "correct": 902,
26
+ "accuracy": 0.6838514025777104,
27
+ "finished": 1313,
28
+ "finish_rate": 0.9954510993176648,
29
+ "mean_completion_tokens": 107.30553449583017
30
+ }
31
+ saved item-level results -> outputs/evals/healing_breadth/reap_math_keep75_seed1224.json
32
+ 2026-07-14T23:29:27-07:00 reap_math_keep75 result: 0.6838514025777104
33
+ 2026-07-14T23:29:27-07:00 materializing math-calibrated Instruct uniform keep-50 checkpoint
34
+ 2026-07-14T23:29:57-07:00 uniform keep-50 checkpoint ready
35
+ 2026-07-14T23:29:57-07:00 starting uniform_math_keep50 healing from outputs/pruned/uniform_keep50
36
+ 2026-07-15T00:43:17-07:00 uniform_math_keep50 healing complete
37
+ 2026-07-15T00:43:18-07:00 evaluating uniform_math_keep50 on full chat GSM8K
38
+ {
39
+ "correct": 673,
40
+ "accuracy": 0.510235026535254,
41
+ "finished": 1302,
42
+ "finish_rate": 0.9871114480667172,
43
+ "mean_completion_tokens": 115.02501895375285
44
+ }
45
+ saved item-level results -> outputs/evals/healing_breadth/uniform_math_keep50_seed1224.json
46
+ 2026-07-15T00:45:11-07:00 uniform_math_keep50 result: 0.510235026535254
47
+ 2026-07-15T00:45:11-07:00 healing breadth queue complete
healed/keep5_math_offpolicy_top128.console.log ADDED
@@ -0,0 +1,324 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ /home/henry/Documents/PythonProjects/variable-reap/.venv/lib/python3.12/site-packages/megablocks/grouped_gemm_util.py:10: UserWarning: Grouped GEMM not available.
2
+ warnings.warn('Grouped GEMM not available.')
3
+ wandb: Tracking run with wandb version 0.28.0
4
+ wandb: W&B syncing is set to `offline` in this directory. Run `wandb online` or set WANDB_MODE=online to enable cloud syncing.
5
+ wandb: Run data is saved locally in outputs/healed/keep5_math_offpolicy_top128/wandb/offline-run-20260713_193950-ixqqkrvl
6
+ wandb: View this run in the terminal with `wandb leet`
7
+
8
+ 12115 cached top-128 chat trajectories / 6,476,634 unique tokens | 52 steps/epoch | 52 total steps | student params 3.70B | teacher overlap=False
9
+ {"step": 1, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.26238401538550016, "tokens": 122002, "cumulative_loss_tokens": 122002, "grad_norm": 4.71875, "lr": 6e-06, "finish_rate": 0.873, "comp_len": 514.8, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 47.1, "frames": {"chat": 237}, "mem_gb": 15.88}
10
+ The attention mask is not set and cannot be inferred from input because pad token is same as eos token. As a consequence, you may observe unexpected behavior. Please pass your input's `attention_mask` to obtain reliable results.
11
+ [eval step 1] sample: 'To find the greatest common divisor (GCD) of 51 and 12767, we can use the Euclidean algorithm. The steps are as follows:\n\n1. **Prime Factorization**:\n - **51**: The prime factorization is \\(3 \\times'
12
+ {"step": 1, "gsm8k_n": 64, "gsm8k_quick_raw": 0.53125, "gsm8k_quick": 0.53125, "gsm8k_quick_chat": 0.5625, "t_eval_s": 94.5}
13
+ {"step": 2, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.26422106620881236, "tokens": 133490, "cumulative_loss_tokens": 255492, "grad_norm": 4.40625, "lr": 9e-06, "finish_rate": 0.823, "comp_len": 563.2, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 55.8, "frames": {"chat": 237}, "mem_gb": 18.67}
14
+ {"step": 3, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.24198635682198535, "tokens": 138158, "cumulative_loss_tokens": 393650, "grad_norm": 3.5, "lr": 1.2e-05, "finish_rate": 0.726, "comp_len": 582.9, "t_data_s": 0.1, "t_rollout_s": 0.0, "t_step_s": 45.7, "frames": {"chat": 237}, "mem_gb": 16.1}
15
+ {"step": 4, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.17819873864831604, "tokens": 115487, "cumulative_loss_tokens": 509137, "grad_norm": 2.28125, "lr": 1.5e-05, "finish_rate": 0.886, "comp_len": 487.3, "t_data_s": 0.1, "t_rollout_s": 0.0, "t_step_s": 39.5, "frames": {"chat": 237}, "mem_gb": 15.97}
16
+ {"step": 5, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.14776355108085704, "tokens": 130289, "cumulative_loss_tokens": 639426, "grad_norm": 1.6484375, "lr": 1.8e-05, "finish_rate": 0.785, "comp_len": 549.7, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 43.1, "frames": {"chat": 237}, "mem_gb": 16.05}
17
+ {"step": 6, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.13656450381276233, "tokens": 114067, "cumulative_loss_tokens": 753493, "grad_norm": 1.34375, "lr": 2.1e-05, "finish_rate": 0.903, "comp_len": 481.3, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 40.2, "frames": {"chat": 237}, "mem_gb": 15.95}
18
+ {"step": 7, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.12682973371786743, "tokens": 143092, "cumulative_loss_tokens": 896585, "grad_norm": 0.875, "lr": 2.4e-05, "finish_rate": 0.755, "comp_len": 603.8, "t_data_s": 0.1, "t_rollout_s": 0.0, "t_step_s": 44.5, "frames": {"chat": 237}, "mem_gb": 16.06}
19
+ {"step": 8, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.11038571776528162, "tokens": 143293, "cumulative_loss_tokens": 1039878, "grad_norm": 0.6640625, "lr": 2.7000000000000002e-05, "finish_rate": 0.764, "comp_len": 604.6, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 44.4, "frames": {"chat": 237}, "mem_gb": 16.03}
20
+ {"step": 9, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.09619075945408695, "tokens": 130128, "cumulative_loss_tokens": 1170006, "grad_norm": 0.67578125, "lr": 3e-05, "finish_rate": 0.823, "comp_len": 549.1, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 42.6, "frames": {"chat": 237}, "mem_gb": 16.05}
21
+ {"step": 10, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.09096234446447897, "tokens": 109943, "cumulative_loss_tokens": 1279949, "grad_norm": 0.69140625, "lr": 3e-05, "finish_rate": 0.924, "comp_len": 463.9, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 38.5, "frames": {"chat": 237}, "mem_gb": 16.01}
22
+ [eval step 10] sample: "To find the greatest common divisor (GCD) of 51 and 12767, we can use Python's `sympy` library, which provides a built-in function `gcd` for computing the GCD of two numbers.\n\nHere's the step-by-step "
23
+ {"step": 10, "gsm8k_n": 64, "gsm8k_quick_raw": 0.5, "gsm8k_quick": 0.5, "gsm8k_quick_chat": 0.609375, "t_eval_s": 88.0}
24
+ {"step": 11, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.10498863942140974, "tokens": 121186, "cumulative_loss_tokens": 1401135, "grad_norm": 0.60546875, "lr": 3e-05, "finish_rate": 0.861, "comp_len": 511.3, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 41.0, "frames": {"chat": 237}, "mem_gb": 18.69}
25
+ {"step": 12, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.09756646872474235, "tokens": 121234, "cumulative_loss_tokens": 1522369, "grad_norm": 0.609375, "lr": 3e-05, "finish_rate": 0.878, "comp_len": 511.5, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 40.9, "frames": {"chat": 237}, "mem_gb": 15.88}
26
+ {"step": 13, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.10299045846225482, "tokens": 135126, "cumulative_loss_tokens": 1657495, "grad_norm": 0.58984375, "lr": 3e-05, "finish_rate": 0.81, "comp_len": 570.2, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 44.0, "frames": {"chat": 237}, "mem_gb": 16.04}
27
+ {"step": 14, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.12756414832580626, "tokens": 119078, "cumulative_loss_tokens": 1776573, "grad_norm": 0.890625, "lr": 3e-05, "finish_rate": 0.857, "comp_len": 502.4, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 41.4, "frames": {"chat": 237}, "mem_gb": 15.9}
28
+ {"step": 15, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.08950029558452756, "tokens": 116533, "cumulative_loss_tokens": 1893106, "grad_norm": 0.46875, "lr": 3e-05, "finish_rate": 0.882, "comp_len": 491.7, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 40.1, "frames": {"chat": 237}, "mem_gb": 15.98}
29
+ {"step": 16, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.08766337552320194, "tokens": 142126, "cumulative_loss_tokens": 2035232, "grad_norm": 0.451171875, "lr": 3e-05, "finish_rate": 0.781, "comp_len": 599.7, "t_data_s": 0.1, "t_rollout_s": 0.0, "t_step_s": 44.0, "frames": {"chat": 237}, "mem_gb": 15.99}
30
+ {"step": 17, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.09789560582727819, "tokens": 139042, "cumulative_loss_tokens": 2174274, "grad_norm": 0.466796875, "lr": 3e-05, "finish_rate": 0.709, "comp_len": 586.7, "t_data_s": 0.1, "t_rollout_s": 0.0, "t_step_s": 47.0, "frames": {"chat": 237}, "mem_gb": 16.14}
31
+ {"step": 18, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.08371686447640507, "tokens": 132346, "cumulative_loss_tokens": 2306620, "grad_norm": 0.453125, "lr": 3e-05, "finish_rate": 0.81, "comp_len": 558.4, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 43.7, "frames": {"chat": 237}, "mem_gb": 16.04}
32
+ {"step": 19, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.07511474406852305, "tokens": 125011, "cumulative_loss_tokens": 2431631, "grad_norm": 0.380859375, "lr": 3e-05, "finish_rate": 0.844, "comp_len": 527.5, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 41.9, "frames": {"chat": 237}, "mem_gb": 16.04}
33
+ {"step": 20, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.07739786637930317, "tokens": 119952, "cumulative_loss_tokens": 2551583, "grad_norm": 0.396484375, "lr": 3e-05, "finish_rate": 0.84, "comp_len": 506.1, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 40.8, "frames": {"chat": 237}, "mem_gb": 16.03}
34
+ [eval step 20] sample: 'To solve the problem of finding the greatest common divisor (GCD) of 51 and 12767 using Python and SymPy, we can follow these steps:\n\n1. **Import the necessary functions from SymPy.**\n2. **Use the `gc'
35
+ {"step": 20, "gsm8k_n": 64, "gsm8k_quick_raw": 0.578125, "gsm8k_quick": 0.578125, "gsm8k_quick_chat": 0.5625, "t_eval_s": 92.7}
36
+ {"step": 21, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.07174015066452584, "tokens": 132305, "cumulative_loss_tokens": 2683888, "grad_norm": 0.388671875, "lr": 3e-05, "finish_rate": 0.852, "comp_len": 558.2, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 42.9, "frames": {"chat": 237}, "mem_gb": 18.69}
37
+ {"step": 22, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.09040485032644428, "tokens": 127244, "cumulative_loss_tokens": 2811132, "grad_norm": 0.46484375, "lr": 3e-05, "finish_rate": 0.802, "comp_len": 536.9, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 44.6, "frames": {"chat": 237}, "mem_gb": 16.07}
38
+ {"step": 23, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.07938295329848784, "tokens": 122990, "cumulative_loss_tokens": 2934122, "grad_norm": 0.4375, "lr": 3e-05, "finish_rate": 0.857, "comp_len": 518.9, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 41.2, "frames": {"chat": 237}, "mem_gb": 15.96}
39
+ {"step": 24, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.0773625176487415, "tokens": 131390, "cumulative_loss_tokens": 3065512, "grad_norm": 0.37109375, "lr": 3e-05, "finish_rate": 0.831, "comp_len": 554.4, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 43.3, "frames": {"chat": 237}, "mem_gb": 16.05}
40
+ {"step": 25, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.09143481954781472, "tokens": 139086, "cumulative_loss_tokens": 3204598, "grad_norm": 0.41796875, "lr": 3e-05, "finish_rate": 0.688, "comp_len": 586.9, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 46.8, "frames": {"chat": 237}, "mem_gb": 16.08}
41
+ checkpoint snapshot queued -> outputs/healed/keep5_math_offpolicy_top128/step0025
42
+ {"step": 26, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.07439380060321867, "tokens": 135786, "cumulative_loss_tokens": 3340384, "grad_norm": 0.400390625, "lr": 3e-05, "finish_rate": 0.785, "comp_len": 572.9, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 43.8, "frames": {"chat": 237}, "mem_gb": 15.97}
43
+ {"step": 27, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.06889912054688525, "tokens": 125008, "cumulative_loss_tokens": 3465392, "grad_norm": 0.353515625, "lr": 3e-05, "finish_rate": 0.861, "comp_len": 527.5, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 42.0, "frames": {"chat": 237}, "mem_gb": 16.0}
44
+ wandb:
45
+ wandb: Run history:
46
+ wandb: comp_len ▄▆▇▂▅▂██▅▁▃▃▆▃▂█▇▆▄▃▆▅▄▆▇▆▄
47
+ wandb: cumulative_loss_tokens ▁▁▂▂▂▂▃▃▃▃▄▄▄▄▅▅▅▆▆▆▆▇▇▇▇██
48
+ wandb: epoch ▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁
49
+ wandb: finish_rate ▆▅▂▇▄▇▃▃▅█▆▇▅▆▇▄▂▅▆▆▆▄▆▅▁▄▆
50
+ wandb: forward_topk_kl ██▇▅▄▃▃▂▂▂▂▂▂▃▂▂▂▂▁▁▁▂▁▁▂▁▁
51
+ wandb: grad_norm █▇▆▄▃▃▂▁▂▂▁▁▁▂▁▁▁▁▁▁▁▁▁▁▁▁▁
52
+ wandb: gsm8k_quick ▄▁█
53
+ wandb: gsm8k_quick_chat ▁█▁
54
+ wandb: gsm8k_quick_raw ▄▁█
55
+ wandb: lr ▁▂▃▄▅▅▆▇███████████████████
56
+ wandb: +6 ...
57
+ wandb:
58
+ wandb: Run summary:
59
+ wandb: comp_len 527.5
60
+ wandb: cumulative_loss_tokens 3465392
61
+ wandb: epoch 0
62
+ wandb: finish_rate 0.861
63
+ wandb: forward_topk_kl 0.0689
64
+ wandb: grad_norm 0.35352
65
+ wandb: gsm8k_quick 0.57812
66
+ wandb: gsm8k_quick_chat 0.5625
67
+ wandb: gsm8k_quick_raw 0.57812
68
+ wandb: lr 3e-05
69
+ wandb: +7 ...
70
+ wandb:
71
+ wandb: You can sync this run to the cloud by running:
72
+ wandb: wandb sync outputs/healed/keep5_math_offpolicy_top128/wandb/offline-run-20260713_193950-ixqqkrvl
73
+ wandb: Find logs at: outputs/healed/keep5_math_offpolicy_top128/wandb/offline-run-20260713_193950-ixqqkrvl/logs
74
+ wandb sync launched in background (pid 539761) -> outputs/healed/keep5_math_offpolicy_top128/wandb_sync.log
75
+ Traceback (most recent call last):
76
+ File "/home/henry/Documents/PythonProjects/variable-reap/scripts/11_distill_on_policy.py", line 871, in <module>
77
+ stdout=log, stderr=subprocess.STDOUT, start_new_session=True,
78
+ ^^^^^^
79
+ File "/home/henry/Documents/PythonProjects/variable-reap/scripts/11_distill_on_policy.py", line 695, in main
80
+ torch.cuda.empty_cache()
81
+
82
+ File "/home/henry/Documents/PythonProjects/variable-reap/src/glean/topk_targets.py", line 322, in distill_topk_microbatches
83
+ (loss * (n_tokens / total_tokens)).backward()
84
+ File "/home/henry/Documents/PythonProjects/variable-reap/.venv/lib/python3.12/site-packages/torch/_tensor.py", line 631, in backward
85
+ torch.autograd.backward(
86
+ File "/home/henry/Documents/PythonProjects/variable-reap/.venv/lib/python3.12/site-packages/torch/autograd/__init__.py", line 379, in backward
87
+ _engine_run_backward(
88
+ File "/home/henry/Documents/PythonProjects/variable-reap/.venv/lib/python3.12/site-packages/torch/autograd/graph.py", line 882, in _engine_run_backward
89
+ return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
90
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
91
+ KeyboardInterrupt
92
+ /home/henry/Documents/PythonProjects/variable-reap/.venv/lib/python3.12/site-packages/megablocks/grouped_gemm_util.py:10: UserWarning: Grouped GEMM not available.
93
+ warnings.warn('Grouped GEMM not available.')
94
+ wandb: [wandb.login()] Loaded credentials for https://api.wandb.ai from /home/henry/.netrc.
95
+ wandb: Currently logged in as: hbfreed to https://api.wandb.ai. Use `wandb login --relogin` to force relogin
96
+ wandb: setting up run ixqqkrvl
97
+ wandb: Tracking run with wandb version 0.28.0
98
+ wandb: Run data is saved locally in outputs/healed/keep5_math_offpolicy_top128/wandb/run-20260713_200636-ixqqkrvl
99
+ wandb: Run `wandb offline` to turn off syncing.
100
+ wandb: Resuming run keep5-math-offpolicy-top128-6.335M
101
+ wandb: ⭐️ View project at https://wandb.ai/hbfreed/glean-heal
102
+ wandb: 🚀 View run at https://wandb.ai/hbfreed/glean-heal/runs/ixqqkrvl
103
+
104
+ resumed student weights from outputs/healed/keep5_math_offpolicy_top128/step0025 (fresh optimizer, step counter at 0)
105
+ 12115 cached top-128 chat trajectories / 6,476,634 unique tokens | 52 steps/epoch | 52 total steps | student params 3.70B | teacher overlap=False
106
+ restored optimizer/scheduler state from step 25
107
+ wandb: updating run metadata
108
+ wandb: uploading summary
109
+ wandb:
110
+ wandb: Run summary:
111
+ wandb: comp_len 527.5
112
+ wandb: cumulative_loss_tokens 3465392
113
+ wandb: epoch 0
114
+ wandb: finish_rate 0.861
115
+ wandb: forward_topk_kl 0.0689
116
+ wandb: grad_norm 0.35352
117
+ wandb: gsm8k_quick 0.57812
118
+ wandb: gsm8k_quick_chat 0.5625
119
+ wandb: gsm8k_quick_raw 0.57812
120
+ wandb: lr 3e-05
121
+ wandb: +7 ...
122
+ wandb:
123
+ wandb: 🚀 View run keep5-math-offpolicy-top128-6.335M at: https://wandb.ai/hbfreed/glean-heal/runs/ixqqkrvl
124
+ wandb: ⭐️ View project at: https://wandb.ai/hbfreed/glean-heal
125
+ wandb: Synced 5 W&B file(s), 0 media file(s), 0 artifact file(s) and 0 other file(s)
126
+ wandb: Find logs at: outputs/healed/keep5_math_offpolicy_top128/wandb/run-20260713_200636-ixqqkrvl/logs
127
+ Traceback (most recent call last):
128
+ File "/home/henry/Documents/PythonProjects/variable-reap/scripts/11_distill_on_policy.py", line 884, in <module>
129
+ main()
130
+ File "/home/henry/Documents/PythonProjects/variable-reap/scripts/11_distill_on_policy.py", line 702, in main
131
+ total_loss, total_tokens = distill_topk_microbatches(
132
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^
133
+ File "/home/henry/Documents/PythonProjects/variable-reap/src/glean/topk_targets.py", line 322, in distill_topk_microbatches
134
+ (loss * (n_tokens / total_tokens)).backward()
135
+ File "/home/henry/Documents/PythonProjects/variable-reap/.venv/lib/python3.12/site-packages/torch/_tensor.py", line 631, in backward
136
+ torch.autograd.backward(
137
+ File "/home/henry/Documents/PythonProjects/variable-reap/.venv/lib/python3.12/site-packages/torch/autograd/__init__.py", line 379, in backward
138
+ _engine_run_backward(
139
+ File "/home/henry/Documents/PythonProjects/variable-reap/.venv/lib/python3.12/site-packages/torch/autograd/graph.py", line 882, in _engine_run_backward
140
+ return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
141
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
142
+ File "/home/henry/Documents/PythonProjects/variable-reap/.venv/lib/python3.12/site-packages/torch/autograd/function.py", line 317, in apply
143
+ return user_fn(self, *args)
144
+ ^^^^^^^^^^^^^^^^^^^^
145
+ File "/home/henry/Documents/PythonProjects/variable-reap/.venv/lib/python3.12/site-packages/torch/utils/checkpoint.py", line 331, in backward
146
+ torch.autograd.backward(outputs_with_grad, args_with_grad)
147
+ File "/home/henry/Documents/PythonProjects/variable-reap/.venv/lib/python3.12/site-packages/torch/autograd/__init__.py", line 379, in backward
148
+ _engine_run_backward(
149
+ File "/home/henry/Documents/PythonProjects/variable-reap/.venv/lib/python3.12/site-packages/torch/autograd/graph.py", line 882, in _engine_run_backward
150
+ return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
151
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
152
+ File "/home/henry/Documents/PythonProjects/variable-reap/.venv/lib/python3.12/site-packages/torch/autograd/function.py", line 317, in apply
153
+ return user_fn(self, *args)
154
+ ^^^^^^^^^^^^^^^^^^^^
155
+ File "/home/henry/Documents/PythonProjects/variable-reap/.venv/lib/python3.12/site-packages/stk/backend/autocast.py", line 36, in decorate_bwd
156
+ return bwd(*args, **kwargs)
157
+ ^^^^^^^^^^^^^^^^^^^^
158
+ File "/home/henry/Documents/PythonProjects/variable-reap/.venv/lib/python3.12/site-packages/stk/backend/sputnik.py", line 284, in backward
159
+ dlhs = _lhs_gradient(op,
160
+ ^^^^^^^^^^^^^^^^^
161
+ File "/home/henry/Documents/PythonProjects/variable-reap/.venv/lib/python3.12/site-packages/stk/backend/sputnik.py", line 71, in _lhs_gradient
162
+ out = _call_helper(op, lhs, a, b, trans_a, trans_b)
163
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
164
+ File "/home/henry/Documents/PythonProjects/variable-reap/.venv/lib/python3.12/site-packages/stk/backend/sputnik.py", line 42, in _call_helper
165
+ return op(*args)
166
+ ^^^^^^^^^
167
+ File "/home/henry/Documents/PythonProjects/variable-reap/.venv/lib/python3.12/site-packages/torch/autograd/function.py", line 596, in apply
168
+ return super().apply(*args, **kwargs) # type: ignore[misc]
169
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
170
+ File "/home/henry/Documents/PythonProjects/variable-reap/.venv/lib/python3.12/site-packages/stk/backend/autocast.py", line 28, in decorate_fwd
171
+ return fwd(*args, **kwargs)
172
+ ^^^^^^^^^^^^^^^^^^^^
173
+ File "/home/henry/Documents/PythonProjects/variable-reap/.venv/lib/python3.12/site-packages/stk/backend/sputnik.py", line 111, in forward
174
+ out = torch.empty(
175
+ ^^^^^^^^^^^^
176
+ torch.OutOfMemoryError: CUDA out of memory. Tried to allocate 76.00 MiB. GPU 0 has a total capacity of 23.56 GiB of which 25.38 MiB is free. Including non-PyTorch memory, this process has 23.51 GiB memory in use. Of the allocated memory 22.98 GiB is allocated by PyTorch, and 216.91 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation. See documentation for Memory Management (https://docs.pytorch.org/docs/stable/notes/cuda.html#optimizing-memory-usage-with-pytorch-cuda-alloc-conf)
177
+ /home/henry/Documents/PythonProjects/variable-reap/.venv/lib/python3.12/site-packages/megablocks/grouped_gemm_util.py:10: UserWarning: Grouped GEMM not available.
178
+ warnings.warn('Grouped GEMM not available.')
179
+ wandb: [wandb.login()] Loaded credentials for https://api.wandb.ai from /home/henry/.netrc.
180
+ wandb: Currently logged in as: hbfreed to https://api.wandb.ai. Use `wandb login --relogin` to force relogin
181
+ wandb: setting up run ixqqkrvl
182
+ wandb: Tracking run with wandb version 0.28.0
183
+ wandb: Run data is saved locally in outputs/healed/keep5_math_offpolicy_top128/wandb/run-20260713_201005-ixqqkrvl
184
+ wandb: Run `wandb offline` to turn off syncing.
185
+ wandb: Resuming run keep5-math-offpolicy-top128-6.335M
186
+ wandb: ⭐️ View project at https://wandb.ai/hbfreed/glean-heal
187
+ wandb: 🚀 View run at https://wandb.ai/hbfreed/glean-heal/runs/ixqqkrvl
188
+
189
+ resumed student weights from outputs/healed/keep5_math_offpolicy_top128/step0025 (fresh optimizer, step counter at 0)
190
+ 12115 cached top-128 chat trajectories / 6,476,634 unique tokens | 52 steps/epoch | 52 total steps | student params 3.70B | teacher overlap=False
191
+ restored optimizer/scheduler state from step 25; rebuilt 0 paged buffers
192
+ {"step": 26, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.07439380060321867, "tokens": 135786, "cumulative_loss_tokens": 3340384, "grad_norm": 0.400390625, "lr": 3e-05, "finish_rate": 0.785, "comp_len": 572.9, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 63.6, "frames": {"chat": 237}, "mem_gb": 15.86}
193
+ wandb: WARNING Tried to log to step 26 that is less than the current step 28. Steps must be monotonically increasing, so this data will be ignored. See https://wandb.me/define-metric to log data out of order.
194
+ wandb: updating run metadata
195
+ wandb: uploading summary
196
+ wandb:
197
+ wandb: Run summary:
198
+ wandb: comp_len 527.5
199
+ wandb: cumulative_loss_tokens 3465392
200
+ wandb: epoch 0
201
+ wandb: finish_rate 0.861
202
+ wandb: forward_topk_kl 0.0689
203
+ wandb: grad_norm 0.35352
204
+ wandb: gsm8k_quick 0.57812
205
+ wandb: gsm8k_quick_chat 0.5625
206
+ wandb: gsm8k_quick_raw 0.57812
207
+ wandb: lr 3e-05
208
+ wandb: +7 ...
209
+ wandb:
210
+ wandb: 🚀 View run keep5-math-offpolicy-top128-6.335M at: https://wandb.ai/hbfreed/glean-heal/runs/ixqqkrvl
211
+ wandb: ⭐️ View project at: https://wandb.ai/hbfreed/glean-heal
212
+ wandb: Synced 5 W&B file(s), 0 media file(s), 0 artifact file(s) and 0 other file(s)
213
+ wandb: Find logs at: outputs/healed/keep5_math_offpolicy_top128/wandb/run-20260713_201005-ixqqkrvl/logs
214
+ Traceback (most recent call last):
215
+ File "/home/henry/Documents/PythonProjects/variable-reap/scripts/11_distill_on_policy.py", line 916, in <module>
216
+ File "/home/henry/Documents/PythonProjects/variable-reap/scripts/11_distill_on_policy.py", line 734, in main
217
+ opt.zero_grad(set_to_none=True)
218
+
219
+ File "/home/henry/Documents/PythonProjects/variable-reap/src/glean/topk_targets.py", line 322, in distill_topk_microbatches
220
+ (loss * (n_tokens / total_tokens)).backward()
221
+ File "/home/henry/Documents/PythonProjects/variable-reap/.venv/lib/python3.12/site-packages/torch/_tensor.py", line 631, in backward
222
+ torch.autograd.backward(
223
+ File "/home/henry/Documents/PythonProjects/variable-reap/.venv/lib/python3.12/site-packages/torch/autograd/__init__.py", line 379, in backward
224
+ _engine_run_backward(
225
+ File "/home/henry/Documents/PythonProjects/variable-reap/.venv/lib/python3.12/site-packages/torch/autograd/graph.py", line 882, in _engine_run_backward
226
+ return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
227
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
228
+ KeyboardInterrupt
229
+ /home/henry/Documents/PythonProjects/variable-reap/.venv/lib/python3.12/site-packages/megablocks/grouped_gemm_util.py:10: UserWarning: Grouped GEMM not available.
230
+ warnings.warn('Grouped GEMM not available.')
231
+ wandb: [wandb.login()] Loaded credentials for https://api.wandb.ai from /home/henry/.netrc.
232
+ wandb: Currently logged in as: hbfreed to https://api.wandb.ai. Use `wandb login --relogin` to force relogin
233
+ wandb: setting up run ixqqkrvl
234
+ wandb: Tracking run with wandb version 0.28.0
235
+ wandb: Run data is saved locally in outputs/healed/keep5_math_offpolicy_top128/wandb/run-20260713_201227-ixqqkrvl
236
+ wandb: Run `wandb offline` to turn off syncing.
237
+ wandb: Resuming run keep5-math-offpolicy-top128-6.335M
238
+ wandb: ⭐️ View project at https://wandb.ai/hbfreed/glean-heal
239
+ wandb: 🚀 View run at https://wandb.ai/hbfreed/glean-heal/runs/ixqqkrvl
240
+
241
+ resumed student weights from outputs/healed/keep5_math_offpolicy_top128/step0025 (fresh optimizer, step counter at 0)
242
+ 12115 cached top-128 chat trajectories / 6,476,634 unique tokens | 52 steps/epoch | 52 total steps | student params 3.70B | teacher overlap=False
243
+ restored optimizer/scheduler state from step 25; rebuilt 0 paged buffers
244
+ {"step": 26, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.07439380060321867, "tokens": 135786, "cumulative_loss_tokens": 3340384, "grad_norm": 0.400390625, "lr": 3e-05, "finish_rate": 0.785, "comp_len": 572.9, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 63.5, "frames": {"chat": 237}, "mem_gb": 15.86}
245
+ /home/henry/Documents/PythonProjects/variable-reap/.venv/lib/python3.12/site-packages/megablocks/grouped_gemm_util.py:10: UserWarning: Grouped GEMM not available.
246
+ warnings.warn('Grouped GEMM not available.')
247
+ wandb: [wandb.login()] Loaded credentials for https://api.wandb.ai from /home/henry/.netrc.
248
+ wandb: Currently logged in as: hbfreed to https://api.wandb.ai. Use `wandb login --relogin` to force relogin
249
+ wandb: Tracking run with wandb version 0.28.0
250
+ wandb: Run data is saved locally in outputs/healed/keep5_math_offpolicy_top128/wandb/run-20260713_210942-ixqqkrvl
251
+ wandb: Run `wandb offline` to turn off syncing.
252
+ wandb: Resuming run keep5-math-offpolicy-top128-6.335M
253
+ wandb: ⭐️ View project at https://wandb.ai/hbfreed/glean-heal
254
+ wandb: 🚀 View run at https://wandb.ai/hbfreed/glean-heal/runs/ixqqkrvl
255
+
256
+ resumed student weights from outputs/healed/keep5_math_offpolicy_top128/step0025 (fresh optimizer, step counter at 0)
257
+ 12115 cached top-128 chat trajectories / 6,476,634 unique tokens | 52 steps/epoch | 52 total steps | student params 3.70B | teacher overlap=False
258
+ restored optimizer/scheduler state from step 25; rebuilt 252 paged buffers
259
+ {"step": 26, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.07439380060321867, "tokens": 135786, "cumulative_loss_tokens": 3340384, "grad_norm": 0.400390625, "lr": 3e-05, "finish_rate": 0.785, "comp_len": 572.9, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 49.7, "frames": {"chat": 237}, "mem_gb": 15.86}
260
+ {"step": 27, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.0688901165153168, "tokens": 125008, "cumulative_loss_tokens": 3465392, "grad_norm": 0.353515625, "lr": 3e-05, "finish_rate": 0.861, "comp_len": 527.5, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 41.4, "frames": {"chat": 237}, "mem_gb": 16.0}
261
+ {"step": 28, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.07942616952093563, "tokens": 123465, "cumulative_loss_tokens": 3588857, "grad_norm": 0.3828125, "lr": 3e-05, "finish_rate": 0.861, "comp_len": 520.9, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 41.7, "frames": {"chat": 237}, "mem_gb": 16.05}
262
+ {"step": 29, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.07290488002502521, "tokens": 125066, "cumulative_loss_tokens": 3713923, "grad_norm": 0.375, "lr": 3e-05, "finish_rate": 0.852, "comp_len": 527.7, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 41.9, "frames": {"chat": 237}, "mem_gb": 16.07}
263
+ {"step": 30, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.06405109558663372, "tokens": 116845, "cumulative_loss_tokens": 3830768, "grad_norm": 0.36328125, "lr": 3e-05, "finish_rate": 0.873, "comp_len": 493.0, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 39.2, "frames": {"chat": 237}, "mem_gb": 16.02}
264
+ The attention mask is not set and cannot be inferred from input because pad token is same as eos token. As a consequence, you may observe unexpected behavior. Please pass your input's `attention_mask` to obtain reliable results.
265
+ [eval step 30] sample: 'To solve the problem of finding the greatest common divisor (GCD) of 51 and 12767 using Python and SymPy, we can follow these steps:\n\n1. **Import SymPy**: SymPy provides a built-in function `gcd` to c'
266
+ {"step": 30, "gsm8k_n": 64, "gsm8k_quick_raw": 0.59375, "gsm8k_quick": 0.59375, "gsm8k_quick_chat": 0.546875, "t_eval_s": 99.6}
267
+ {"step": 31, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.06380675256369958, "tokens": 118784, "cumulative_loss_tokens": 3949552, "grad_norm": 0.35546875, "lr": 3e-05, "finish_rate": 0.848, "comp_len": 501.2, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 40.7, "frames": {"chat": 237}, "mem_gb": 18.69}
268
+ {"step": 32, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.07116466968754641, "tokens": 126692, "cumulative_loss_tokens": 4076244, "grad_norm": 0.365234375, "lr": 3e-05, "finish_rate": 0.819, "comp_len": 534.6, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 42.7, "frames": {"chat": 237}, "mem_gb": 16.07}
269
+ {"step": 33, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.07169101262025185, "tokens": 127534, "cumulative_loss_tokens": 4203778, "grad_norm": 0.361328125, "lr": 3e-05, "finish_rate": 0.819, "comp_len": 538.1, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 43.8, "frames": {"chat": 237}, "mem_gb": 16.1}
270
+ {"step": 34, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.058325077411156355, "tokens": 117601, "cumulative_loss_tokens": 4321379, "grad_norm": 0.3359375, "lr": 3e-05, "finish_rate": 0.916, "comp_len": 496.2, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 39.4, "frames": {"chat": 237}, "mem_gb": 15.85}
271
+ {"step": 35, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.09290853046750798, "tokens": 124664, "cumulative_loss_tokens": 4446043, "grad_norm": 0.5703125, "lr": 3e-05, "finish_rate": 0.844, "comp_len": 526.0, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 40.7, "frames": {"chat": 237}, "mem_gb": 16.0}
272
+ {"step": 36, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.0604027349406713, "tokens": 120567, "cumulative_loss_tokens": 4566610, "grad_norm": 0.3359375, "lr": 3e-05, "finish_rate": 0.873, "comp_len": 508.7, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 40.4, "frames": {"chat": 237}, "mem_gb": 15.91}
273
+ {"step": 37, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.06577399537383462, "tokens": 134445, "cumulative_loss_tokens": 4701055, "grad_norm": 0.328125, "lr": 3e-05, "finish_rate": 0.759, "comp_len": 567.3, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 43.3, "frames": {"chat": 237}, "mem_gb": 16.05}
274
+ {"step": 38, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.0683866965607533, "tokens": 131408, "cumulative_loss_tokens": 4832463, "grad_norm": 0.353515625, "lr": 3e-05, "finish_rate": 0.785, "comp_len": 554.5, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 45.6, "frames": {"chat": 237}, "mem_gb": 16.18}
275
+ {"step": 39, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.05965062385041473, "tokens": 128642, "cumulative_loss_tokens": 4961105, "grad_norm": 0.328125, "lr": 3e-05, "finish_rate": 0.852, "comp_len": 542.8, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 41.0, "frames": {"chat": 237}, "mem_gb": 16.04}
276
+ {"step": 40, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.05333902896525456, "tokens": 125546, "cumulative_loss_tokens": 5086651, "grad_norm": 0.33984375, "lr": 3e-05, "finish_rate": 0.852, "comp_len": 529.7, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 40.9, "frames": {"chat": 237}, "mem_gb": 15.95}
277
+ [eval step 40] sample: 'To solve the problem of finding the greatest common divisor (GCD) of 51 and 12767 using Python and SymPy, we can follow these steps:\n\n1. **Import SymPy**: SymPy provides a built-in function `gcd` to c'
278
+ {"step": 40, "gsm8k_n": 64, "gsm8k_quick_raw": 0.546875, "gsm8k_quick": 0.546875, "gsm8k_quick_chat": 0.546875, "t_eval_s": 91.8}
279
+ {"step": 41, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.05690588536906052, "tokens": 139086, "cumulative_loss_tokens": 5225737, "grad_norm": 0.318359375, "lr": 3e-05, "finish_rate": 0.781, "comp_len": 586.9, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 43.6, "frames": {"chat": 237}, "mem_gb": 18.68}
280
+ {"step": 42, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.058374759728881946, "tokens": 118340, "cumulative_loss_tokens": 5344077, "grad_norm": 0.31640625, "lr": 3e-05, "finish_rate": 0.924, "comp_len": 499.3, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 40.0, "frames": {"chat": 237}, "mem_gb": 15.95}
281
+ {"step": 43, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.07019935327672881, "tokens": 122214, "cumulative_loss_tokens": 5466291, "grad_norm": 0.35546875, "lr": 3e-05, "finish_rate": 0.823, "comp_len": 515.7, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 41.1, "frames": {"chat": 237}, "mem_gb": 16.02}
282
+ {"step": 44, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.06907294226702002, "tokens": 127145, "cumulative_loss_tokens": 5593436, "grad_norm": 0.34765625, "lr": 3e-05, "finish_rate": 0.857, "comp_len": 536.5, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 44.0, "frames": {"chat": 237}, "mem_gb": 16.05}
283
+ {"step": 45, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.05835399558657501, "tokens": 129772, "cumulative_loss_tokens": 5723208, "grad_norm": 0.31640625, "lr": 3e-05, "finish_rate": 0.81, "comp_len": 547.6, "t_data_s": 0.1, "t_rollout_s": 0.0, "t_step_s": 43.6, "frames": {"chat": 237}, "mem_gb": 15.92}
284
+ {"step": 46, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.059369281406764264, "tokens": 114125, "cumulative_loss_tokens": 5837333, "grad_norm": 0.3359375, "lr": 3e-05, "finish_rate": 0.928, "comp_len": 481.5, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 38.4, "frames": {"chat": 237}, "mem_gb": 15.94}
285
+ {"step": 47, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.05725491126893169, "tokens": 116580, "cumulative_loss_tokens": 5953913, "grad_norm": 0.345703125, "lr": 3e-05, "finish_rate": 0.928, "comp_len": 491.9, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 40.3, "frames": {"chat": 237}, "mem_gb": 15.99}
286
+ {"step": 48, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.05294738627785973, "tokens": 124833, "cumulative_loss_tokens": 6078746, "grad_norm": 0.314453125, "lr": 3e-05, "finish_rate": 0.869, "comp_len": 526.7, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 41.2, "frames": {"chat": 237}, "mem_gb": 16.04}
287
+ {"step": 49, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.05482089092956944, "tokens": 127946, "cumulative_loss_tokens": 6206692, "grad_norm": 0.302734375, "lr": 3e-05, "finish_rate": 0.844, "comp_len": 539.9, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 41.7, "frames": {"chat": 237}, "mem_gb": 15.92}
288
+ {"step": 50, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.08619242395973124, "tokens": 128977, "cumulative_loss_tokens": 6335669, "grad_norm": 0.390625, "lr": 3e-05, "finish_rate": 0.746, "comp_len": 575.8, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 44.6, "frames": {"chat": 224}, "mem_gb": 16.05}
289
+ [eval step 50] sample: 'To solve the problem of finding the greatest common divisor (GCD) of 51 and 12767 using Python and SymPy, we can follow these steps:\n\n1. **Import the necessary library:**\n We will use the `sympy` li'
290
+ {"step": 50, "gsm8k_n": 64, "gsm8k_quick_raw": 0.515625, "gsm8k_quick": 0.515625, "gsm8k_quick_chat": 0.53125, "t_eval_s": 95.6}
291
+ checkpoint snapshot queued -> outputs/healed/keep5_math_offpolicy_top128/step0050
292
+ wandb: updating run metadata
293
+ wandb: uploading output.log; uploading wandb-summary.json; uploading config.yaml
294
+ wandb:
295
+ wandb: Run history:
296
+ wandb: comp_len ▇▄▄▄▂▂▅▅▂▄▃▇▆▅▄█▂▃▅▅▁▂▄▅▇
297
+ wandb: cumulative_loss_tokens ▁▁▂▂▂▂▃▃▃▄▄▄▄▅▅▅▆▆▆▇▇▇▇██
298
+ wandb: epoch ▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁
299
+ wandb: finish_rate ▃▅▅▅▆▅▄▄█▅▆▂▃▅▅▂█▄▅▃██▆▅▁
300
+ wandb: forward_topk_kl ▅▄▆▄▃▃▄▄▂█▂▃▄▂▁▂▂▄▄▂▂▂▁▁▇
301
+ wandb: grad_norm ▄▂▃▃▃▂▃▃▂█▂▂▂▂▂▁▁▂▂▁▂▂▁▁▃
302
+ wandb: gsm8k_quick █▄▁
303
+ wandb: gsm8k_quick_chat ██▁
304
+ wandb: gsm8k_quick_raw █▄▁
305
+ wandb: lr ▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁
306
+ wandb: +6 ...
307
+ wandb:
308
+ wandb: Run summary:
309
+ wandb: comp_len 575.8
310
+ wandb: cumulative_loss_tokens 6335669
311
+ wandb: epoch 0
312
+ wandb: finish_rate 0.746
313
+ wandb: forward_topk_kl 0.08619
314
+ wandb: grad_norm 0.39062
315
+ wandb: gsm8k_quick 0.51562
316
+ wandb: gsm8k_quick_chat 0.53125
317
+ wandb: gsm8k_quick_raw 0.51562
318
+ wandb: lr 3e-05
319
+ wandb: +7 ...
320
+ wandb:
321
+ wandb: 🚀 View run keep5-math-offpolicy-top128-6.335M at: https://wandb.ai/hbfreed/glean-heal/runs/ixqqkrvl
322
+ wandb: ⭐️ View project at: https://wandb.ai/hbfreed/glean-heal
323
+ wandb: Synced 5 W&B file(s), 0 media file(s), 0 artifact file(s) and 0 other file(s)
324
+ wandb: Find logs at: outputs/healed/keep5_math_offpolicy_top128/wandb/run-20260713_210942-ixqqkrvl/logs
healed/knee0924_queue.log ADDED
@@ -0,0 +1,43 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ [12:06:52] keep50 artifact exists
2
+ [12:06:52] short-healing keep50
3
+ [12:42:54] keep50 done: {"pre_heal_ppl":25.590167999267578,"post_heal_ppl":18.254446029663086}
4
+ [12:42:54] keep40 artifact exists
5
+ [12:42:54] short-healing keep40
6
+ [13:17:33] keep40 done: {"pre_heal_ppl":34.65605163574219,"post_heal_ppl":21.057231903076172}
7
+ [13:17:33] materializing keep30 (CPU surgery)
8
+ [13:17:49] short-healing keep30
9
+ [13:50:35] keep30 done: {"pre_heal_ppl":69.09920501708984,"post_heal_ppl":25.705188751220703}
10
+ [13:50:35] materializing keep25 (CPU surgery)
11
+ [13:50:49] short-healing keep25
12
+ [14:22:49] keep25 done: {"pre_heal_ppl":105.4487533569336,"post_heal_ppl":28.708465576171875}
13
+ [14:22:49] materializing keep20 (CPU surgery)
14
+ [14:23:04] short-healing keep20
15
+ [14:54:27] keep20 done: {"pre_heal_ppl":164.80226135253906,"post_heal_ppl":33.18513870239258}
16
+ [14:54:27] knee heal queue complete
17
+ [
18
+ {
19
+ "student": "outputs/pruned/knee0924/keep20",
20
+ "pre_heal_ppl": 164.80226135253906,
21
+ "post_heal_ppl": 33.18513870239258
22
+ },
23
+ {
24
+ "student": "outputs/pruned/knee0924/keep25",
25
+ "pre_heal_ppl": 105.4487533569336,
26
+ "post_heal_ppl": 28.708465576171875
27
+ },
28
+ {
29
+ "student": "outputs/pruned/knee0924/keep30",
30
+ "pre_heal_ppl": 69.09920501708984,
31
+ "post_heal_ppl": 25.705188751220703
32
+ },
33
+ {
34
+ "student": "outputs/pruned/knee0924/keep40",
35
+ "pre_heal_ppl": 34.65605163574219,
36
+ "post_heal_ppl": 21.057231903076172
37
+ },
38
+ {
39
+ "student": "outputs/pruned/knee0924/keep50",
40
+ "pre_heal_ppl": 25.590167999267578,
41
+ "post_heal_ppl": 18.254446029663086
42
+ }
43
+ ]
healed/opd_warm.log ADDED
@@ -0,0 +1,59 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ wandb: Tracking run with wandb version 0.28.0
2
+ wandb: W&B syncing is set to `offline` in this directory. Run `wandb online` or set WANDB_MODE=online to enable cloud syncing.
3
+ wandb: Run data is saved locally in outputs/healed/opd_warm_keep50/wandb/offline-run-20260802_001724-yclwxiwa
4
+ wandb: View this run in the terminal with `wandb leet`
5
+
6
+ teacher converted to fused MoE path (allenai/OLMoE-1B-7B-0125-Instruct)
7
+
8
+ reference anchor loaded from outputs/healed/keep50_offpolicy_warmup_s1224/step0150 on cuda:0 (beta=0.05)
9
+
10
+ starting vllm rollout server on GPU 2 (port 8377) ...
11
+ vllm server healthy in 36s
12
+ mixture distillation: 58360 gold trajectories from outputs/teacher_trajectories/dolci_combined_top128, lambda=0.5 decay=0.0
13
+
14
+ The attention mask is not set and cannot be inferred from input because pad token is same as eos token. As a consequence, you may observe unexpected behavior. Please pass your input's `attention_mask` to obtain reliable results.
15
+ 168191 prompts | 2627 steps/epoch | 120 total steps | student params 3.70B | teacher overlap=True
16
+ {"step": 1, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 1.0715529867857179, "tokens": 330677, "cumulative_loss_tokens": 330677, "grad_norm": 18.0, "lr": 2.0000000000000003e-06, "finish_rate": 0.66, "comp_len": 1291.7, "dropped_truncated": 0, "gold_loss": 0.4816, "gold_lambda": 0.5, "rep_ratio": 7.359, "t_data_s": 0.0, "t_rollout_s": 117.2, "t_step_s": 230.0, "t_refresh_s": 0.3, "mem_gb": 10.4, "mem_gb_teacher": 20.7}
17
+ [eval step 1] sample: 'To "faye" "n" "m" "n" "m" "n" "m" "n" "m" "n" "m" "n" "m" "n" "m" "n" "m" "n" "m" "n" "m'
18
+ {"step": 1, "gsm8k_n": 256, "gsm8k_quick_chat": 0.46875, "t_eval_s": 40.3}
19
+ {"step": 2, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 1.0476409321224727, "tokens": 305425, "cumulative_loss_tokens": 636102, "grad_norm": 19.25, "lr": 3e-06, "finish_rate": 0.699, "comp_len": 1193.1, "dropped_truncated": 0, "gold_loss": 0.4636, "gold_lambda": 0.5, "rep_ratio": 6.842, "t_data_s": 0.0, "t_rollout_s": 113.7, "t_step_s": 214.3, "t_refresh_s": 0.3, "mem_gb": 10.55, "mem_gb_teacher": 20.67}
20
+ {"step": 3, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 1.2240935077801478, "tokens": 302635, "cumulative_loss_tokens": 938737, "grad_norm": 16.75, "lr": 4.000000000000001e-06, "finish_rate": 0.711, "comp_len": 1182.2, "dropped_truncated": 0, "gold_loss": 0.4989, "gold_lambda": 0.5, "rep_ratio": 8.206, "t_data_s": 0.0, "t_rollout_s": 109.6, "t_step_s": 211.3, "t_refresh_s": 0.3, "mem_gb": 10.46, "mem_gb_teacher": 20.63}
21
+ {"step": 4, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 1.1232984279951, "tokens": 314442, "cumulative_loss_tokens": 1253179, "grad_norm": 13.625, "lr": 5e-06, "finish_rate": 0.723, "comp_len": 1228.3, "dropped_truncated": 0, "gold_loss": 0.4989, "gold_lambda": 0.5, "rep_ratio": 4.007, "t_data_s": 0.0, "t_rollout_s": 112.2, "t_step_s": 215.0, "t_refresh_s": 0.3, "mem_gb": 10.24, "mem_gb_teacher": 20.57}
22
+ {"step": 5, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 1.196489499796483, "tokens": 314169, "cumulative_loss_tokens": 1567348, "grad_norm": 12.0625, "lr": 6e-06, "finish_rate": 0.715, "comp_len": 1227.2, "dropped_truncated": 0, "gold_loss": 1.7581, "gold_lambda": 0.5, "rep_ratio": 7.457, "t_data_s": 0.0, "t_rollout_s": 113.5, "t_step_s": 213.6, "t_refresh_s": 0.3, "mem_gb": 10.48, "mem_gb_teacher": 20.64}
23
+ {"step": 6, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 1.174327937240284, "tokens": 302861, "cumulative_loss_tokens": 1870209, "grad_norm": 10.125, "lr": 7e-06, "finish_rate": 0.715, "comp_len": 1183.1, "dropped_truncated": 0, "gold_loss": 1.4087, "gold_lambda": 0.5, "rep_ratio": 6.804, "t_data_s": 0.0, "t_rollout_s": 112.3, "t_step_s": 214.1, "t_refresh_s": 0.3, "mem_gb": 10.49, "mem_gb_teacher": 20.64}
24
+ {"step": 7, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 1.142888064626811, "tokens": 259225, "cumulative_loss_tokens": 2129434, "grad_norm": 8.0625, "lr": 8.000000000000001e-06, "finish_rate": 0.875, "comp_len": 1012.6, "dropped_truncated": 0, "gold_loss": 1.7253, "gold_lambda": 0.5, "rep_ratio": 2.291, "t_data_s": 0.0, "t_rollout_s": 85.8, "t_step_s": 180.9, "t_refresh_s": 0.3, "mem_gb": 10.5, "mem_gb_teacher": 20.66}
25
+ {"step": 8, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 1.2367885471050992, "tokens": 275165, "cumulative_loss_tokens": 2404599, "grad_norm": 10.0625, "lr": 9e-06, "finish_rate": 0.777, "comp_len": 1074.9, "dropped_truncated": 0, "gold_loss": 1.791, "gold_lambda": 0.5, "rep_ratio": 4.278, "t_data_s": 0.0, "t_rollout_s": 101.9, "t_step_s": 198.9, "t_refresh_s": 0.3, "mem_gb": 10.37, "mem_gb_teacher": 20.6}
26
+ {"step": 9, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 1.2960635659850794, "tokens": 287956, "cumulative_loss_tokens": 2692555, "grad_norm": 10.5, "lr": 1e-05, "finish_rate": 0.777, "comp_len": 1124.8, "dropped_truncated": 0, "gold_loss": 0.4427, "gold_lambda": 0.5, "rep_ratio": 3.539, "t_data_s": 0.0, "t_rollout_s": 107.3, "t_step_s": 217.4, "t_refresh_s": 0.3, "mem_gb": 10.37, "mem_gb_teacher": 20.61}
27
+ {"step": 10, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 1.1268998827677594, "tokens": 256196, "cumulative_loss_tokens": 2948751, "grad_norm": 8.1875, "lr": 1e-05, "finish_rate": 0.852, "comp_len": 1000.8, "dropped_truncated": 0, "gold_loss": 0.4179, "gold_lambda": 0.5, "rep_ratio": 5.761, "t_data_s": 0.0, "t_rollout_s": 83.3, "t_step_s": 177.5, "t_refresh_s": 0.3, "mem_gb": 10.63, "mem_gb_teacher": 20.7}
28
+ [eval step 10] sample: 'To be a-m-on-n-on-on-on-on-on-on-on-on-on-on-on-on-on-on-on-on-on-on-on-on-on-on-on-on-on-on-on-'
29
+ {"step": 11, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 1.1852989046363867, "tokens": 249794, "cumulative_loss_tokens": 3198545, "grad_norm": 9.4375, "lr": 1e-05, "finish_rate": 0.879, "comp_len": 975.8, "dropped_truncated": 0, "gold_loss": 1.8732, "gold_lambda": 0.5, "rep_ratio": 2.592, "t_data_s": 0.0, "t_rollout_s": 79.1, "t_step_s": 170.9, "t_refresh_s": 0.3, "mem_gb": 10.17, "mem_gb_teacher": 20.53}
30
+ {"step": 12, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 1.0672082473087046, "tokens": 235329, "cumulative_loss_tokens": 3433874, "grad_norm": 6.40625, "lr": 1e-05, "finish_rate": 0.918, "comp_len": 919.3, "dropped_truncated": 0, "gold_loss": 1.7626, "gold_lambda": 0.5, "rep_ratio": 2.465, "t_data_s": 0.0, "t_rollout_s": 72.1, "t_step_s": 161.9, "t_refresh_s": 0.3, "mem_gb": 10.51, "mem_gb_teacher": 20.65}
31
+ {"step": 13, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 1.1896850641550707, "tokens": 196953, "cumulative_loss_tokens": 3630827, "grad_norm": 6.78125, "lr": 1e-05, "finish_rate": 0.945, "comp_len": 769.3, "dropped_truncated": 0, "gold_loss": 1.8544, "gold_lambda": 0.5, "rep_ratio": 2.39, "t_data_s": 0.0, "t_rollout_s": 57.2, "t_step_s": 138.9, "t_refresh_s": 0.3, "mem_gb": 10.31, "mem_gb_teacher": 20.59}
32
+ {"step": 14, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 1.110568549902207, "tokens": 195354, "cumulative_loss_tokens": 3826181, "grad_norm": 5.25, "lr": 1e-05, "finish_rate": 0.938, "comp_len": 763.1, "dropped_truncated": 0, "gold_loss": 1.7907, "gold_lambda": 0.5, "rep_ratio": 2.662, "t_data_s": 0.0, "t_rollout_s": 57.3, "t_step_s": 140.7, "t_refresh_s": 0.3, "mem_gb": 10.2, "mem_gb_teacher": 20.55}
33
+ {"step": 15, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 1.0880019511972423, "tokens": 211631, "cumulative_loss_tokens": 4037812, "grad_norm": 4.78125, "lr": 1e-05, "finish_rate": 0.938, "comp_len": 826.7, "dropped_truncated": 0, "gold_loss": 0.4573, "gold_lambda": 0.5, "rep_ratio": 4.91, "t_data_s": 0.0, "t_rollout_s": 61.5, "t_step_s": 147.1, "t_refresh_s": 0.3, "mem_gb": 10.26, "mem_gb_teacher": 20.53}
34
+ {"step": 16, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.9728942313239658, "tokens": 194169, "cumulative_loss_tokens": 4231981, "grad_norm": 7.90625, "lr": 1e-05, "finish_rate": 0.953, "comp_len": 758.5, "dropped_truncated": 0, "gold_loss": 0.4647, "gold_lambda": 0.5, "rep_ratio": 12.498, "t_data_s": 0.0, "t_rollout_s": 57.5, "t_step_s": 140.4, "t_refresh_s": 0.3, "mem_gb": 10.38, "mem_gb_teacher": 20.56}
35
+ {"step": 17, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 1.0340790669212048, "tokens": 182968, "cumulative_loss_tokens": 4414949, "grad_norm": 8.5625, "lr": 1e-05, "finish_rate": 0.945, "comp_len": 714.7, "dropped_truncated": 0, "gold_loss": 1.5083, "gold_lambda": 0.5, "rep_ratio": 2.658, "t_data_s": 0.0, "t_rollout_s": 55.6, "t_step_s": 137.3, "t_refresh_s": 0.3, "mem_gb": 10.54, "mem_gb_teacher": 20.66}
36
+ {"step": 18, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.9951681053661686, "tokens": 192973, "cumulative_loss_tokens": 4607922, "grad_norm": 4.09375, "lr": 1e-05, "finish_rate": 0.965, "comp_len": 753.8, "dropped_truncated": 0, "gold_loss": 1.5254, "gold_lambda": 0.5, "rep_ratio": 2.551, "t_data_s": 0.0, "t_rollout_s": 58.0, "t_step_s": 139.2, "t_refresh_s": 0.3, "mem_gb": 10.28, "mem_gb_teacher": 20.58}
37
+ {"step": 19, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.941273376354514, "tokens": 172849, "cumulative_loss_tokens": 4780771, "grad_norm": 3.65625, "lr": 1e-05, "finish_rate": 0.969, "comp_len": 675.2, "dropped_truncated": 0, "gold_loss": 1.7476, "gold_lambda": 0.5, "rep_ratio": 2.772, "t_data_s": 0.0, "t_rollout_s": 52.5, "t_step_s": 129.2, "t_refresh_s": 0.3, "mem_gb": 10.44, "mem_gb_teacher": 20.63}
38
+ {"step": 20, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.8855722967075427, "tokens": 178588, "cumulative_loss_tokens": 4959359, "grad_norm": 3.34375, "lr": 1e-05, "finish_rate": 0.969, "comp_len": 697.6, "dropped_truncated": 0, "gold_loss": 1.8831, "gold_lambda": 0.5, "rep_ratio": 2.626, "t_data_s": 0.0, "t_rollout_s": 53.1, "t_step_s": 129.8, "t_refresh_s": 0.3, "mem_gb": 10.64, "mem_gb_teacher": 20.63}
39
+ [eval step 20] sample: 'To be a a "m" of the "m" of the "m" of the "m" of the "m" of the "m" of the "m" of the "m" of the "m" of the "m" of the "m" of the "m" of the'
40
+ {"step": 20, "gsm8k_n": 256, "gsm8k_quick_chat": 0.49609375, "t_eval_s": 43.9}
41
+ {"step": 21, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 1.0167943671227804, "tokens": 170868, "cumulative_loss_tokens": 5130227, "grad_norm": 4.125, "lr": 1e-05, "finish_rate": 0.988, "comp_len": 667.5, "dropped_truncated": 0, "gold_loss": 0.4302, "gold_lambda": 0.5, "rep_ratio": 2.259, "t_data_s": 0.0, "t_rollout_s": 50.9, "t_step_s": 126.6, "t_refresh_s": 0.3, "mem_gb": 10.41, "mem_gb_teacher": 20.55}
42
+ {"step": 22, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.9240483859857905, "tokens": 186012, "cumulative_loss_tokens": 5316239, "grad_norm": 3.453125, "lr": 1e-05, "finish_rate": 0.973, "comp_len": 726.6, "dropped_truncated": 0, "gold_loss": 0.4496, "gold_lambda": 0.5, "rep_ratio": 2.576, "t_data_s": 0.0, "t_rollout_s": 54.9, "t_step_s": 133.4, "t_refresh_s": 0.3, "mem_gb": 10.37, "mem_gb_teacher": 20.58}
43
+ {"step": 23, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.9889739579658967, "tokens": 161101, "cumulative_loss_tokens": 5477340, "grad_norm": 3.6875, "lr": 1e-05, "finish_rate": 0.973, "comp_len": 629.3, "dropped_truncated": 0, "gold_loss": 1.1295, "gold_lambda": 0.5, "rep_ratio": 2.562, "t_data_s": 0.0, "t_rollout_s": 49.2, "t_step_s": 126.1, "t_refresh_s": 0.3, "mem_gb": 10.19, "mem_gb_teacher": 20.54}
44
+ {"step": 24, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.9001673450655785, "tokens": 172167, "cumulative_loss_tokens": 5649507, "grad_norm": 2.84375, "lr": 1e-05, "finish_rate": 0.98, "comp_len": 672.5, "dropped_truncated": 0, "gold_loss": 1.184, "gold_lambda": 0.5, "rep_ratio": 2.392, "t_data_s": 0.0, "t_rollout_s": 51.2, "t_step_s": 125.6, "t_refresh_s": 0.3, "mem_gb": 10.42, "mem_gb_teacher": 20.61}
45
+ {"step": 25, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.8294563550778241, "tokens": 193603, "cumulative_loss_tokens": 5843110, "grad_norm": 3.203125, "lr": 1e-05, "finish_rate": 0.949, "comp_len": 756.3, "dropped_truncated": 0, "gold_loss": 2.1916, "gold_lambda": 0.5, "rep_ratio": 2.737, "t_data_s": 0.0, "t_rollout_s": 57.4, "t_step_s": 137.7, "t_refresh_s": 0.3, "mem_gb": 10.55, "mem_gb_teacher": 20.67}
46
+ {"step": 26, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.8520757204397251, "tokens": 184343, "cumulative_loss_tokens": 6027453, "grad_norm": 2.6875, "lr": 1e-05, "finish_rate": 0.969, "comp_len": 720.1, "dropped_truncated": 0, "gold_loss": 1.8995, "gold_lambda": 0.5, "rep_ratio": 2.86, "t_data_s": 0.0, "t_rollout_s": 54.2, "t_step_s": 131.3, "t_refresh_s": 0.3, "mem_gb": 10.35, "mem_gb_teacher": 20.59}
47
+ {"step": 27, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.9107435231664042, "tokens": 190215, "cumulative_loss_tokens": 6217668, "grad_norm": 2.625, "lr": 1e-05, "finish_rate": 0.953, "comp_len": 743.0, "dropped_truncated": 0, "gold_loss": 1.4011, "gold_lambda": 0.5, "rep_ratio": 2.559, "t_data_s": 0.0, "t_rollout_s": 55.9, "t_step_s": 134.6, "t_refresh_s": 0.3, "mem_gb": 10.41, "mem_gb_teacher": 20.62}
48
+ {"step": 28, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.9077256934040764, "tokens": 178659, "cumulative_loss_tokens": 6396327, "grad_norm": 2.1875, "lr": 1e-05, "finish_rate": 0.977, "comp_len": 697.9, "dropped_truncated": 0, "gold_loss": 1.4574, "gold_lambda": 0.5, "rep_ratio": 2.687, "t_data_s": 0.0, "t_rollout_s": 51.5, "t_step_s": 126.5, "t_refresh_s": 0.3, "mem_gb": 10.37, "mem_gb_teacher": 20.54}
49
+ {"step": 29, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.9122361920726376, "tokens": 179195, "cumulative_loss_tokens": 6575522, "grad_norm": 1.7109375, "lr": 1e-05, "finish_rate": 0.949, "comp_len": 700.0, "dropped_truncated": 0, "gold_loss": 0.4425, "gold_lambda": 0.5, "rep_ratio": 2.463, "t_data_s": 0.0, "t_rollout_s": 54.0, "t_step_s": 131.5, "t_refresh_s": 0.3, "mem_gb": 10.38, "mem_gb_teacher": 20.6}
50
+ {"step": 30, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.8639218128417365, "tokens": 196883, "cumulative_loss_tokens": 6772405, "grad_norm": 1.625, "lr": 1e-05, "finish_rate": 0.973, "comp_len": 769.1, "dropped_truncated": 0, "gold_loss": 0.4074, "gold_lambda": 0.5, "rep_ratio": 2.545, "t_data_s": 0.0, "t_rollout_s": 57.7, "t_step_s": 138.1, "t_refresh_s": 0.3, "mem_gb": 10.48, "mem_gb_teacher": 20.64}
51
+ [eval step 30] sample: 'To solve the problem of determining the number of days in a month that has a positive count, we need to analyze the data provided and identify the pattern.\n\n**Step 1: Analyze the Data**\n\nThe data prov'
52
+ {"step": 31, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.8510895792700558, "tokens": 203228, "cumulative_loss_tokens": 6975633, "grad_norm": 2.484375, "lr": 1e-05, "finish_rate": 0.98, "comp_len": 793.9, "dropped_truncated": 0, "gold_loss": 1.7625, "gold_lambda": 0.5, "rep_ratio": 2.537, "t_data_s": 0.0, "t_rollout_s": 59.8, "t_step_s": 139.0, "t_refresh_s": 0.3, "mem_gb": 10.1, "mem_gb_teacher": 20.51}
53
+ {"step": 32, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.799281217758318, "tokens": 186923, "cumulative_loss_tokens": 7162556, "grad_norm": 2.125, "lr": 1e-05, "finish_rate": 0.969, "comp_len": 730.2, "dropped_truncated": 0, "gold_loss": 1.4991, "gold_lambda": 0.5, "rep_ratio": 2.655, "t_data_s": 0.0, "t_rollout_s": 55.1, "t_step_s": 134.5, "t_refresh_s": 0.3, "mem_gb": 10.43, "mem_gb_teacher": 20.6}
54
+ {"step": 33, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.9108344851361305, "tokens": 206685, "cumulative_loss_tokens": 7369241, "grad_norm": 2.609375, "lr": 1e-05, "finish_rate": 0.941, "comp_len": 807.4, "dropped_truncated": 0, "gold_loss": 1.8247, "gold_lambda": 0.5, "rep_ratio": 2.723, "t_data_s": 0.0, "t_rollout_s": 61.8, "t_step_s": 145.9, "t_refresh_s": 0.3, "mem_gb": 10.34, "mem_gb_teacher": 20.6}
55
+ {"step": 34, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.8392004683972054, "tokens": 200636, "cumulative_loss_tokens": 7569877, "grad_norm": 5.21875, "lr": 1e-05, "finish_rate": 0.953, "comp_len": 783.7, "dropped_truncated": 0, "gold_loss": 2.3876, "gold_lambda": 0.5, "rep_ratio": 2.624, "t_data_s": 0.0, "t_rollout_s": 59.5, "t_step_s": 140.6, "t_refresh_s": 0.3, "mem_gb": 10.38, "mem_gb_teacher": 20.56}
56
+ {"step": 35, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.8846296758133513, "tokens": 180440, "cumulative_loss_tokens": 7750317, "grad_norm": 2.015625, "lr": 1e-05, "finish_rate": 0.98, "comp_len": 704.8, "dropped_truncated": 0, "gold_loss": 0.3641, "gold_lambda": 0.5, "rep_ratio": 2.604, "t_data_s": 0.0, "t_rollout_s": 51.9, "t_step_s": 127.8, "t_refresh_s": 0.3, "mem_gb": 10.24, "mem_gb_teacher": 20.53}
57
+ {"step": 36, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.8721849152298049, "tokens": 194494, "cumulative_loss_tokens": 7944811, "grad_norm": 1.859375, "lr": 1e-05, "finish_rate": 0.957, "comp_len": 759.7, "dropped_truncated": 0, "gold_loss": 0.4256, "gold_lambda": 0.5, "rep_ratio": 2.591, "t_data_s": 0.0, "t_rollout_s": 56.4, "t_step_s": 136.2, "t_refresh_s": 0.3, "mem_gb": 10.27, "mem_gb_teacher": 20.57}
58
+ {"step": 37, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.849565157371755, "tokens": 186626, "cumulative_loss_tokens": 8131437, "grad_norm": 2.390625, "lr": 1e-05, "finish_rate": 0.977, "comp_len": 729.0, "dropped_truncated": 0, "gold_loss": 1.6035, "gold_lambda": 0.5, "rep_ratio": 2.528, "t_data_s": 0.0, "t_rollout_s": 54.0, "t_step_s": 132.0, "t_refresh_s": 0.3, "mem_gb": 10.26, "mem_gb_teacher": 20.53}
59
+ {"step": 38, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.7732366005655155, "tokens": 210982, "cumulative_loss_tokens": 8342419, "grad_norm": 4.96875, "lr": 1e-05, "finish_rate": 0.969, "comp_len": 824.1, "dropped_truncated": 0, "gold_loss": 1.5962, "gold_lambda": 0.5, "rep_ratio": 2.4, "t_data_s": 0.0, "t_rollout_s": 63.0, "t_step_s": 146.4, "t_refresh_s": 0.3, "mem_gb": 10.31, "mem_gb_teacher": 20.56}
healed/opd_warm_fixed.log ADDED
@@ -0,0 +1,186 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ wandb: Tracking run with wandb version 0.28.0
2
+ wandb: W&B syncing is set to `offline` in this directory. Run `wandb online` or set WANDB_MODE=online to enable cloud syncing.
3
+ wandb: Run data is saved locally in outputs/healed/opd_warm_fixed_keep50/wandb/offline-run-20260802_065345-ubll0rd3
4
+ wandb: View this run in the terminal with `wandb leet`
5
+
6
+
7
+ reference anchor loaded from outputs/healed/keep50_warmup_fixed_s1224/step0150 on cuda:0 (beta=0.05)
8
+
9
+ starting vllm rollout server on GPU 2 (port 8377) ...
10
+ vllm server healthy in 36s
11
+ mixture distillation: 58360 gold trajectories from outputs/teacher_trajectories/dolci_combined_top128, lambda=0.5 decay=0.0
12
+
13
+ The attention mask is not set and cannot be inferred from input because pad token is same as eos token. As a consequence, you may observe unexpected behavior. Please pass your input's `attention_mask` to obtain reliable results.
14
+ 168191 prompts | 2627 steps/epoch | 120 total steps | student params 3.70B | teacher overlap=True
15
+ {"step": 1, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.44963827088755237, "tokens": 194827, "cumulative_loss_tokens": 194827, "grad_norm": 2.0625, "lr": 2.0000000000000003e-06, "finish_rate": 0.957, "comp_len": 761.0, "dropped_truncated": 0, "gold_loss": 0.2115, "gold_lambda": 0.5, "rep_ratio": 2.401, "t_data_s": 0.0, "t_rollout_s": 56.4, "t_step_s": 182.8, "t_refresh_s": 0.3, "mem_gb": 17.24, "mem_gb_teacher": 20.26}
16
+ [eval step 1] sample: 'Misy fitsapana manokana ve mba hanombanana ny fahaiza-mamorona?'
17
+ {"step": 1, "gsm8k_n": 256, "gsm8k_quick_chat": 0.65234375, "t_eval_s": 23.8}
18
+ {"step": 2, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3940014742786295, "tokens": 187987, "cumulative_loss_tokens": 382814, "grad_norm": 2.0, "lr": 3e-06, "finish_rate": 0.992, "comp_len": 734.3, "dropped_truncated": 0, "gold_loss": 0.2068, "gold_lambda": 0.5, "rep_ratio": 2.369, "t_data_s": 0.0, "t_rollout_s": 56.2, "t_step_s": 157.9, "t_refresh_s": 0.3, "mem_gb": 17.22, "mem_gb_teacher": 20.24}
19
+ {"step": 3, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3842378264528005, "tokens": 175719, "cumulative_loss_tokens": 558533, "grad_norm": 1.9296875, "lr": 4.000000000000001e-06, "finish_rate": 0.969, "comp_len": 686.4, "dropped_truncated": 0, "gold_loss": 0.2386, "gold_lambda": 0.5, "rep_ratio": 2.451, "t_data_s": 0.0, "t_rollout_s": 52.3, "t_step_s": 150.2, "t_refresh_s": 0.3, "mem_gb": 17.33, "mem_gb_teacher": 20.27}
20
+ {"step": 4, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3169176116394187, "tokens": 183284, "cumulative_loss_tokens": 741817, "grad_norm": 1.703125, "lr": 5e-06, "finish_rate": 0.973, "comp_len": 716.0, "dropped_truncated": 0, "gold_loss": 0.2391, "gold_lambda": 0.5, "rep_ratio": 2.682, "t_data_s": 0.0, "t_rollout_s": 53.4, "t_step_s": 156.8, "t_refresh_s": 0.3, "mem_gb": 17.1, "mem_gb_teacher": 20.22}
21
+ {"step": 5, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.31469852179097824, "tokens": 169650, "cumulative_loss_tokens": 911467, "grad_norm": 2.625, "lr": 6e-06, "finish_rate": 0.988, "comp_len": 662.7, "dropped_truncated": 0, "gold_loss": 1.3211, "gold_lambda": 0.5, "rep_ratio": 2.526, "t_data_s": 0.0, "t_rollout_s": 50.1, "t_step_s": 144.7, "t_refresh_s": 0.3, "mem_gb": 17.17, "mem_gb_teacher": 20.24}
22
+ {"step": 6, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3012831195711605, "tokens": 182143, "cumulative_loss_tokens": 1093610, "grad_norm": 2.203125, "lr": 7e-06, "finish_rate": 0.949, "comp_len": 711.5, "dropped_truncated": 0, "gold_loss": 0.9775, "gold_lambda": 0.5, "rep_ratio": 2.403, "t_data_s": 0.0, "t_rollout_s": 56.4, "t_step_s": 154.9, "t_refresh_s": 0.3, "mem_gb": 17.26, "mem_gb_teacher": 20.26}
23
+ {"step": 7, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.2884359588736265, "tokens": 182618, "cumulative_loss_tokens": 1276228, "grad_norm": 2.640625, "lr": 8.000000000000001e-06, "finish_rate": 0.977, "comp_len": 713.4, "dropped_truncated": 0, "gold_loss": 1.3168, "gold_lambda": 0.5, "rep_ratio": 2.266, "t_data_s": 0.0, "t_rollout_s": 53.8, "t_step_s": 154.1, "t_refresh_s": 0.3, "mem_gb": 17.47, "mem_gb_teacher": 20.28}
24
+ {"step": 8, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.25864222868998754, "tokens": 182352, "cumulative_loss_tokens": 1458580, "grad_norm": 2.9375, "lr": 9e-06, "finish_rate": 0.969, "comp_len": 712.3, "dropped_truncated": 0, "gold_loss": 1.3305, "gold_lambda": 0.5, "rep_ratio": 2.607, "t_data_s": 0.0, "t_rollout_s": 53.3, "t_step_s": 153.4, "t_refresh_s": 0.3, "mem_gb": 17.53, "mem_gb_teacher": 20.29}
25
+ {"step": 9, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.4240309239163299, "tokens": 161654, "cumulative_loss_tokens": 1620234, "grad_norm": 1.8046875, "lr": 1e-05, "finish_rate": 0.98, "comp_len": 631.5, "dropped_truncated": 0, "gold_loss": 0.2139, "gold_lambda": 0.5, "rep_ratio": 2.185, "t_data_s": 0.0, "t_rollout_s": 49.1, "t_step_s": 143.8, "t_refresh_s": 0.3, "mem_gb": 17.52, "mem_gb_teacher": 20.28}
26
+ {"step": 10, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3422506856649129, "tokens": 182466, "cumulative_loss_tokens": 1802700, "grad_norm": 1.4453125, "lr": 1e-05, "finish_rate": 0.973, "comp_len": 712.8, "dropped_truncated": 0, "gold_loss": 0.2024, "gold_lambda": 0.5, "rep_ratio": 2.523, "t_data_s": 0.0, "t_rollout_s": 53.5, "t_step_s": 153.2, "t_refresh_s": 0.3, "mem_gb": 17.34, "mem_gb_teacher": 20.25}
27
+ [eval step 10] sample: 'Misy fitsapana manokana ve mba hanombanana ny fahaiza-mamorona?'
28
+ {"step": 11, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3194563395594054, "tokens": 162668, "cumulative_loss_tokens": 1965368, "grad_norm": 2.875, "lr": 1e-05, "finish_rate": 0.996, "comp_len": 635.4, "dropped_truncated": 0, "gold_loss": 1.4128, "gold_lambda": 0.5, "rep_ratio": 2.353, "t_data_s": 0.0, "t_rollout_s": 49.0, "t_step_s": 144.6, "t_refresh_s": 0.3, "mem_gb": 17.31, "mem_gb_teacher": 20.25}
29
+ {"step": 12, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.30020437081658174, "tokens": 172670, "cumulative_loss_tokens": 2138038, "grad_norm": 2.5, "lr": 1e-05, "finish_rate": 0.984, "comp_len": 674.5, "dropped_truncated": 0, "gold_loss": 1.3403, "gold_lambda": 0.5, "rep_ratio": 2.41, "t_data_s": 0.0, "t_rollout_s": 52.5, "t_step_s": 152.4, "t_refresh_s": 0.3, "mem_gb": 17.45, "mem_gb_teacher": 20.3}
30
+ {"step": 13, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.4090317845232767, "tokens": 160408, "cumulative_loss_tokens": 2298446, "grad_norm": 3.203125, "lr": 1e-05, "finish_rate": 0.992, "comp_len": 626.6, "dropped_truncated": 0, "gold_loss": 1.4213, "gold_lambda": 0.5, "rep_ratio": 2.23, "t_data_s": 0.0, "t_rollout_s": 47.7, "t_step_s": 141.4, "t_refresh_s": 0.3, "mem_gb": 17.07, "mem_gb_teacher": 20.25}
31
+ {"step": 14, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.34899632673731384, "tokens": 159410, "cumulative_loss_tokens": 2457856, "grad_norm": 2.6875, "lr": 1e-05, "finish_rate": 0.98, "comp_len": 622.7, "dropped_truncated": 0, "gold_loss": 1.3901, "gold_lambda": 0.5, "rep_ratio": 2.422, "t_data_s": 0.0, "t_rollout_s": 47.6, "t_step_s": 141.7, "t_refresh_s": 0.3, "mem_gb": 17.12, "mem_gb_teacher": 20.25}
32
+ {"step": 15, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.34543543075337274, "tokens": 166862, "cumulative_loss_tokens": 2624718, "grad_norm": 1.15625, "lr": 1e-05, "finish_rate": 0.98, "comp_len": 651.8, "dropped_truncated": 0, "gold_loss": 0.2286, "gold_lambda": 0.5, "rep_ratio": 2.459, "t_data_s": 0.0, "t_rollout_s": 48.7, "t_step_s": 145.8, "t_refresh_s": 0.3, "mem_gb": 17.25, "mem_gb_teacher": 20.24}
33
+ {"step": 16, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3638701808236742, "tokens": 154124, "cumulative_loss_tokens": 2778842, "grad_norm": 1.2109375, "lr": 1e-05, "finish_rate": 0.988, "comp_len": 602.0, "dropped_truncated": 0, "gold_loss": 0.2253, "gold_lambda": 0.5, "rep_ratio": 2.369, "t_data_s": 0.0, "t_rollout_s": 47.2, "t_step_s": 141.5, "t_refresh_s": 0.3, "mem_gb": 17.09, "mem_gb_teacher": 20.23}
34
+ {"step": 17, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3916856637959853, "tokens": 153093, "cumulative_loss_tokens": 2931935, "grad_norm": 2.0625, "lr": 1e-05, "finish_rate": 0.992, "comp_len": 598.0, "dropped_truncated": 0, "gold_loss": 1.0651, "gold_lambda": 0.5, "rep_ratio": 2.331, "t_data_s": 0.0, "t_rollout_s": 47.5, "t_step_s": 149.3, "t_refresh_s": 0.3, "mem_gb": 17.26, "mem_gb_teacher": 20.27}
35
+ {"step": 18, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3789892320845085, "tokens": 167800, "cumulative_loss_tokens": 3099735, "grad_norm": 1.96875, "lr": 1e-05, "finish_rate": 0.984, "comp_len": 655.5, "dropped_truncated": 0, "gold_loss": 1.0761, "gold_lambda": 0.5, "rep_ratio": 2.334, "t_data_s": 0.0, "t_rollout_s": 52.0, "t_step_s": 153.1, "t_refresh_s": 0.3, "mem_gb": 17.18, "mem_gb_teacher": 20.24}
36
+ {"step": 19, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.35074403341776794, "tokens": 167173, "cumulative_loss_tokens": 3266908, "grad_norm": 2.390625, "lr": 1e-05, "finish_rate": 0.988, "comp_len": 653.0, "dropped_truncated": 0, "gold_loss": 1.3153, "gold_lambda": 0.5, "rep_ratio": 2.337, "t_data_s": 0.0, "t_rollout_s": 50.2, "t_step_s": 143.5, "t_refresh_s": 0.3, "mem_gb": 17.32, "mem_gb_teacher": 20.25}
37
+ {"step": 20, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.2626116644125829, "tokens": 190018, "cumulative_loss_tokens": 3456926, "grad_norm": 2.375, "lr": 1e-05, "finish_rate": 0.969, "comp_len": 742.3, "dropped_truncated": 0, "gold_loss": 1.4092, "gold_lambda": 0.5, "rep_ratio": 2.615, "t_data_s": 0.0, "t_rollout_s": 55.8, "t_step_s": 158.7, "t_refresh_s": 0.3, "mem_gb": 17.61, "mem_gb_teacher": 20.29}
38
+ [eval step 20] sample: 'Misy fitapana manokana ny fahaiza-mamorona.'
39
+ {"step": 20, "gsm8k_n": 256, "gsm8k_quick_chat": 0.6171875, "t_eval_s": 21.4}
40
+ {"step": 21, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.36320468789413923, "tokens": 161040, "cumulative_loss_tokens": 3617966, "grad_norm": 1.4921875, "lr": 1e-05, "finish_rate": 0.977, "comp_len": 629.1, "dropped_truncated": 0, "gold_loss": 0.2034, "gold_lambda": 0.5, "rep_ratio": 2.491, "t_data_s": 0.0, "t_rollout_s": 47.5, "t_step_s": 141.8, "t_refresh_s": 0.3, "mem_gb": 17.31, "mem_gb_teacher": 20.25}
41
+ {"step": 22, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3560513313296494, "tokens": 186196, "cumulative_loss_tokens": 3804162, "grad_norm": 1.140625, "lr": 1e-05, "finish_rate": 0.973, "comp_len": 727.3, "dropped_truncated": 0, "gold_loss": 0.2012, "gold_lambda": 0.5, "rep_ratio": 2.621, "t_data_s": 0.0, "t_rollout_s": 56.3, "t_step_s": 157.9, "t_refresh_s": 0.3, "mem_gb": 17.48, "mem_gb_teacher": 20.28}
42
+ {"step": 23, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.4002737975893586, "tokens": 172291, "cumulative_loss_tokens": 3976453, "grad_norm": 1.5546875, "lr": 1e-05, "finish_rate": 0.973, "comp_len": 673.0, "dropped_truncated": 0, "gold_loss": 0.7825, "gold_lambda": 0.5, "rep_ratio": 2.324, "t_data_s": 0.0, "t_rollout_s": 50.2, "t_step_s": 146.6, "t_refresh_s": 0.3, "mem_gb": 17.11, "mem_gb_teacher": 20.25}
43
+ {"step": 24, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3151660050172998, "tokens": 164852, "cumulative_loss_tokens": 4141305, "grad_norm": 1.6875, "lr": 1e-05, "finish_rate": 0.984, "comp_len": 644.0, "dropped_truncated": 0, "gold_loss": 0.801, "gold_lambda": 0.5, "rep_ratio": 2.331, "t_data_s": 0.0, "t_rollout_s": 49.6, "t_step_s": 143.3, "t_refresh_s": 0.3, "mem_gb": 17.59, "mem_gb_teacher": 20.31}
44
+ {"step": 25, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.313829974284698, "tokens": 196304, "cumulative_loss_tokens": 4337609, "grad_norm": 2.234375, "lr": 1e-05, "finish_rate": 0.965, "comp_len": 766.8, "dropped_truncated": 0, "gold_loss": 1.6377, "gold_lambda": 0.5, "rep_ratio": 2.688, "t_data_s": 0.0, "t_rollout_s": 57.5, "t_step_s": 156.7, "t_refresh_s": 0.3, "mem_gb": 17.76, "mem_gb_teacher": 20.32}
45
+ {"step": 26, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.27340404820149083, "tokens": 174461, "cumulative_loss_tokens": 4512070, "grad_norm": 1.9375, "lr": 1e-05, "finish_rate": 0.977, "comp_len": 681.5, "dropped_truncated": 0, "gold_loss": 1.3934, "gold_lambda": 0.5, "rep_ratio": 2.549, "t_data_s": 0.0, "t_rollout_s": 52.1, "t_step_s": 146.5, "t_refresh_s": 0.3, "mem_gb": 17.47, "mem_gb_teacher": 20.28}
46
+ {"step": 27, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3913407735489407, "tokens": 181479, "cumulative_loss_tokens": 4693549, "grad_norm": 1.8671875, "lr": 1e-05, "finish_rate": 0.973, "comp_len": 708.9, "dropped_truncated": 0, "gold_loss": 1.0072, "gold_lambda": 0.5, "rep_ratio": 2.356, "t_data_s": 0.0, "t_rollout_s": 52.5, "t_step_s": 146.6, "t_refresh_s": 0.3, "mem_gb": 17.33, "mem_gb_teacher": 20.26}
47
+ {"step": 28, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3790620736299349, "tokens": 177593, "cumulative_loss_tokens": 4871142, "grad_norm": 1.6015625, "lr": 1e-05, "finish_rate": 0.98, "comp_len": 693.7, "dropped_truncated": 0, "gold_loss": 1.04, "gold_lambda": 0.5, "rep_ratio": 2.369, "t_data_s": 0.0, "t_rollout_s": 51.6, "t_step_s": 147.2, "t_refresh_s": 0.3, "mem_gb": 17.29, "mem_gb_teacher": 20.25}
48
+ {"step": 29, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3367005885269224, "tokens": 169115, "cumulative_loss_tokens": 5040257, "grad_norm": 1.34375, "lr": 1e-05, "finish_rate": 0.957, "comp_len": 660.6, "dropped_truncated": 0, "gold_loss": 0.2224, "gold_lambda": 0.5, "rep_ratio": 2.308, "t_data_s": 0.0, "t_rollout_s": 52.3, "t_step_s": 150.0, "t_refresh_s": 0.3, "mem_gb": 17.5, "mem_gb_teacher": 20.28}
49
+ {"step": 30, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3335959742327791, "tokens": 192932, "cumulative_loss_tokens": 5233189, "grad_norm": 1.765625, "lr": 1e-05, "finish_rate": 0.961, "comp_len": 753.6, "dropped_truncated": 0, "gold_loss": 0.1972, "gold_lambda": 0.5, "rep_ratio": 2.476, "t_data_s": 0.0, "t_rollout_s": 57.4, "t_step_s": 161.6, "t_refresh_s": 0.3, "mem_gb": 17.65, "mem_gb_teacher": 20.3}
50
+ [eval step 30] sample: 'Misy fitapana manokana ny fahaiza-mamorona.'
51
+ {"step": 31, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3171257981787659, "tokens": 191095, "cumulative_loss_tokens": 5424284, "grad_norm": 1.8046875, "lr": 1e-05, "finish_rate": 0.969, "comp_len": 746.5, "dropped_truncated": 0, "gold_loss": 1.2779, "gold_lambda": 0.5, "rep_ratio": 2.497, "t_data_s": 0.0, "t_rollout_s": 56.9, "t_step_s": 156.9, "t_refresh_s": 0.3, "mem_gb": 17.18, "mem_gb_teacher": 20.26}
52
+ {"step": 32, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3311199295843264, "tokens": 187659, "cumulative_loss_tokens": 5611943, "grad_norm": 1.6171875, "lr": 1e-05, "finish_rate": 0.973, "comp_len": 733.0, "dropped_truncated": 0, "gold_loss": 1.0749, "gold_lambda": 0.5, "rep_ratio": 2.448, "t_data_s": 0.0, "t_rollout_s": 56.4, "t_step_s": 161.5, "t_refresh_s": 0.3, "mem_gb": 17.37, "mem_gb_teacher": 20.29}
53
+ {"step": 33, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3653596817669372, "tokens": 162176, "cumulative_loss_tokens": 5774119, "grad_norm": 1.96875, "lr": 1e-05, "finish_rate": 0.977, "comp_len": 633.5, "dropped_truncated": 0, "gold_loss": 1.3587, "gold_lambda": 0.5, "rep_ratio": 2.488, "t_data_s": 0.0, "t_rollout_s": 50.0, "t_step_s": 147.9, "t_refresh_s": 0.3, "mem_gb": 17.77, "mem_gb_teacher": 20.31}
54
+ {"step": 34, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3303874825869901, "tokens": 175637, "cumulative_loss_tokens": 5949756, "grad_norm": 2.125, "lr": 1e-05, "finish_rate": 0.977, "comp_len": 686.1, "dropped_truncated": 0, "gold_loss": 1.6698, "gold_lambda": 0.5, "rep_ratio": 2.444, "t_data_s": 0.0, "t_rollout_s": 53.0, "t_step_s": 150.9, "t_refresh_s": 0.3, "mem_gb": 17.3, "mem_gb_teacher": 20.27}
55
+ {"step": 35, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.338662115701657, "tokens": 164097, "cumulative_loss_tokens": 6113853, "grad_norm": 1.2578125, "lr": 1e-05, "finish_rate": 0.984, "comp_len": 641.0, "dropped_truncated": 0, "gold_loss": 0.1611, "gold_lambda": 0.5, "rep_ratio": 2.39, "t_data_s": 0.0, "t_rollout_s": 48.7, "t_step_s": 140.7, "t_refresh_s": 0.3, "mem_gb": 17.31, "mem_gb_teacher": 20.25}
56
+ {"step": 36, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3760627886982097, "tokens": 180353, "cumulative_loss_tokens": 6294206, "grad_norm": 1.171875, "lr": 1e-05, "finish_rate": 0.949, "comp_len": 704.5, "dropped_truncated": 0, "gold_loss": 0.1814, "gold_lambda": 0.5, "rep_ratio": 2.598, "t_data_s": 0.0, "t_rollout_s": 54.3, "t_step_s": 154.0, "t_refresh_s": 0.3, "mem_gb": 17.34, "mem_gb_teacher": 20.27}
57
+ {"step": 37, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3777555677864465, "tokens": 179099, "cumulative_loss_tokens": 6473305, "grad_norm": 1.734375, "lr": 1e-05, "finish_rate": 0.988, "comp_len": 699.6, "dropped_truncated": 0, "gold_loss": 1.1694, "gold_lambda": 0.5, "rep_ratio": 2.291, "t_data_s": 0.0, "t_rollout_s": 52.4, "t_step_s": 147.9, "t_refresh_s": 0.3, "mem_gb": 17.28, "mem_gb_teacher": 20.25}
58
+ {"step": 38, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.2655141159261545, "tokens": 183837, "cumulative_loss_tokens": 6657142, "grad_norm": 1.7890625, "lr": 1e-05, "finish_rate": 0.973, "comp_len": 718.1, "dropped_truncated": 0, "gold_loss": 1.0682, "gold_lambda": 0.5, "rep_ratio": 2.283, "t_data_s": 0.0, "t_rollout_s": 52.4, "t_step_s": 149.0, "t_refresh_s": 0.3, "mem_gb": 16.97, "mem_gb_teacher": 20.22}
59
+ {"step": 39, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.26126282006497425, "tokens": 181345, "cumulative_loss_tokens": 6838487, "grad_norm": 1.140625, "lr": 1e-05, "finish_rate": 0.969, "comp_len": 708.4, "dropped_truncated": 0, "gold_loss": 0.1805, "gold_lambda": 0.5, "rep_ratio": 2.463, "t_data_s": 0.0, "t_rollout_s": 53.9, "t_step_s": 152.2, "t_refresh_s": 0.3, "mem_gb": 17.13, "mem_gb_teacher": 20.23}
60
+ {"step": 40, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.4208335484820406, "tokens": 166853, "cumulative_loss_tokens": 7005340, "grad_norm": 1.3125, "lr": 1e-05, "finish_rate": 0.977, "comp_len": 651.8, "dropped_truncated": 0, "gold_loss": 0.1644, "gold_lambda": 0.5, "rep_ratio": 2.677, "t_data_s": 0.0, "t_rollout_s": 48.8, "t_step_s": 140.2, "t_refresh_s": 0.3, "mem_gb": 17.35, "mem_gb_teacher": 20.29}
61
+ [eval step 40] sample: 'Masy fitapana manokana ve mba hanombanana ny fahaiza-mamorona?\n\nMasy fitapana (Masy) is a traditional Indian dish made from a combination of spices, including flava, a Indian seasoning, and pananas, w'
62
+ {"step": 40, "gsm8k_n": 256, "gsm8k_quick_chat": 0.62890625, "t_eval_s": 21.6}
63
+ {"step": 41, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.33376682261101664, "tokens": 161263, "cumulative_loss_tokens": 7166603, "grad_norm": 1.765625, "lr": 1e-05, "finish_rate": 0.98, "comp_len": 629.9, "dropped_truncated": 0, "gold_loss": 1.0706, "gold_lambda": 0.5, "rep_ratio": 2.454, "t_data_s": 0.0, "t_rollout_s": 47.2, "t_step_s": 136.0, "t_refresh_s": 0.3, "mem_gb": 17.22, "mem_gb_teacher": 20.26}
64
+ {"step": 42, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.34152158943635763, "tokens": 171806, "cumulative_loss_tokens": 7338409, "grad_norm": 1.8125, "lr": 1e-05, "finish_rate": 0.992, "comp_len": 671.1, "dropped_truncated": 0, "gold_loss": 1.0227, "gold_lambda": 0.5, "rep_ratio": 2.453, "t_data_s": 0.0, "t_rollout_s": 50.1, "t_step_s": 141.3, "t_refresh_s": 0.3, "mem_gb": 17.12, "mem_gb_teacher": 20.23}
65
+ {"step": 43, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.30709751060902213, "tokens": 173000, "cumulative_loss_tokens": 7511409, "grad_norm": 1.765625, "lr": 1e-05, "finish_rate": 0.973, "comp_len": 675.8, "dropped_truncated": 0, "gold_loss": 1.2339, "gold_lambda": 0.5, "rep_ratio": 2.397, "t_data_s": 0.0, "t_rollout_s": 51.4, "t_step_s": 144.4, "t_refresh_s": 0.3, "mem_gb": 17.07, "mem_gb_teacher": 20.24}
66
+ {"step": 44, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.38481537799269605, "tokens": 160690, "cumulative_loss_tokens": 7672099, "grad_norm": 2.046875, "lr": 1e-05, "finish_rate": 0.98, "comp_len": 627.7, "dropped_truncated": 0, "gold_loss": 1.3762, "gold_lambda": 0.5, "rep_ratio": 2.414, "t_data_s": 0.0, "t_rollout_s": 47.9, "t_step_s": 138.1, "t_refresh_s": 0.3, "mem_gb": 17.33, "mem_gb_teacher": 20.25}
67
+ {"step": 45, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3474364483801316, "tokens": 163851, "cumulative_loss_tokens": 7835950, "grad_norm": 1.609375, "lr": 1e-05, "finish_rate": 0.973, "comp_len": 640.0, "dropped_truncated": 0, "gold_loss": 1.1411, "gold_lambda": 0.5, "rep_ratio": 2.204, "t_data_s": 0.0, "t_rollout_s": 49.1, "t_step_s": 143.4, "t_refresh_s": 0.3, "mem_gb": 17.42, "mem_gb_teacher": 20.27}
68
+ {"step": 46, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.2480004573858833, "tokens": 174070, "cumulative_loss_tokens": 8010020, "grad_norm": 1.796875, "lr": 1e-05, "finish_rate": 0.973, "comp_len": 680.0, "dropped_truncated": 0, "gold_loss": 1.1273, "gold_lambda": 0.5, "rep_ratio": 2.398, "t_data_s": 0.0, "t_rollout_s": 50.5, "t_step_s": 144.5, "t_refresh_s": 0.3, "mem_gb": 17.11, "mem_gb_teacher": 20.24}
69
+ {"step": 47, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3238008658347637, "tokens": 175419, "cumulative_loss_tokens": 8185439, "grad_norm": 1.15625, "lr": 1e-05, "finish_rate": 0.965, "comp_len": 685.2, "dropped_truncated": 0, "gold_loss": 0.2272, "gold_lambda": 0.5, "rep_ratio": 2.395, "t_data_s": 0.0, "t_rollout_s": 52.7, "t_step_s": 151.6, "t_refresh_s": 0.3, "mem_gb": 17.06, "mem_gb_teacher": 20.24}
70
+ {"step": 48, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.26486295279158406, "tokens": 174191, "cumulative_loss_tokens": 8359630, "grad_norm": 1.0078125, "lr": 1e-05, "finish_rate": 0.965, "comp_len": 680.4, "dropped_truncated": 0, "gold_loss": 0.2168, "gold_lambda": 0.5, "rep_ratio": 2.41, "t_data_s": 0.0, "t_rollout_s": 51.6, "t_step_s": 148.9, "t_refresh_s": 0.3, "mem_gb": 17.0, "mem_gb_teacher": 20.23}
71
+ {"step": 49, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3201779053440231, "tokens": 175865, "cumulative_loss_tokens": 8535495, "grad_norm": 1.84375, "lr": 1e-05, "finish_rate": 0.988, "comp_len": 687.0, "dropped_truncated": 0, "gold_loss": 1.2798, "gold_lambda": 0.5, "rep_ratio": 2.35, "t_data_s": 0.0, "t_rollout_s": 52.2, "t_step_s": 147.4, "t_refresh_s": 0.3, "mem_gb": 17.22, "mem_gb_teacher": 20.26}
72
+ {"step": 50, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3491487042176898, "tokens": 173904, "cumulative_loss_tokens": 8709399, "grad_norm": 1.609375, "lr": 1e-05, "finish_rate": 0.988, "comp_len": 679.3, "dropped_truncated": 0, "gold_loss": 1.1223, "gold_lambda": 0.5, "rep_ratio": 2.19, "t_data_s": 0.0, "t_rollout_s": 52.0, "t_step_s": 146.4, "t_refresh_s": 0.3, "mem_gb": 17.47, "mem_gb_teacher": 20.27}
73
+ [eval step 50] sample: 'Masy fitapana manokana ve mba hanombanana ny fahaiza-mamorona?\n\nMasy fitapana (Masy) is a traditional Indian dish made from a combination of spices, including flava, a Indian seasoning, and a Indian v'
74
+ {"step": 51, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.4068168763832476, "tokens": 149202, "cumulative_loss_tokens": 8858601, "grad_norm": 1.84375, "lr": 1e-05, "finish_rate": 0.984, "comp_len": 582.8, "dropped_truncated": 0, "gold_loss": 0.9763, "gold_lambda": 0.5, "rep_ratio": 2.411, "t_data_s": 0.0, "t_rollout_s": 44.8, "t_step_s": 130.7, "t_refresh_s": 0.3, "mem_gb": 17.24, "mem_gb_teacher": 20.26}
75
+ {"step": 52, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3949036657737358, "tokens": 157801, "cumulative_loss_tokens": 9016402, "grad_norm": 1.5390625, "lr": 1e-05, "finish_rate": 0.996, "comp_len": 616.4, "dropped_truncated": 0, "gold_loss": 0.7635, "gold_lambda": 0.5, "rep_ratio": 2.429, "t_data_s": 0.0, "t_rollout_s": 48.4, "t_step_s": 137.8, "t_refresh_s": 0.3, "mem_gb": 16.94, "mem_gb_teacher": 20.22}
76
+ {"step": 53, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3542952813312953, "tokens": 173066, "cumulative_loss_tokens": 9189468, "grad_norm": 1.375, "lr": 1e-05, "finish_rate": 0.961, "comp_len": 676.0, "dropped_truncated": 0, "gold_loss": 0.888, "gold_lambda": 0.5, "rep_ratio": 2.359, "t_data_s": 0.0, "t_rollout_s": 52.3, "t_step_s": 149.8, "t_refresh_s": 0.3, "mem_gb": 17.61, "mem_gb_teacher": 20.31}
77
+ {"step": 54, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3200471835924996, "tokens": 183743, "cumulative_loss_tokens": 9373211, "grad_norm": 1.6796875, "lr": 1e-05, "finish_rate": 0.965, "comp_len": 717.7, "dropped_truncated": 0, "gold_loss": 1.0623, "gold_lambda": 0.5, "rep_ratio": 2.51, "t_data_s": 0.0, "t_rollout_s": 54.3, "t_step_s": 153.0, "t_refresh_s": 0.3, "mem_gb": 17.3, "mem_gb_teacher": 20.26}
78
+ {"step": 55, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3881143610330633, "tokens": 161627, "cumulative_loss_tokens": 9534838, "grad_norm": 2.515625, "lr": 1e-05, "finish_rate": 0.98, "comp_len": 631.4, "dropped_truncated": 0, "gold_loss": 1.6456, "gold_lambda": 0.5, "rep_ratio": 2.152, "t_data_s": 0.0, "t_rollout_s": 50.7, "t_step_s": 143.3, "t_refresh_s": 0.3, "mem_gb": 17.43, "mem_gb_teacher": 20.29}
79
+ {"step": 56, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.37364923851747267, "tokens": 168945, "cumulative_loss_tokens": 9703783, "grad_norm": 1.6953125, "lr": 1e-05, "finish_rate": 0.957, "comp_len": 659.9, "dropped_truncated": 0, "gold_loss": 1.1163, "gold_lambda": 0.5, "rep_ratio": 2.215, "t_data_s": 0.0, "t_rollout_s": 51.2, "t_step_s": 148.9, "t_refresh_s": 0.3, "mem_gb": 17.32, "mem_gb_teacher": 20.25}
80
+ {"step": 57, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.4587748102189034, "tokens": 168237, "cumulative_loss_tokens": 9872020, "grad_norm": 1.5390625, "lr": 1e-05, "finish_rate": 0.977, "comp_len": 657.2, "dropped_truncated": 0, "gold_loss": 0.9107, "gold_lambda": 0.5, "rep_ratio": 2.531, "t_data_s": 0.0, "t_rollout_s": 49.2, "t_step_s": 144.3, "t_refresh_s": 0.3, "mem_gb": 17.16, "mem_gb_teacher": 20.23}
81
+ {"step": 58, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.26425971459508646, "tokens": 184409, "cumulative_loss_tokens": 10056429, "grad_norm": 1.5, "lr": 1e-05, "finish_rate": 0.957, "comp_len": 720.3, "dropped_truncated": 0, "gold_loss": 1.125, "gold_lambda": 0.5, "rep_ratio": 2.684, "t_data_s": 0.0, "t_rollout_s": 54.7, "t_step_s": 153.4, "t_refresh_s": 0.3, "mem_gb": 17.21, "mem_gb_teacher": 20.25}
82
+ {"step": 59, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.2719602728308547, "tokens": 189369, "cumulative_loss_tokens": 10245798, "grad_norm": 1.671875, "lr": 1e-05, "finish_rate": 0.945, "comp_len": 739.7, "dropped_truncated": 0, "gold_loss": 0.9619, "gold_lambda": 0.5, "rep_ratio": 2.581, "t_data_s": 0.0, "t_rollout_s": 56.5, "t_step_s": 154.0, "t_refresh_s": 0.3, "mem_gb": 17.25, "mem_gb_teacher": 20.24}
83
+ {"step": 60, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.35498727428496857, "tokens": 166983, "cumulative_loss_tokens": 10412781, "grad_norm": 1.84375, "lr": 1e-05, "finish_rate": 0.953, "comp_len": 652.3, "dropped_truncated": 0, "gold_loss": 1.1915, "gold_lambda": 0.5, "rep_ratio": 2.391, "t_data_s": 0.0, "t_rollout_s": 49.4, "t_step_s": 142.7, "t_refresh_s": 0.3, "mem_gb": 17.2, "mem_gb_teacher": 20.25}
84
+ [eval step 60] sample: 'Misy fitapana manokana ny fahaiza-mamorona, mba hanombanana, is a term that describes a situation where a person is engaging in activities that are not conducive to their well-being or health, often d'
85
+ {"step": 60, "gsm8k_n": 256, "gsm8k_quick_chat": 0.6171875, "t_eval_s": 21.5}
86
+ {"step": 61, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.36016367042313613, "tokens": 195255, "cumulative_loss_tokens": 10608036, "grad_norm": 1.1640625, "lr": 1e-05, "finish_rate": 0.973, "comp_len": 762.7, "dropped_truncated": 0, "gold_loss": 0.2184, "gold_lambda": 0.5, "rep_ratio": 2.441, "t_data_s": 0.0, "t_rollout_s": 56.4, "t_step_s": 157.9, "t_refresh_s": 0.3, "mem_gb": 17.14, "mem_gb_teacher": 20.25}
87
+ {"step": 62, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.2672880692024946, "tokens": 183249, "cumulative_loss_tokens": 10791285, "grad_norm": 1.0625, "lr": 1e-05, "finish_rate": 0.98, "comp_len": 715.8, "dropped_truncated": 0, "gold_loss": 0.2236, "gold_lambda": 0.5, "rep_ratio": 2.47, "t_data_s": 0.0, "t_rollout_s": 54.1, "t_step_s": 151.8, "t_refresh_s": 0.3, "mem_gb": 17.37, "mem_gb_teacher": 20.26}
88
+ {"step": 63, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.272046200843007, "tokens": 169354, "cumulative_loss_tokens": 10960639, "grad_norm": 1.078125, "lr": 1e-05, "finish_rate": 0.984, "comp_len": 661.5, "dropped_truncated": 0, "gold_loss": 0.2677, "gold_lambda": 0.5, "rep_ratio": 2.399, "t_data_s": 0.0, "t_rollout_s": 50.5, "t_step_s": 146.3, "t_refresh_s": 0.3, "mem_gb": 16.99, "mem_gb_teacher": 20.22}
89
+ {"step": 64, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.36498770492525573, "tokens": 162379, "cumulative_loss_tokens": 11123018, "grad_norm": 1.3359375, "lr": 1e-05, "finish_rate": 0.984, "comp_len": 634.3, "dropped_truncated": 0, "gold_loss": 0.2643, "gold_lambda": 0.5, "rep_ratio": 2.482, "t_data_s": 0.0, "t_rollout_s": 50.7, "t_step_s": 148.6, "t_refresh_s": 0.3, "mem_gb": 17.24, "mem_gb_teacher": 20.25}
90
+ {"step": 65, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3866863870528368, "tokens": 155402, "cumulative_loss_tokens": 11278420, "grad_norm": 1.5390625, "lr": 1e-05, "finish_rate": 0.984, "comp_len": 607.0, "dropped_truncated": 0, "gold_loss": 0.8323, "gold_lambda": 0.5, "rep_ratio": 2.378, "t_data_s": 0.0, "t_rollout_s": 47.0, "t_step_s": 136.8, "t_refresh_s": 0.3, "mem_gb": 17.26, "mem_gb_teacher": 20.27}
91
+ {"step": 66, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.2787221595843288, "tokens": 183694, "cumulative_loss_tokens": 11462114, "grad_norm": 1.875, "lr": 1e-05, "finish_rate": 0.98, "comp_len": 717.6, "dropped_truncated": 0, "gold_loss": 1.1385, "gold_lambda": 0.5, "rep_ratio": 2.538, "t_data_s": 0.0, "t_rollout_s": 53.9, "t_step_s": 152.2, "t_refresh_s": 0.3, "mem_gb": 17.04, "mem_gb_teacher": 20.22}
92
+ {"step": 67, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.27035813788552, "tokens": 176040, "cumulative_loss_tokens": 11638154, "grad_norm": 1.71875, "lr": 1e-05, "finish_rate": 0.949, "comp_len": 687.7, "dropped_truncated": 0, "gold_loss": 0.9716, "gold_lambda": 0.5, "rep_ratio": 2.298, "t_data_s": 0.0, "t_rollout_s": 51.3, "t_step_s": 142.2, "t_refresh_s": 0.3, "mem_gb": 17.37, "mem_gb_teacher": 20.28}
93
+ {"step": 68, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.28064265166714214, "tokens": 163139, "cumulative_loss_tokens": 11801293, "grad_norm": 1.53125, "lr": 1e-05, "finish_rate": 0.996, "comp_len": 637.3, "dropped_truncated": 0, "gold_loss": 0.9747, "gold_lambda": 0.5, "rep_ratio": 2.236, "t_data_s": 0.0, "t_rollout_s": 48.4, "t_step_s": 137.7, "t_refresh_s": 0.3, "mem_gb": 17.12, "mem_gb_teacher": 20.22}
94
+ {"step": 69, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.26903990990510324, "tokens": 171609, "cumulative_loss_tokens": 11972902, "grad_norm": 2.0625, "lr": 1e-05, "finish_rate": 0.973, "comp_len": 670.3, "dropped_truncated": 0, "gold_loss": 1.3383, "gold_lambda": 0.5, "rep_ratio": 2.405, "t_data_s": 0.0, "t_rollout_s": 51.2, "t_step_s": 145.0, "t_refresh_s": 0.3, "mem_gb": 17.11, "mem_gb_teacher": 20.23}
95
+ {"step": 70, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3467570015864296, "tokens": 158831, "cumulative_loss_tokens": 12131733, "grad_norm": 2.0625, "lr": 1e-05, "finish_rate": 0.992, "comp_len": 620.4, "dropped_truncated": 0, "gold_loss": 1.2858, "gold_lambda": 0.5, "rep_ratio": 2.309, "t_data_s": 0.0, "t_rollout_s": 48.3, "t_step_s": 136.2, "t_refresh_s": 0.3, "mem_gb": 17.13, "mem_gb_teacher": 20.26}
96
+ [eval step 70] sample: 'Misy fitapana manokana ny fahaiza-mamorona, mba hanombanana, is a term used to describe a situation where a person is engaging in a sexual relationship with more than one person at the same time, ofte'
97
+ {"step": 71, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3942991111679687, "tokens": 131738, "cumulative_loss_tokens": 12263471, "grad_norm": 1.671875, "lr": 1e-05, "finish_rate": 0.996, "comp_len": 514.6, "dropped_truncated": 0, "gold_loss": 0.9927, "gold_lambda": 0.5, "rep_ratio": 2.479, "t_data_s": 0.0, "t_rollout_s": 42.0, "t_step_s": 123.0, "t_refresh_s": 0.3, "mem_gb": 17.01, "mem_gb_teacher": 20.22}
98
+ {"step": 72, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.2885391192831585, "tokens": 177469, "cumulative_loss_tokens": 12440940, "grad_norm": 1.71875, "lr": 1e-05, "finish_rate": 0.965, "comp_len": 693.2, "dropped_truncated": 0, "gold_loss": 1.2827, "gold_lambda": 0.5, "rep_ratio": 2.388, "t_data_s": 0.0, "t_rollout_s": 52.2, "t_step_s": 148.6, "t_refresh_s": 0.3, "mem_gb": 17.1, "mem_gb_teacher": 20.24}
99
+ {"step": 73, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3441034620278951, "tokens": 153979, "cumulative_loss_tokens": 12594919, "grad_norm": 1.1875, "lr": 1e-05, "finish_rate": 0.988, "comp_len": 601.5, "dropped_truncated": 0, "gold_loss": 0.2327, "gold_lambda": 0.5, "rep_ratio": 2.452, "t_data_s": 0.0, "t_rollout_s": 47.3, "t_step_s": 140.2, "t_refresh_s": 0.3, "mem_gb": 16.98, "mem_gb_teacher": 20.22}
100
+ {"step": 74, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.2929858381382192, "tokens": 160418, "cumulative_loss_tokens": 12755337, "grad_norm": 1.0703125, "lr": 1e-05, "finish_rate": 0.969, "comp_len": 626.6, "dropped_truncated": 0, "gold_loss": 0.2401, "gold_lambda": 0.5, "rep_ratio": 2.408, "t_data_s": 0.0, "t_rollout_s": 48.7, "t_step_s": 144.0, "t_refresh_s": 0.3, "mem_gb": 17.21, "mem_gb_teacher": 20.24}
101
+ {"step": 75, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.35599221217074245, "tokens": 160043, "cumulative_loss_tokens": 12915380, "grad_norm": 1.5703125, "lr": 1e-05, "finish_rate": 0.98, "comp_len": 625.2, "dropped_truncated": 0, "gold_loss": 1.0205, "gold_lambda": 0.5, "rep_ratio": 2.571, "t_data_s": 0.0, "t_rollout_s": 47.8, "t_step_s": 136.3, "t_refresh_s": 0.3, "mem_gb": 17.06, "mem_gb_teacher": 20.22}
102
+ {"step": 76, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.24107641524497922, "tokens": 196064, "cumulative_loss_tokens": 13111444, "grad_norm": 1.5078125, "lr": 1e-05, "finish_rate": 0.984, "comp_len": 765.9, "dropped_truncated": 0, "gold_loss": 0.945, "gold_lambda": 0.5, "rep_ratio": 2.558, "t_data_s": 0.0, "t_rollout_s": 56.1, "t_step_s": 154.5, "t_refresh_s": 0.3, "mem_gb": 17.26, "mem_gb_teacher": 20.27}
103
+ {"step": 77, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.2918642508978692, "tokens": 168701, "cumulative_loss_tokens": 13280145, "grad_norm": 1.484375, "lr": 1e-05, "finish_rate": 0.969, "comp_len": 659.0, "dropped_truncated": 0, "gold_loss": 0.9558, "gold_lambda": 0.5, "rep_ratio": 2.496, "t_data_s": 0.0, "t_rollout_s": 51.7, "t_step_s": 146.1, "t_refresh_s": 0.3, "mem_gb": 17.23, "mem_gb_teacher": 20.24}
104
+ {"step": 78, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3819451671292889, "tokens": 133293, "cumulative_loss_tokens": 13413438, "grad_norm": 1.5625, "lr": 1e-05, "finish_rate": 0.988, "comp_len": 520.7, "dropped_truncated": 0, "gold_loss": 1.0728, "gold_lambda": 0.5, "rep_ratio": 2.506, "t_data_s": 0.0, "t_rollout_s": 43.8, "t_step_s": 128.9, "t_refresh_s": 0.3, "mem_gb": 17.52, "mem_gb_teacher": 20.28}
105
+ {"step": 79, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.2689037877672512, "tokens": 178569, "cumulative_loss_tokens": 13592007, "grad_norm": 1.6953125, "lr": 1e-05, "finish_rate": 0.934, "comp_len": 697.5, "dropped_truncated": 0, "gold_loss": 1.2626, "gold_lambda": 0.5, "rep_ratio": 2.574, "t_data_s": 0.0, "t_rollout_s": 53.3, "t_step_s": 150.3, "t_refresh_s": 0.3, "mem_gb": 17.2, "mem_gb_teacher": 20.24}
106
+ {"step": 80, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.31916272058224376, "tokens": 170669, "cumulative_loss_tokens": 13762676, "grad_norm": 1.7109375, "lr": 1e-05, "finish_rate": 0.977, "comp_len": 666.7, "dropped_truncated": 0, "gold_loss": 1.2327, "gold_lambda": 0.5, "rep_ratio": 2.53, "t_data_s": 0.0, "t_rollout_s": 51.5, "t_step_s": 145.4, "t_refresh_s": 0.3, "mem_gb": 17.48, "mem_gb_teacher": 20.28}
107
+ [eval step 80] sample: 'Misy fitapana manokana ny fahaiza-mamorona, mba hanombanana, is a term that describes a situation where a person is engaging in activities that are not conducive to their well-being or health, often d'
108
+ {"step": 80, "gsm8k_n": 256, "gsm8k_quick_chat": 0.59765625, "t_eval_s": 21.4}
109
+ {"step": 81, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.2485308842504299, "tokens": 186465, "cumulative_loss_tokens": 13949141, "grad_norm": 1.953125, "lr": 1e-05, "finish_rate": 0.969, "comp_len": 728.4, "dropped_truncated": 0, "gold_loss": 1.1764, "gold_lambda": 0.5, "rep_ratio": 2.547, "t_data_s": 0.0, "t_rollout_s": 55.5, "t_step_s": 152.9, "t_refresh_s": 0.3, "mem_gb": 17.07, "mem_gb_teacher": 20.23}
110
+ {"step": 82, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.25967463855320316, "tokens": 190239, "cumulative_loss_tokens": 14139380, "grad_norm": 1.6484375, "lr": 1e-05, "finish_rate": 0.965, "comp_len": 743.1, "dropped_truncated": 0, "gold_loss": 1.3093, "gold_lambda": 0.5, "rep_ratio": 2.526, "t_data_s": 0.0, "t_rollout_s": 57.3, "t_step_s": 158.5, "t_refresh_s": 0.3, "mem_gb": 17.07, "mem_gb_teacher": 20.24}
111
+ {"step": 83, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.33700879973768977, "tokens": 182259, "cumulative_loss_tokens": 14321639, "grad_norm": 1.6875, "lr": 1e-05, "finish_rate": 0.973, "comp_len": 711.9, "dropped_truncated": 0, "gold_loss": 1.0788, "gold_lambda": 0.5, "rep_ratio": 2.505, "t_data_s": 0.0, "t_rollout_s": 55.0, "t_step_s": 152.8, "t_refresh_s": 0.3, "mem_gb": 17.47, "mem_gb_teacher": 20.27}
112
+ {"step": 84, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.2682194592400478, "tokens": 170072, "cumulative_loss_tokens": 14491711, "grad_norm": 1.3984375, "lr": 1e-05, "finish_rate": 0.973, "comp_len": 664.3, "dropped_truncated": 0, "gold_loss": 0.7509, "gold_lambda": 0.5, "rep_ratio": 2.597, "t_data_s": 0.0, "t_rollout_s": 51.4, "t_step_s": 144.0, "t_refresh_s": 0.3, "mem_gb": 17.44, "mem_gb_teacher": 20.27}
113
+ {"step": 85, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.23743917622813884, "tokens": 199223, "cumulative_loss_tokens": 14690934, "grad_norm": 1.8359375, "lr": 1e-05, "finish_rate": 0.961, "comp_len": 778.2, "dropped_truncated": 0, "gold_loss": 0.8562, "gold_lambda": 0.5, "rep_ratio": 2.627, "t_data_s": 0.0, "t_rollout_s": 58.4, "t_step_s": 159.8, "t_refresh_s": 0.3, "mem_gb": 17.48, "mem_gb_teacher": 20.28}
114
+ {"step": 86, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.40081178832628034, "tokens": 179491, "cumulative_loss_tokens": 14870425, "grad_norm": 1.96875, "lr": 1e-05, "finish_rate": 0.969, "comp_len": 701.1, "dropped_truncated": 0, "gold_loss": 0.7619, "gold_lambda": 0.5, "rep_ratio": 2.334, "t_data_s": 0.0, "t_rollout_s": 55.1, "t_step_s": 154.6, "t_refresh_s": 0.3, "mem_gb": 17.68, "mem_gb_teacher": 20.32}
115
+ {"step": 87, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.31119211183407003, "tokens": 161086, "cumulative_loss_tokens": 15031511, "grad_norm": 1.609375, "lr": 1e-05, "finish_rate": 0.965, "comp_len": 629.2, "dropped_truncated": 0, "gold_loss": 1.1431, "gold_lambda": 0.5, "rep_ratio": 2.341, "t_data_s": 0.0, "t_rollout_s": 48.7, "t_step_s": 139.4, "t_refresh_s": 0.3, "mem_gb": 17.28, "mem_gb_teacher": 20.25}
116
+ {"step": 88, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.21312830959642717, "tokens": 172945, "cumulative_loss_tokens": 15204456, "grad_norm": 1.53125, "lr": 1e-05, "finish_rate": 0.973, "comp_len": 675.6, "dropped_truncated": 0, "gold_loss": 1.0969, "gold_lambda": 0.5, "rep_ratio": 2.558, "t_data_s": 0.0, "t_rollout_s": 52.2, "t_step_s": 148.0, "t_refresh_s": 0.3, "mem_gb": 17.49, "mem_gb_teacher": 20.28}
117
+ {"step": 89, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.32083614396197874, "tokens": 197781, "cumulative_loss_tokens": 15402237, "grad_norm": 1.796875, "lr": 1e-05, "finish_rate": 0.957, "comp_len": 772.6, "dropped_truncated": 0, "gold_loss": 1.1581, "gold_lambda": 0.5, "rep_ratio": 2.371, "t_data_s": 0.0, "t_rollout_s": 59.1, "t_step_s": 158.0, "t_refresh_s": 0.3, "mem_gb": 17.13, "mem_gb_teacher": 20.24}
118
+ {"step": 90, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3539003647110563, "tokens": 177862, "cumulative_loss_tokens": 15580099, "grad_norm": 1.8046875, "lr": 1e-05, "finish_rate": 0.957, "comp_len": 694.8, "dropped_truncated": 0, "gold_loss": 1.1751, "gold_lambda": 0.5, "rep_ratio": 2.484, "t_data_s": 0.0, "t_rollout_s": 53.6, "t_step_s": 152.0, "t_refresh_s": 0.3, "mem_gb": 17.37, "mem_gb_teacher": 20.27}
119
+ [eval step 90] sample: 'Masy fitapana manokana ny fahaiza-mamorona?'
120
+ {"step": 91, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3003133533208709, "tokens": 190730, "cumulative_loss_tokens": 15770829, "grad_norm": 1.1328125, "lr": 1e-05, "finish_rate": 0.977, "comp_len": 745.0, "dropped_truncated": 0, "gold_loss": 0.2162, "gold_lambda": 0.5, "rep_ratio": 2.401, "t_data_s": 0.0, "t_rollout_s": 56.0, "t_step_s": 156.9, "t_refresh_s": 0.3, "mem_gb": 17.01, "mem_gb_teacher": 20.25}
121
+ {"step": 92, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3466024366871271, "tokens": 169647, "cumulative_loss_tokens": 15940476, "grad_norm": 1.34375, "lr": 1e-05, "finish_rate": 0.969, "comp_len": 662.7, "dropped_truncated": 0, "gold_loss": 0.2275, "gold_lambda": 0.5, "rep_ratio": 2.384, "t_data_s": 0.0, "t_rollout_s": 50.4, "t_step_s": 144.5, "t_refresh_s": 0.3, "mem_gb": 17.16, "mem_gb_teacher": 20.24}
122
+ {"step": 93, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.23514000853877456, "tokens": 167638, "cumulative_loss_tokens": 16108114, "grad_norm": 1.8359375, "lr": 1e-05, "finish_rate": 0.98, "comp_len": 654.8, "dropped_truncated": 0, "gold_loss": 1.6188, "gold_lambda": 0.5, "rep_ratio": 2.248, "t_data_s": 0.0, "t_rollout_s": 50.4, "t_step_s": 145.5, "t_refresh_s": 0.3, "mem_gb": 17.14, "mem_gb_teacher": 20.25}
123
+ {"step": 94, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.31567410076907493, "tokens": 176303, "cumulative_loss_tokens": 16284417, "grad_norm": 1.6484375, "lr": 1e-05, "finish_rate": 0.992, "comp_len": 688.7, "dropped_truncated": 0, "gold_loss": 1.4047, "gold_lambda": 0.5, "rep_ratio": 2.276, "t_data_s": 0.0, "t_rollout_s": 52.5, "t_step_s": 150.1, "t_refresh_s": 0.3, "mem_gb": 17.19, "mem_gb_teacher": 20.23}
124
+ {"step": 95, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3306478749962856, "tokens": 173098, "cumulative_loss_tokens": 16457515, "grad_norm": 1.6796875, "lr": 1e-05, "finish_rate": 0.973, "comp_len": 676.2, "dropped_truncated": 0, "gold_loss": 1.3459, "gold_lambda": 0.5, "rep_ratio": 2.403, "t_data_s": 0.0, "t_rollout_s": 51.7, "t_step_s": 145.0, "t_refresh_s": 0.3, "mem_gb": 17.25, "mem_gb_teacher": 20.24}
125
+ {"step": 96, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.2879815189713935, "tokens": 176694, "cumulative_loss_tokens": 16634209, "grad_norm": 1.4140625, "lr": 1e-05, "finish_rate": 0.977, "comp_len": 690.2, "dropped_truncated": 0, "gold_loss": 1.2006, "gold_lambda": 0.5, "rep_ratio": 2.383, "t_data_s": 0.0, "t_rollout_s": 51.7, "t_step_s": 148.5, "t_refresh_s": 0.3, "mem_gb": 17.54, "mem_gb_teacher": 20.28}
126
+ {"step": 97, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.23541706224584372, "tokens": 193571, "cumulative_loss_tokens": 16827780, "grad_norm": 2.34375, "lr": 1e-05, "finish_rate": 0.973, "comp_len": 756.1, "dropped_truncated": 0, "gold_loss": 1.2615, "gold_lambda": 0.5, "rep_ratio": 2.438, "t_data_s": 0.0, "t_rollout_s": 57.9, "t_step_s": 158.2, "t_refresh_s": 0.3, "mem_gb": 17.75, "mem_gb_teacher": 20.32}
127
+ {"step": 98, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.34905908556710274, "tokens": 149760, "cumulative_loss_tokens": 16977540, "grad_norm": 1.75, "lr": 1e-05, "finish_rate": 0.992, "comp_len": 585.0, "dropped_truncated": 0, "gold_loss": 1.1712, "gold_lambda": 0.5, "rep_ratio": 2.325, "t_data_s": 0.0, "t_rollout_s": 47.2, "t_step_s": 139.0, "t_refresh_s": 0.3, "mem_gb": 17.11, "mem_gb_teacher": 20.28}
128
+ {"step": 99, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.2776076079423005, "tokens": 183255, "cumulative_loss_tokens": 17160795, "grad_norm": 0.9765625, "lr": 1e-05, "finish_rate": 0.988, "comp_len": 715.8, "dropped_truncated": 0, "gold_loss": 0.2535, "gold_lambda": 0.5, "rep_ratio": 2.47, "t_data_s": 0.0, "t_rollout_s": 53.3, "t_step_s": 153.1, "t_refresh_s": 0.3, "mem_gb": 17.33, "mem_gb_teacher": 20.25}
129
+ {"step": 100, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.33610000682790214, "tokens": 155488, "cumulative_loss_tokens": 17316283, "grad_norm": 1.125, "lr": 1e-05, "finish_rate": 0.977, "comp_len": 607.4, "dropped_truncated": 0, "gold_loss": 0.2397, "gold_lambda": 0.5, "rep_ratio": 2.429, "t_data_s": 0.0, "t_rollout_s": 46.8, "t_step_s": 138.1, "t_refresh_s": 0.3, "mem_gb": 17.13, "mem_gb_teacher": 20.23}
130
+ [eval step 100] sample: 'Misy fitapana manokana ny fahaiza-mamorona, mnaya ny fahaiza-mamorona, mnaya ny fahaiza-mamorona, mnaya ny fahaiza-mamorona, mnaya ny fahaiza'
131
+ {"step": 100, "gsm8k_n": 256, "gsm8k_quick_chat": 0.6171875, "t_eval_s": 12.9}
132
+ {"step": 101, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.2934597294326003, "tokens": 180157, "cumulative_loss_tokens": 17496440, "grad_norm": 1.6328125, "lr": 1e-05, "finish_rate": 0.973, "comp_len": 703.7, "dropped_truncated": 0, "gold_loss": 1.1732, "gold_lambda": 0.5, "rep_ratio": 2.543, "t_data_s": 0.0, "t_rollout_s": 53.7, "t_step_s": 152.7, "t_refresh_s": 0.3, "mem_gb": 17.36, "mem_gb_teacher": 20.26}
133
+ {"step": 102, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.38382341174255885, "tokens": 152575, "cumulative_loss_tokens": 17649015, "grad_norm": 1.8203125, "lr": 1e-05, "finish_rate": 0.984, "comp_len": 596.0, "dropped_truncated": 0, "gold_loss": 1.3037, "gold_lambda": 0.5, "rep_ratio": 2.116, "t_data_s": 0.0, "t_rollout_s": 46.9, "t_step_s": 137.3, "t_refresh_s": 0.3, "mem_gb": 17.14, "mem_gb_teacher": 20.25}
134
+ {"step": 103, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3232517142888692, "tokens": 178507, "cumulative_loss_tokens": 17827522, "grad_norm": 1.109375, "lr": 1e-05, "finish_rate": 0.984, "comp_len": 697.3, "dropped_truncated": 0, "gold_loss": 0.2206, "gold_lambda": 0.5, "rep_ratio": 2.433, "t_data_s": 0.0, "t_rollout_s": 52.6, "t_step_s": 148.0, "t_refresh_s": 0.3, "mem_gb": 17.63, "mem_gb_teacher": 20.3}
135
+ {"step": 104, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.258444819787859, "tokens": 163920, "cumulative_loss_tokens": 17991442, "grad_norm": 1.0, "lr": 1e-05, "finish_rate": 0.988, "comp_len": 640.3, "dropped_truncated": 0, "gold_loss": 0.2333, "gold_lambda": 0.5, "rep_ratio": 2.313, "t_data_s": 0.0, "t_rollout_s": 49.4, "t_step_s": 142.5, "t_refresh_s": 0.3, "mem_gb": 17.27, "mem_gb_teacher": 20.27}
136
+ {"step": 105, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3201432822026658, "tokens": 134492, "cumulative_loss_tokens": 18125934, "grad_norm": 1.4921875, "lr": 1e-05, "finish_rate": 0.984, "comp_len": 525.4, "dropped_truncated": 0, "gold_loss": 0.8977, "gold_lambda": 0.5, "rep_ratio": 2.268, "t_data_s": 0.0, "t_rollout_s": 43.1, "t_step_s": 129.5, "t_refresh_s": 0.3, "mem_gb": 17.43, "mem_gb_teacher": 20.27}
137
+ {"step": 106, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.38415758914838366, "tokens": 173745, "cumulative_loss_tokens": 18299679, "grad_norm": 1.703125, "lr": 1e-05, "finish_rate": 0.973, "comp_len": 678.7, "dropped_truncated": 0, "gold_loss": 1.2116, "gold_lambda": 0.5, "rep_ratio": 2.414, "t_data_s": 0.0, "t_rollout_s": 53.3, "t_step_s": 151.0, "t_refresh_s": 0.3, "mem_gb": 17.51, "mem_gb_teacher": 20.3}
138
+ {"step": 107, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.27767968833353346, "tokens": 150360, "cumulative_loss_tokens": 18450039, "grad_norm": 1.625, "lr": 1e-05, "finish_rate": 0.984, "comp_len": 587.3, "dropped_truncated": 0, "gold_loss": 1.1874, "gold_lambda": 0.5, "rep_ratio": 2.226, "t_data_s": 0.0, "t_rollout_s": 46.6, "t_step_s": 135.4, "t_refresh_s": 0.3, "mem_gb": 17.1, "mem_gb_teacher": 20.25}
139
+ {"step": 108, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.2810955602886575, "tokens": 212108, "cumulative_loss_tokens": 18662147, "grad_norm": 1.890625, "lr": 1e-05, "finish_rate": 0.953, "comp_len": 828.5, "dropped_truncated": 0, "gold_loss": 1.2019, "gold_lambda": 0.5, "rep_ratio": 2.377, "t_data_s": 0.0, "t_rollout_s": 63.5, "t_step_s": 167.3, "t_refresh_s": 0.3, "mem_gb": 17.55, "mem_gb_teacher": 20.3}
140
+ {"step": 109, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3388609603948928, "tokens": 163795, "cumulative_loss_tokens": 18825942, "grad_norm": 1.515625, "lr": 1e-05, "finish_rate": 0.984, "comp_len": 639.8, "dropped_truncated": 0, "gold_loss": 0.8848, "gold_lambda": 0.5, "rep_ratio": 2.577, "t_data_s": 0.0, "t_rollout_s": 49.4, "t_step_s": 142.1, "t_refresh_s": 0.3, "mem_gb": 17.91, "mem_gb_teacher": 20.35}
141
+ {"step": 110, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.354563303726803, "tokens": 163003, "cumulative_loss_tokens": 18988945, "grad_norm": 1.4765625, "lr": 1e-05, "finish_rate": 0.977, "comp_len": 636.7, "dropped_truncated": 0, "gold_loss": 0.99, "gold_lambda": 0.5, "rep_ratio": 2.477, "t_data_s": 0.0, "t_rollout_s": 48.9, "t_step_s": 142.6, "t_refresh_s": 0.3, "mem_gb": 17.17, "mem_gb_teacher": 20.24}
142
+ [eval step 110] sample: 'Misy fitapana manokana ny fahaiza-mamorona, mnaya ny fahaiza-mamorona, mnaya ny fahaiza-mamorona, mnaya ny fahaiza-mamorona, mnaya ny fahaiza'
143
+ {"step": 111, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3537407948279594, "tokens": 195368, "cumulative_loss_tokens": 19184313, "grad_norm": 1.6015625, "lr": 1e-05, "finish_rate": 0.961, "comp_len": 763.2, "dropped_truncated": 0, "gold_loss": 1.1307, "gold_lambda": 0.5, "rep_ratio": 2.286, "t_data_s": 0.0, "t_rollout_s": 57.3, "t_step_s": 157.0, "t_refresh_s": 0.3, "mem_gb": 17.21, "mem_gb_teacher": 20.25}
144
+ {"step": 112, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.2772862368117265, "tokens": 204542, "cumulative_loss_tokens": 19388855, "grad_norm": 2.03125, "lr": 1e-05, "finish_rate": 0.965, "comp_len": 799.0, "dropped_truncated": 0, "gold_loss": 1.339, "gold_lambda": 0.5, "rep_ratio": 2.278, "t_data_s": 0.0, "t_rollout_s": 59.7, "t_step_s": 161.2, "t_refresh_s": 0.3, "mem_gb": 17.17, "mem_gb_teacher": 20.26}
145
+ {"step": 113, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.310700205462893, "tokens": 153520, "cumulative_loss_tokens": 19542375, "grad_norm": 1.515625, "lr": 1e-05, "finish_rate": 0.98, "comp_len": 599.7, "dropped_truncated": 0, "gold_loss": 0.8762, "gold_lambda": 0.5, "rep_ratio": 2.643, "t_data_s": 0.0, "t_rollout_s": 46.8, "t_step_s": 137.2, "t_refresh_s": 0.3, "mem_gb": 16.91, "mem_gb_teacher": 20.22}
146
+ {"step": 114, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.41260987902655, "tokens": 157613, "cumulative_loss_tokens": 19699988, "grad_norm": 1.4453125, "lr": 1e-05, "finish_rate": 0.996, "comp_len": 615.7, "dropped_truncated": 0, "gold_loss": 0.8989, "gold_lambda": 0.5, "rep_ratio": 2.193, "t_data_s": 0.0, "t_rollout_s": 47.3, "t_step_s": 133.6, "t_refresh_s": 0.3, "mem_gb": 17.36, "mem_gb_teacher": 20.26}wandb:
147
+ wandb: Run history:
148
+ wandb: comp_len ▆▆▄▆▄▄▅▄▄▇▄▄▄▃▅▄▆▆▅▃▄▇▅▆▅▅▅▅▅▅▆▅▄▁▄█▃▂▃▄
149
+ wandb: cumulative_loss_tokens ▁▁▁▁▁▂▂▂▂▃▃▃▃▃▃▄▄▄▄▄▄▅▅▅▅▅▅▅▅▆▆▆▇▇▇▇▇███
150
+ wandb: dropped_truncated ▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁
151
+ wandb: epoch ▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁
152
+ wandb: finish_rate ▇▄▇▁▄▆▇▇▆▅▂▄▅▅▅▅▅▅▅▃▆▂▅▆█▆▄▅▃▅▅▇▅▅▇▆▇▆▅▃
153
+ wandb: gold_lambda ▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁
154
+ wandb: gold_loss ▅▇▁▁▇▁▄▄▇▁▆▁▆▅█▆▁▁▄▆▅▁▅▆▆▆▄▄▄▆▇▆▁▆▁▆▄▄▁▅
155
+ wandb: grad_norm ▄▃▆▇▂█▁▄▅▃▄▂▃▃▂▃▁▃▂▃▂▃▂▄▁▃▂▄▂▃▂▂▃▁▂▂▂▃▄▂
156
+ wandb: gsm8k_quick_chat █▃▅▃▁▃▅
157
+ wandb: lr ▁▇██████████████████████████████████████
158
+ wandb: +10 ...
159
+ wandb:
160
+ wandb: Run summary:
161
+ wandb: comp_len 627.2
162
+ wandb: cumulative_loss_tokens 20669838
163
+ wandb: dropped_truncated 0
164
+ wandb: epoch 0
165
+ wandb: finish_rate 1
166
+ wandb: gold_lambda 0.5
167
+ wandb: gold_loss 1.0218
168
+ wandb: grad_norm 1.5
169
+ wandb: gsm8k_quick_chat 0.625
170
+ wandb: lr 1e-05
171
+ wandb: +11 ...
172
+ wandb:
173
+ wandb: You can sync this run to the cloud by running:
174
+ wandb: wandb sync outputs/healed/opd_warm_fixed_keep50/wandb/offline-run-20260802_065345-ubll0rd3
175
+ wandb: Find logs at: outputs/healed/opd_warm_fixed_keep50/wandb/offline-run-20260802_065345-ubll0rd3/logs
176
+
177
+ {"step": 115, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.4028610580933015, "tokens": 147120, "cumulative_loss_tokens": 19847108, "grad_norm": 1.28125, "lr": 1e-05, "finish_rate": 0.996, "comp_len": 574.7, "dropped_truncated": 0, "gold_loss": 0.1728, "gold_lambda": 0.5, "rep_ratio": 2.412, "t_data_s": 0.0, "t_rollout_s": 46.1, "t_step_s": 130.4, "t_refresh_s": 0.3, "mem_gb": 16.95, "mem_gb_teacher": 20.22}
178
+ {"step": 116, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.34066545782229457, "tokens": 173292, "cumulative_loss_tokens": 20020400, "grad_norm": 1.0625, "lr": 1e-05, "finish_rate": 0.98, "comp_len": 676.9, "dropped_truncated": 0, "gold_loss": 0.1759, "gold_lambda": 0.5, "rep_ratio": 2.363, "t_data_s": 0.0, "t_rollout_s": 51.2, "t_step_s": 144.9, "t_refresh_s": 0.3, "mem_gb": 17.43, "mem_gb_teacher": 20.28}
179
+ {"step": 117, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3317595209318613, "tokens": 152431, "cumulative_loss_tokens": 20172831, "grad_norm": 1.46875, "lr": 1e-05, "finish_rate": 0.988, "comp_len": 595.4, "dropped_truncated": 0, "gold_loss": 1.0952, "gold_lambda": 0.5, "rep_ratio": 2.472, "t_data_s": 0.0, "t_rollout_s": 47.2, "t_step_s": 134.1, "t_refresh_s": 0.3, "mem_gb": 17.4, "mem_gb_teacher": 20.27}
180
+ {"step": 118, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.30969671645999375, "tokens": 141862, "cumulative_loss_tokens": 20314693, "grad_norm": 1.734375, "lr": 1e-05, "finish_rate": 0.973, "comp_len": 554.1, "dropped_truncated": 0, "gold_loss": 1.3548, "gold_lambda": 0.5, "rep_ratio": 2.267, "t_data_s": 0.0, "t_rollout_s": 44.3, "t_step_s": 131.8, "t_refresh_s": 0.3, "mem_gb": 17.6, "mem_gb_teacher": 20.29}
181
+ {"step": 119, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.24082254176214968, "tokens": 194590, "cumulative_loss_tokens": 20509283, "grad_norm": 1.4921875, "lr": 1e-05, "finish_rate": 0.961, "comp_len": 760.1, "dropped_truncated": 0, "gold_loss": 1.0994, "gold_lambda": 0.5, "rep_ratio": 2.633, "t_data_s": 0.0, "t_rollout_s": 56.9, "t_step_s": 154.8, "t_refresh_s": 0.3, "mem_gb": 17.21, "mem_gb_teacher": 20.24}
182
+ {"step": 120, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3365587571631956, "tokens": 160555, "cumulative_loss_tokens": 20669838, "grad_norm": 1.5, "lr": 1e-05, "finish_rate": 1.0, "comp_len": 627.2, "dropped_truncated": 0, "gold_loss": 1.0218, "gold_lambda": 0.5, "rep_ratio": 2.389, "t_data_s": 0.0, "t_rollout_s": 46.6, "t_step_s": 126.9, "t_refresh_s": 0.0, "mem_gb": 16.8, "mem_gb_teacher": 20.2}
183
+ [eval step 120] sample: 'Misy fitapana manokana ny fahaiza-mamorona, mba hanombanana, is a term used to describe a condition where a person has a severe allergic reaction to a specific food or substance, often due to a sensit'
184
+ {"step": 120, "gsm8k_n": 256, "gsm8k_quick_chat": 0.625, "t_eval_s": 31.8}
185
+ checkpoint snapshot queued -> outputs/healed/opd_warm_fixed_keep50/step0120
186
+ wandb sync launched in background (pid 251372) -> outputs/healed/opd_warm_fixed_keep50/wandb_sync.log
healed/warm_chain.sh ADDED
@@ -0,0 +1,50 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #!/bin/bash
2
+ # Chain: wait for 1e-5 cold OPD (pid 89246) -> off-policy warmup heal of math
3
+ # keep-50 (regenerating the grid artifact deleted by drop_cell_checkpoints) ->
4
+ # warm-start OPD at 1e-5 (full Stable-OPD recipe, same as the cold control).
5
+ cd /home/henry/Documents/PythonProjects/variable-reap
6
+ LOGDIR=outputs/healed
7
+ echo "$(date +%T) chain: waiting for cold 1e-5 OPD (pid 89246) to finish"
8
+ while ps -p 89246 >/dev/null 2>&1; do sleep 300; done
9
+ echo "$(date +%T) chain: cold run done; settling 2 min"
10
+ sleep 120
11
+
12
+ echo "$(date +%T) chain: stage 1 - off-policy warmup heal (150 steps)"
13
+ PYTHONUNBUFFERED=1 PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True \
14
+ CUDA_VISIBLE_DEVICES=0 .venv/bin/python scripts/11_distill_on_policy.py \
15
+ --training-mode off-policy --kl-direction forward \
16
+ --topk-targets outputs/teacher_trajectories/dolci_math_curated_opd_top128 \
17
+ --off-policy-frames chat --student-device cuda:0 \
18
+ --student outputs/pruned/glean-0125inst-math-keep50 \
19
+ --lr 3e-5 --optimizer adamw8bit --epochs 3 --sweep 150 \
20
+ --micro-batch 3 --loss-tokens-per-step 120000 --gsm8k-every 0 \
21
+ --save-every 50 --seed 1224 \
22
+ --out-dir outputs/healed/keep50_offpolicy_warmup_s1224 \
23
+ --wandb --wandb-mode online --wandb-project glean-heal \
24
+ --wandb-run-name offpolicy-warmup-keep50-s1224 \
25
+ > "$LOGDIR/warmup_keep50.log" 2>&1
26
+ # do not gate on exit code (CUDA teardown segfault lesson); check the artifact
27
+ if [ ! -d outputs/healed/keep50_offpolicy_warmup_s1224/step0150 ]; then
28
+ echo "$(date +%T) chain: WARMUP FAILED (no step0150), aborting"; exit 1
29
+ fi
30
+ echo "$(date +%T) chain: warmup done; settling 2 min"
31
+ sleep 120
32
+
33
+ echo "$(date +%T) chain: stage 2 - warm-start OPD at 1e-5"
34
+ .venv/bin/python scripts/11_distill_on_policy.py \
35
+ --student outputs/healed/keep50_offpolicy_warmup_s1224/step0150 \
36
+ --teacher allenai/OLMoE-1B-7B-0125-Instruct \
37
+ --training-mode on-policy --kl-direction reverse \
38
+ --dataset allenai/Dolci-Instruct-RL \
39
+ --lr 1e-5 --optimizer adamw8bit --epochs 2 --sweep 120 \
40
+ --max-new-tokens 2048 --group-size 4 --prompts-per-step 256 --micro-batch 4 \
41
+ --seed 1223 --save-every 1000 --gsm8k-every 20 --gsm8k-max-new-tokens 1024 \
42
+ --fast-teacher --liger-loss --reference-kl-beta 0.05 \
43
+ --gold-mix-lambda 0.5 --gold-loss ce \
44
+ --gold-topk-targets outputs/teacher_trajectories/dolci_combined_top128 \
45
+ --rollout-engine vllm --vllm-gpu 2 --vllm-refresh-every 1 \
46
+ --student-device cuda:1 --teacher-device cuda:0 \
47
+ --out-dir outputs/healed/opd_warm_keep50 \
48
+ --wandb --wandb-project glean-heal --wandb-run-name opd-warm-keep50-s1223 \
49
+ > "$LOGDIR/opd_warm.log" 2>&1
50
+ echo "$(date +%T) chain: warm OPD exited"
healed/warm_chain2.sh ADDED
@@ -0,0 +1,81 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #!/bin/bash
2
+ # Chain v2 (post fused-MoE-no-autograd fix): off-policy warmup with the native
3
+ # differentiable student -> hard quality gate -> warm-start OPD at 1e-5.
4
+ # Neither stage uses --fast-student/--fast-teacher: the fork kernels are
5
+ # forward-only since the 07-25 re-pin (froze experts in every run since).
6
+ cd /home/henry/Documents/PythonProjects/variable-reap
7
+ LOGDIR=outputs/healed
8
+
9
+ echo "$(date +%T) chain2: stage 1 - off-policy warmup heal (150 steps, native)"
10
+ PYTHONUNBUFFERED=1 PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True \
11
+ CUDA_VISIBLE_DEVICES=0 .venv/bin/python scripts/11_distill_on_policy.py \
12
+ --training-mode off-policy --kl-direction forward \
13
+ --topk-targets outputs/teacher_trajectories/dolci_math_curated_opd_top128 \
14
+ --off-policy-frames chat --student-device cuda:0 \
15
+ --student outputs/pruned/glean-0125inst-math-keep50 \
16
+ --lr 3e-5 --optimizer adamw8bit --epochs 3 --sweep 150 \
17
+ --micro-batch 3 --loss-tokens-per-step 120000 --gsm8k-every 0 \
18
+ --save-every 50 --seed 1224 \
19
+ --out-dir outputs/healed/keep50_warmup_fixed_s1224 \
20
+ --wandb --wandb-mode online --wandb-project glean-heal \
21
+ --wandb-run-name warmup-fixed-keep50-s1224 \
22
+ > "$LOGDIR/warmup_fixed.log" 2>&1
23
+ if [ ! -d outputs/healed/keep50_warmup_fixed_s1224/step0150 ]; then
24
+ echo "$(date +%T) chain2: WARMUP FAILED (no step0150)"; exit 1
25
+ fi
26
+
27
+ echo "$(date +%T) chain2: gate - probe warmup checkpoint"
28
+ .venv/bin/python - <<'EOF' > "$LOGDIR/warmup_gate.log" 2>&1
29
+ import sys, torch, glob
30
+ from safetensors import safe_open
31
+ from transformers import AutoTokenizer, AutoModelForCausalLM
32
+
33
+ def get(path, key):
34
+ for f in glob.glob(f"{path}/*.safetensors"):
35
+ with safe_open(f, framework="pt") as sf:
36
+ if key in sf.keys(): return sf.get_tensor(key)
37
+
38
+ k = "model.layers.0.mlp.experts.0.gate_proj.weight"
39
+ d = (get("outputs/healed/keep50_warmup_fixed_s1224/step0150", k).float()
40
+ - get("outputs/pruned/glean-0125inst-math-keep50", k).float()).abs().max().item()
41
+ print(f"expert max|delta| = {d:.3e}")
42
+ assert d > 1e-4, "experts did not train"
43
+
44
+ tok = AutoTokenizer.from_pretrained("allenai/OLMoE-1B-7B-0125-Instruct")
45
+ m = AutoModelForCausalLM.from_pretrained(
46
+ "outputs/healed/keep50_warmup_fixed_s1224/step0150",
47
+ dtype=torch.bfloat16, trust_remote_code=True).cuda().eval()
48
+ q = ("Natalia sold clips to 48 of her friends in April, and then she sold "
49
+ "half as many clips in May. How many clips did Natalia sell altogether "
50
+ "in April and May?")
51
+ ids = tok.apply_chat_template([{"role":"user","content":q}],
52
+ add_generation_prompt=True, return_tensors="pt").cuda()
53
+ with torch.no_grad():
54
+ out = m.generate(ids, max_new_tokens=200, do_sample=False)
55
+ text = tok.decode(out[0, ids.shape[1]:], skip_special_tokens=True)
56
+ print(text)
57
+ assert "72" in text, "did not produce the correct answer 72"
58
+ print("GATE PASSED")
59
+ EOF
60
+ if ! grep -q "GATE PASSED" "$LOGDIR/warmup_gate.log"; then
61
+ echo "$(date +%T) chain2: GATE FAILED - not launching OPD"; exit 1
62
+ fi
63
+
64
+ echo "$(date +%T) chain2: stage 2 - warm-start OPD at 1e-5 (native both)"
65
+ .venv/bin/python scripts/11_distill_on_policy.py \
66
+ --student outputs/healed/keep50_warmup_fixed_s1224/step0150 \
67
+ --teacher allenai/OLMoE-1B-7B-0125-Instruct \
68
+ --training-mode on-policy --kl-direction reverse \
69
+ --dataset allenai/Dolci-Instruct-RL \
70
+ --lr 1e-5 --optimizer adamw8bit --epochs 2 --sweep 120 \
71
+ --max-new-tokens 2048 --group-size 4 --prompts-per-step 256 --micro-batch 4 \
72
+ --seed 1223 --save-every 1000 --gsm8k-every 20 --gsm8k-max-new-tokens 1024 \
73
+ --liger-loss --reference-kl-beta 0.05 \
74
+ --gold-mix-lambda 0.5 --gold-loss ce \
75
+ --gold-topk-targets outputs/teacher_trajectories/dolci_combined_top128 \
76
+ --rollout-engine vllm --vllm-gpu 2 --vllm-refresh-every 1 \
77
+ --student-device cuda:1 --teacher-device cuda:0 \
78
+ --out-dir outputs/healed/opd_warm_fixed_keep50 \
79
+ --wandb --wandb-project glean-heal --wandb-run-name opd-warm-fixed-keep50-s1223 \
80
+ > "$LOGDIR/opd_warm_fixed.log" 2>&1
81
+ echo "$(date +%T) chain2: warm OPD exited"
healed/warmup_fixed.log ADDED
@@ -0,0 +1,213 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ wandb: [wandb.login()] Loaded credentials for https://api.wandb.ai from /home/henry/.netrc.
2
+ wandb: Currently logged in as: hbfreed to https://api.wandb.ai. Use `wandb login --relogin` to force relogin
3
+ wandb: Tracking run with wandb version 0.28.0
4
+ wandb: Run data is saved locally in outputs/healed/keep50_warmup_fixed_s1224/wandb/run-20260802_025702-sep45w2j
5
+ wandb: Run `wandb offline` to turn off syncing.
6
+ wandb: Syncing run warmup-fixed-keep50-s1224
7
+ wandb: ⭐️ View project at https://wandb.ai/hbfreed/glean-heal
8
+ wandb: 🚀 View run at https://wandb.ai/hbfreed/glean-heal/runs/sep45w2j
9
+
10
+ 12115 cached top-128 chat trajectories / 6,476,634 unique tokens | 53 steps/epoch | 150 total steps | student params 3.70B | teacher overlap=False
11
+ {"step": 1, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.21896971870896717, "tokens": 120000, "cumulative_loss_tokens": 120000, "grad_norm": 4.6875, "lr": 6e-06, "finish_rate": 0.907, "comp_len": 508.5, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.337, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 99.1, "frames": {"chat": 236}, "mem_gb": 15.78, "mem_gb_teacher": 15.78}
12
+ The attention mask is not set and cannot be inferred from input because pad token is same as eos token. As a consequence, you may observe unexpected behavior. Please pass your input's `attention_mask` to obtain reliable results.
13
+ [eval step 1] sample: 'To solve the given system of equations:\n\\[\n\\begin{align*}\na + b &= k, \\\\\nk + m &= p, \\\\\np + a &= r, \\\\\nb + m + r &= 18,\n\\end{align*}\n\\]\nwe need to determine the values'
14
+ {"step": 2, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.27215903437460465, "tokens": 120000, "cumulative_loss_tokens": 240000, "grad_norm": 4.84375, "lr": 9e-06, "finish_rate": 0.781, "comp_len": 558.1, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.474, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 90.6, "frames": {"chat": 215}, "mem_gb": 16.05, "mem_gb_teacher": 16.05}
15
+ {"step": 3, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.2780314308715363, "tokens": 120000, "cumulative_loss_tokens": 360000, "grad_norm": 4.09375, "lr": 1.2e-05, "finish_rate": 0.825, "comp_len": 553.0, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.376, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 90.6, "frames": {"chat": 217}, "mem_gb": 15.93, "mem_gb_teacher": 15.93}
16
+ {"step": 4, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.21216022065331538, "tokens": 120000, "cumulative_loss_tokens": 480000, "grad_norm": 2.546875, "lr": 1.5e-05, "finish_rate": 0.8, "comp_len": 585.4, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.366, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 86.0, "frames": {"chat": 205}, "mem_gb": 15.99, "mem_gb_teacher": 15.99}
17
+ {"step": 5, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.13948054732351253, "tokens": 120000, "cumulative_loss_tokens": 600000, "grad_norm": 1.6484375, "lr": 1.8e-05, "finish_rate": 0.834, "comp_len": 524.0, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.318, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 94.5, "frames": {"chat": 229}, "mem_gb": 15.96, "mem_gb_teacher": 15.96}
18
+ {"step": 6, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.2525411546646928, "tokens": 120000, "cumulative_loss_tokens": 720000, "grad_norm": 2.578125, "lr": 2.1e-05, "finish_rate": 0.812, "comp_len": 538.1, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.336, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 92.0, "frames": {"chat": 223}, "mem_gb": 16.03, "mem_gb_teacher": 16.03}
19
+ {"step": 7, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.1408244575532774, "tokens": 120000, "cumulative_loss_tokens": 840000, "grad_norm": 1.171875, "lr": 2.4e-05, "finish_rate": 0.708, "comp_len": 594.1, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.42, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 86.3, "frames": {"chat": 202}, "mem_gb": 16.07, "mem_gb_teacher": 16.07}
20
+ {"step": 8, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.16170320059585697, "tokens": 120000, "cumulative_loss_tokens": 960000, "grad_norm": 1.171875, "lr": 2.7000000000000002e-05, "finish_rate": 0.77, "comp_len": 574.2, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.453, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 87.6, "frames": {"chat": 209}, "mem_gb": 16.04, "mem_gb_teacher": 16.04}
21
+ {"step": 9, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.1060452919805112, "tokens": 120000, "cumulative_loss_tokens": 1080000, "grad_norm": 0.703125, "lr": 3e-05, "finish_rate": 0.885, "comp_len": 528.6, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.397, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 93.7, "frames": {"chat": 227}, "mem_gb": 16.02, "mem_gb_teacher": 16.02}
22
+ {"step": 10, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.11166963671234746, "tokens": 120000, "cumulative_loss_tokens": 1200000, "grad_norm": 0.66796875, "lr": 3e-05, "finish_rate": 0.848, "comp_len": 521.7, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.546, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 95.5, "frames": {"chat": 230}, "mem_gb": 16.09, "mem_gb_teacher": 16.09}
23
+ [eval step 10] sample: "To solve this system of equations, we need to determine the values of \\(a\\), \\(b\\), \\(k\\), \\(m\\), and \\(p\\) given the constraints that each letter represents a non-zero digit.\n\nLet's break down the pr"
24
+ {"step": 11, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.09883926218959192, "tokens": 120000, "cumulative_loss_tokens": 1320000, "grad_norm": 0.671875, "lr": 3e-05, "finish_rate": 0.879, "comp_len": 519.5, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.426, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 94.7, "frames": {"chat": 231}, "mem_gb": 15.94, "mem_gb_teacher": 15.94}
25
+ {"step": 12, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.09610702018613616, "tokens": 120000, "cumulative_loss_tokens": 1440000, "grad_norm": 0.55078125, "lr": 3e-05, "finish_rate": 0.882, "comp_len": 489.8, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.246, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 99.3, "frames": {"chat": 245}, "mem_gb": 16.02, "mem_gb_teacher": 16.02}
26
+ {"step": 13, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.09665392311389248, "tokens": 120000, "cumulative_loss_tokens": 1560000, "grad_norm": 0.5703125, "lr": 3e-05, "finish_rate": 0.81, "comp_len": 571.4, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.375, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 87.0, "frames": {"chat": 210}, "mem_gb": 16.02, "mem_gb_teacher": 16.02}
27
+ {"step": 14, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.1129512736000431, "tokens": 120000, "cumulative_loss_tokens": 1680000, "grad_norm": 0.578125, "lr": 3e-05, "finish_rate": 0.758, "comp_len": 568.7, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.312, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 89.4, "frames": {"chat": 211}, "mem_gb": 16.02, "mem_gb_teacher": 16.02}
28
+ {"step": 15, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.09338119786353782, "tokens": 120000, "cumulative_loss_tokens": 1800000, "grad_norm": 0.453125, "lr": 3e-05, "finish_rate": 0.819, "comp_len": 543.0, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.413, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 92.0, "frames": {"chat": 221}, "mem_gb": 16.08, "mem_gb_teacher": 16.08}
29
+ {"step": 16, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.08123978671409811, "tokens": 120000, "cumulative_loss_tokens": 1920000, "grad_norm": 0.490234375, "lr": 3e-05, "finish_rate": 0.912, "comp_len": 480.0, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.366, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 101.9, "frames": {"chat": 250}, "mem_gb": 15.89, "mem_gb_teacher": 15.89}
30
+ {"step": 17, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.08966803371421993, "tokens": 120000, "cumulative_loss_tokens": 2040000, "grad_norm": 0.49609375, "lr": 3e-05, "finish_rate": 0.79, "comp_len": 524.0, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.402, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 95.6, "frames": {"chat": 229}, "mem_gb": 16.06, "mem_gb_teacher": 16.06}
31
+ {"step": 18, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.06860631760672356, "tokens": 120000, "cumulative_loss_tokens": 2160000, "grad_norm": 0.408203125, "lr": 3e-05, "finish_rate": 0.888, "comp_len": 480.0, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.479, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 101.4, "frames": {"chat": 250}, "mem_gb": 16.04, "mem_gb_teacher": 16.04}
32
+ {"step": 19, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.08370754941549773, "tokens": 120000, "cumulative_loss_tokens": 2280000, "grad_norm": 0.421875, "lr": 3e-05, "finish_rate": 0.844, "comp_len": 519.5, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.363, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 95.2, "frames": {"chat": 231}, "mem_gb": 15.91, "mem_gb_teacher": 15.91}
33
+ {"step": 20, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.07473204600410537, "tokens": 120000, "cumulative_loss_tokens": 2400000, "grad_norm": 0.41796875, "lr": 3e-05, "finish_rate": 0.844, "comp_len": 535.7, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.433, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 92.5, "frames": {"chat": 224}, "mem_gb": 15.96, "mem_gb_teacher": 15.96}
34
+ [eval step 20] sample: 'To solve the system of equations given:\n\n\\[\n\\begin{align*}\na + b &= k \\\\\nk + m &= p \\\\\np + a &= r \\\\\nb + m + r &= 18\n\\end{align*}\n\\]\n\nwe need to determine the values of \\('
35
+ {"step": 21, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.08024157820896556, "tokens": 120000, "cumulative_loss_tokens": 2520000, "grad_norm": 0.486328125, "lr": 3e-05, "finish_rate": 0.802, "comp_len": 566.0, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.329, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 88.4, "frames": {"chat": 212}, "mem_gb": 16.0, "mem_gb_teacher": 16.0}
36
+ {"step": 22, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.0721184289892825, "tokens": 120000, "cumulative_loss_tokens": 2640000, "grad_norm": 0.435546875, "lr": 3e-05, "finish_rate": 0.87, "comp_len": 504.2, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.376, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 97.3, "frames": {"chat": 238}, "mem_gb": 15.95, "mem_gb_teacher": 15.95}
37
+ {"step": 23, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.07248173923180438, "tokens": 120000, "cumulative_loss_tokens": 2760000, "grad_norm": 0.443359375, "lr": 3e-05, "finish_rate": 0.903, "comp_len": 466.9, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.445, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 103.5, "frames": {"chat": 257}, "mem_gb": 15.83, "mem_gb_teacher": 15.83}
38
+ {"step": 24, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.0655906916304181, "tokens": 120000, "cumulative_loss_tokens": 2880000, "grad_norm": 0.36328125, "lr": 3e-05, "finish_rate": 0.868, "comp_len": 528.6, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.405, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 93.3, "frames": {"chat": 227}, "mem_gb": 16.02, "mem_gb_teacher": 16.02}
39
+ {"step": 25, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.07533252080177578, "tokens": 120000, "cumulative_loss_tokens": 3000000, "grad_norm": 0.380859375, "lr": 3e-05, "finish_rate": 0.838, "comp_len": 526.3, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.359, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 93.6, "frames": {"chat": 228}, "mem_gb": 16.05, "mem_gb_teacher": 16.05}
40
+ {"step": 26, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.07692283792655605, "tokens": 120000, "cumulative_loss_tokens": 3120000, "grad_norm": 0.40625, "lr": 3e-05, "finish_rate": 0.803, "comp_len": 515.0, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.542, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 97.0, "frames": {"chat": 233}, "mem_gb": 16.04, "mem_gb_teacher": 16.04}
41
+ {"step": 27, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.06803224476923546, "tokens": 120000, "cumulative_loss_tokens": 3240000, "grad_norm": 0.3828125, "lr": 3e-05, "finish_rate": 0.863, "comp_len": 515.0, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.351, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 96.2, "frames": {"chat": 233}, "mem_gb": 16.04, "mem_gb_teacher": 16.04}
42
+ {"step": 28, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.09600000411309302, "tokens": 120000, "cumulative_loss_tokens": 3360000, "grad_norm": 0.46484375, "lr": 3e-05, "finish_rate": 0.731, "comp_len": 609.1, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.388, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 85.0, "frames": {"chat": 197}, "mem_gb": 16.13, "mem_gb_teacher": 16.13}
43
+ {"step": 29, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.0838949200007754, "tokens": 120000, "cumulative_loss_tokens": 3480000, "grad_norm": 0.404296875, "lr": 3e-05, "finish_rate": 0.862, "comp_len": 502.1, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.384, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 98.3, "frames": {"chat": 239}, "mem_gb": 15.88, "mem_gb_teacher": 15.88}
44
+ {"step": 30, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.08550032369385784, "tokens": 120000, "cumulative_loss_tokens": 3600000, "grad_norm": 0.640625, "lr": 3e-05, "finish_rate": 0.83, "comp_len": 535.7, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.302, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 91.8, "frames": {"chat": 224}, "mem_gb": 15.93, "mem_gb_teacher": 15.93}
45
+ [eval step 30] sample: 'To solve the given system of equations for \\(a\\), \\(b\\), \\(k\\), \\(m\\), and \\(p\\), we need to follow these steps:\n\n1. **Understand the Equations:**\n \\[\n \\begin{align*}\n a + b &= k \\\\'
46
+ {"step": 31, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.06893708595448794, "tokens": 120000, "cumulative_loss_tokens": 3720000, "grad_norm": 0.3828125, "lr": 3e-05, "finish_rate": 0.788, "comp_len": 553.0, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.388, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 90.4, "frames": {"chat": 217}, "mem_gb": 16.04, "mem_gb_teacher": 16.04}
47
+ {"step": 32, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.0662639382578743, "tokens": 120000, "cumulative_loss_tokens": 3840000, "grad_norm": 0.376953125, "lr": 3e-05, "finish_rate": 0.876, "comp_len": 497.9, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.259, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 98.4, "frames": {"chat": 241}, "mem_gb": 16.04, "mem_gb_teacher": 16.04}
48
+ {"step": 33, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.06439438750580885, "tokens": 120000, "cumulative_loss_tokens": 3960000, "grad_norm": 0.39453125, "lr": 3e-05, "finish_rate": 0.835, "comp_len": 550.5, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.407, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 89.8, "frames": {"chat": 218}, "mem_gb": 16.02, "mem_gb_teacher": 16.02}
49
+ {"step": 34, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.06970151182319968, "tokens": 120000, "cumulative_loss_tokens": 4080000, "grad_norm": 0.353515625, "lr": 3e-05, "finish_rate": 0.767, "comp_len": 582.5, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.439, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 86.2, "frames": {"chat": 206}, "mem_gb": 16.03, "mem_gb_teacher": 16.03}
50
+ {"step": 35, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.06962556957538861, "tokens": 120000, "cumulative_loss_tokens": 4200000, "grad_norm": 0.37890625, "lr": 3e-05, "finish_rate": 0.845, "comp_len": 517.2, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.512, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 96.4, "frames": {"chat": 232}, "mem_gb": 16.07, "mem_gb_teacher": 16.07}
51
+ {"step": 36, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.07808773103132843, "tokens": 120000, "cumulative_loss_tokens": 4320000, "grad_norm": 0.419921875, "lr": 3e-05, "finish_rate": 0.771, "comp_len": 550.5, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.432, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 92.0, "frames": {"chat": 218}, "mem_gb": 16.09, "mem_gb_teacher": 16.09}
52
+ {"step": 37, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.07417992857682208, "tokens": 120000, "cumulative_loss_tokens": 4440000, "grad_norm": 0.376953125, "lr": 3e-05, "finish_rate": 0.779, "comp_len": 563.4, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.428, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 88.9, "frames": {"chat": 213}, "mem_gb": 16.05, "mem_gb_teacher": 16.05}
53
+ {"step": 38, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.07201246263841167, "tokens": 120000, "cumulative_loss_tokens": 4560000, "grad_norm": 0.369140625, "lr": 3e-05, "finish_rate": 0.887, "comp_len": 483.9, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.527, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 100.9, "frames": {"chat": 248}, "mem_gb": 16.02, "mem_gb_teacher": 16.02}
54
+ {"step": 39, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.05988815330729509, "tokens": 120000, "cumulative_loss_tokens": 4680000, "grad_norm": 0.373046875, "lr": 3e-05, "finish_rate": 0.803, "comp_len": 550.5, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.333, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 90.1, "frames": {"chat": 218}, "mem_gb": 16.08, "mem_gb_teacher": 16.08}
55
+ {"step": 40, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.05963528015368308, "tokens": 120000, "cumulative_loss_tokens": 4800000, "grad_norm": 0.3359375, "lr": 3e-05, "finish_rate": 0.851, "comp_len": 543.0, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.401, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 91.5, "frames": {"chat": 221}, "mem_gb": 16.04, "mem_gb_teacher": 16.04}
56
+ [eval step 40] sample: "To solve the given system of equations, we need to determine the values of \\(a\\), \\(b\\), \\(k\\), \\(m\\), and \\(p\\) such that each letter represents a non-zero digit. Let's break down the problem step-by"
57
+ {"step": 41, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.05816531496203194, "tokens": 120000, "cumulative_loss_tokens": 4920000, "grad_norm": 0.345703125, "lr": 3e-05, "finish_rate": 0.894, "comp_len": 508.5, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.396, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 96.1, "frames": {"chat": 236}, "mem_gb": 15.97, "mem_gb_teacher": 15.97}
58
+ {"step": 42, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.06005277933338657, "tokens": 120000, "cumulative_loss_tokens": 5040000, "grad_norm": 0.326171875, "lr": 3e-05, "finish_rate": 0.858, "comp_len": 487.8, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.338, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 100.6, "frames": {"chat": 246}, "mem_gb": 15.89, "mem_gb_teacher": 15.89}
59
+ {"step": 43, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.06171362210111692, "tokens": 120000, "cumulative_loss_tokens": 5160000, "grad_norm": 0.35546875, "lr": 3e-05, "finish_rate": 0.838, "comp_len": 512.8, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.48, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 96.2, "frames": {"chat": 234}, "mem_gb": 16.14, "mem_gb_teacher": 16.14}
60
+ {"step": 44, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.055975045030827945, "tokens": 120000, "cumulative_loss_tokens": 5280000, "grad_norm": 0.3515625, "lr": 3e-05, "finish_rate": 0.748, "comp_len": 594.1, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.474, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 85.4, "frames": {"chat": 202}, "mem_gb": 16.03, "mem_gb_teacher": 16.03}
61
+ {"step": 45, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.06047188884726105, "tokens": 120000, "cumulative_loss_tokens": 5400000, "grad_norm": 0.3359375, "lr": 3e-05, "finish_rate": 0.811, "comp_len": 553.0, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.36, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 90.3, "frames": {"chat": 217}, "mem_gb": 16.04, "mem_gb_teacher": 16.04}
62
+ {"step": 46, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.05655016646341731, "tokens": 120000, "cumulative_loss_tokens": 5520000, "grad_norm": 0.357421875, "lr": 3e-05, "finish_rate": 0.866, "comp_len": 535.7, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.47, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 92.3, "frames": {"chat": 224}, "mem_gb": 16.04, "mem_gb_teacher": 16.04}
63
+ {"step": 47, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.0790889122961089, "tokens": 120000, "cumulative_loss_tokens": 5640000, "grad_norm": 0.828125, "lr": 3e-05, "finish_rate": 0.753, "comp_len": 558.1, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.285, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 89.5, "frames": {"chat": 215}, "mem_gb": 16.05, "mem_gb_teacher": 16.05}
64
+ {"step": 48, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.05377363468687981, "tokens": 120000, "cumulative_loss_tokens": 5760000, "grad_norm": 0.353515625, "lr": 3e-05, "finish_rate": 0.884, "comp_len": 463.3, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.356, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 104.3, "frames": {"chat": 259}, "mem_gb": 15.97, "mem_gb_teacher": 15.97}
65
+ {"step": 49, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.06014201375305032, "tokens": 120000, "cumulative_loss_tokens": 5880000, "grad_norm": 0.3359375, "lr": 3e-05, "finish_rate": 0.829, "comp_len": 571.4, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.475, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 86.9, "frames": {"chat": 210}, "mem_gb": 16.04, "mem_gb_teacher": 16.04}
66
+ {"step": 50, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.07395202046850076, "tokens": 120000, "cumulative_loss_tokens": 6000000, "grad_norm": 0.416015625, "lr": 3e-05, "finish_rate": 0.77, "comp_len": 563.4, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.304, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 89.3, "frames": {"chat": 213}, "mem_gb": 16.09, "mem_gb_teacher": 16.09}
67
+ [eval step 50] sample: "To solve the given system of equations, we need to determine the values of \\(a\\), \\(b\\), \\(k\\), \\(m\\), and \\(p\\) such that each letter represents a non-zero digit. Let's break down the problem step-by"
68
+ checkpoint snapshot queued -> outputs/healed/keep50_warmup_fixed_s1224/step0050
69
+ {"step": 51, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.05859705059945894, "tokens": 120000, "cumulative_loss_tokens": 6120000, "grad_norm": 0.34765625, "lr": 3e-05, "finish_rate": 0.815, "comp_len": 540.5, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.453, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 90.8, "frames": {"chat": 222}, "mem_gb": 16.0, "mem_gb_teacher": 16.0}
70
+ {"step": 52, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.0539097297622667, "tokens": 120000, "cumulative_loss_tokens": 6240000, "grad_norm": 0.353515625, "lr": 3e-05, "finish_rate": 0.889, "comp_len": 510.6, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.439, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 96.1, "frames": {"chat": 235}, "mem_gb": 16.05, "mem_gb_teacher": 16.05}
71
+ {"step": 53, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.0594748610290233, "tokens": 120000, "cumulative_loss_tokens": 6360000, "grad_norm": 0.328125, "lr": 3e-05, "finish_rate": 0.798, "comp_len": 576.9, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.417, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 86.8, "frames": {"chat": 208}, "mem_gb": 16.01, "mem_gb_teacher": 16.01}
72
+ {"step": 54, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.045099887475029875, "tokens": 120000, "cumulative_loss_tokens": 6480000, "grad_norm": 0.275390625, "lr": 3e-05, "finish_rate": 0.733, "comp_len": 628.3, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.413, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 80.2, "frames": {"chat": 191}, "mem_gb": 16.05, "mem_gb_teacher": 16.05}
73
+ {"step": 55, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.03569839329215077, "tokens": 120000, "cumulative_loss_tokens": 6600000, "grad_norm": 0.2451171875, "lr": 3e-05, "finish_rate": 0.845, "comp_len": 547.9, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.523, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 90.0, "frames": {"chat": 219}, "mem_gb": 16.04, "mem_gb_teacher": 16.04}
74
+ {"step": 56, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.04039377661425, "tokens": 120000, "cumulative_loss_tokens": 6720000, "grad_norm": 0.302734375, "lr": 3e-05, "finish_rate": 0.778, "comp_len": 579.7, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.459, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 86.0, "frames": {"chat": 207}, "mem_gb": 16.05, "mem_gb_teacher": 16.05}
75
+ {"step": 57, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.055602129854184265, "tokens": 120000, "cumulative_loss_tokens": 6840000, "grad_norm": 0.330078125, "lr": 3e-05, "finish_rate": 0.755, "comp_len": 576.9, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.449, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 88.0, "frames": {"chat": 208}, "mem_gb": 16.01, "mem_gb_teacher": 16.01}
76
+ {"step": 58, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.035877808295966436, "tokens": 120000, "cumulative_loss_tokens": 6960000, "grad_norm": 0.279296875, "lr": 3e-05, "finish_rate": 0.799, "comp_len": 547.9, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.288, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 90.8, "frames": {"chat": 219}, "mem_gb": 16.04, "mem_gb_teacher": 16.04}
77
+ {"step": 59, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.03478414489501932, "tokens": 120000, "cumulative_loss_tokens": 7080000, "grad_norm": 0.2578125, "lr": 3e-05, "finish_rate": 0.915, "comp_len": 487.8, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.329, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 99.7, "frames": {"chat": 246}, "mem_gb": 15.92, "mem_gb_teacher": 15.92}
78
+ {"step": 60, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.04739725781680706, "tokens": 120000, "cumulative_loss_tokens": 7200000, "grad_norm": 0.330078125, "lr": 3e-05, "finish_rate": 0.704, "comp_len": 582.5, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.437, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 87.5, "frames": {"chat": 206}, "mem_gb": 16.07, "mem_gb_teacher": 16.07}
79
+ [eval step 60] sample: 'To solve the given system of equations for \\(a\\), \\(b\\), \\(k\\), \\(m\\), and \\(p\\), we need to determine the values of these variables such that each letter represents a non-zero digit.\n\nThe equations a'
80
+ {"step": 61, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.03820183755345643, "tokens": 120000, "cumulative_loss_tokens": 7320000, "grad_norm": 0.275390625, "lr": 3e-05, "finish_rate": 0.876, "comp_len": 515.0, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.375, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 95.5, "frames": {"chat": 233}, "mem_gb": 16.05, "mem_gb_teacher": 16.05}
81
+ {"step": 62, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.04388709897000032, "tokens": 120000, "cumulative_loss_tokens": 7440000, "grad_norm": 0.275390625, "lr": 3e-05, "finish_rate": 0.847, "comp_len": 524.0, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.363, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 93.8, "frames": {"chat": 229}, "mem_gb": 15.92, "mem_gb_teacher": 15.92}
82
+ {"step": 63, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.032643947703313705, "tokens": 120000, "cumulative_loss_tokens": 7560000, "grad_norm": 0.236328125, "lr": 3e-05, "finish_rate": 0.864, "comp_len": 508.5, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.285, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 96.4, "frames": {"chat": 236}, "mem_gb": 15.95, "mem_gb_teacher": 15.95}
83
+ {"step": 64, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.04086883016227123, "tokens": 120000, "cumulative_loss_tokens": 7680000, "grad_norm": 0.2451171875, "lr": 3e-05, "finish_rate": 0.87, "comp_len": 502.1, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.37, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 98.3, "frames": {"chat": 239}, "mem_gb": 15.84, "mem_gb_teacher": 15.84}
84
+ {"step": 65, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.03435394472128246, "tokens": 120000, "cumulative_loss_tokens": 7800000, "grad_norm": 0.2177734375, "lr": 3e-05, "finish_rate": 0.867, "comp_len": 497.9, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.349, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 99.1, "frames": {"chat": 241}, "mem_gb": 15.96, "mem_gb_teacher": 15.96}
85
+ {"step": 66, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.036759108127855385, "tokens": 120000, "cumulative_loss_tokens": 7920000, "grad_norm": 0.2294921875, "lr": 3e-05, "finish_rate": 0.863, "comp_len": 531.0, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.44, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 93.4, "frames": {"chat": 226}, "mem_gb": 15.92, "mem_gb_teacher": 15.92}
86
+ {"step": 67, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.031785604377323765, "tokens": 120000, "cumulative_loss_tokens": 8040000, "grad_norm": 0.224609375, "lr": 3e-05, "finish_rate": 0.893, "comp_len": 512.8, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.32, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 95.3, "frames": {"chat": 234}, "mem_gb": 16.05, "mem_gb_teacher": 16.05}
87
+ {"step": 68, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.03207274362700991, "tokens": 120000, "cumulative_loss_tokens": 8160000, "grad_norm": 0.22265625, "lr": 3e-05, "finish_rate": 0.914, "comp_len": 466.9, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.401, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 102.7, "frames": {"chat": 257}, "mem_gb": 16.04, "mem_gb_teacher": 16.04}
88
+ {"step": 69, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.04908341010484534, "tokens": 120000, "cumulative_loss_tokens": 8280000, "grad_norm": 0.2890625, "lr": 3e-05, "finish_rate": 0.76, "comp_len": 576.9, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.523, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 89.3, "frames": {"chat": 208}, "mem_gb": 16.1, "mem_gb_teacher": 16.1}
89
+ {"step": 70, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.043880132612407516, "tokens": 120000, "cumulative_loss_tokens": 8400000, "grad_norm": 0.267578125, "lr": 3e-05, "finish_rate": 0.763, "comp_len": 568.7, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.522, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 89.2, "frames": {"chat": 211}, "mem_gb": 16.07, "mem_gb_teacher": 16.07}
90
+ [eval step 70] sample: "To solve the given system of equations for \\(a\\), \\(b\\), \\(k\\), \\(m\\), and \\(p\\), we need to ensure that each letter represents a non-zero digit. Let's break down the problem step-by-step and solve it"
91
+ {"step": 71, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.04658220293604148, "tokens": 120000, "cumulative_loss_tokens": 8520000, "grad_norm": 0.302734375, "lr": 3e-05, "finish_rate": 0.806, "comp_len": 528.6, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.407, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 93.8, "frames": {"chat": 227}, "mem_gb": 16.05, "mem_gb_teacher": 16.05}
92
+ {"step": 72, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.04197514075435077, "tokens": 120000, "cumulative_loss_tokens": 8640000, "grad_norm": 0.275390625, "lr": 3e-05, "finish_rate": 0.796, "comp_len": 568.7, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.434, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 89.1, "frames": {"chat": 211}, "mem_gb": 16.03, "mem_gb_teacher": 16.03}
93
+ {"step": 73, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.03433373855294194, "tokens": 120000, "cumulative_loss_tokens": 8760000, "grad_norm": 0.248046875, "lr": 3e-05, "finish_rate": 0.861, "comp_len": 504.2, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.403, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 98.0, "frames": {"chat": 238}, "mem_gb": 16.04, "mem_gb_teacher": 16.04}
94
+ {"step": 74, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.035312130354298275, "tokens": 120000, "cumulative_loss_tokens": 8880000, "grad_norm": 0.2255859375, "lr": 3e-05, "finish_rate": 0.835, "comp_len": 506.3, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.517, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 98.2, "frames": {"chat": 237}, "mem_gb": 16.08, "mem_gb_teacher": 16.08}
95
+ {"step": 75, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.04418459653495035, "tokens": 120000, "cumulative_loss_tokens": 9000000, "grad_norm": 0.265625, "lr": 3e-05, "finish_rate": 0.721, "comp_len": 576.9, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.339, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 88.6, "frames": {"chat": 208}, "mem_gb": 16.08, "mem_gb_teacher": 16.08}
96
+ {"step": 76, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.03462567985369048, "tokens": 120000, "cumulative_loss_tokens": 9120000, "grad_norm": 0.212890625, "lr": 3e-05, "finish_rate": 0.801, "comp_len": 543.0, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.588, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 91.3, "frames": {"chat": 221}, "mem_gb": 16.17, "mem_gb_teacher": 16.17}
97
+ {"step": 77, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.0355419263230792, "tokens": 120000, "cumulative_loss_tokens": 9240000, "grad_norm": 0.244140625, "lr": 3e-05, "finish_rate": 0.853, "comp_len": 517.2, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.479, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 96.6, "frames": {"chat": 232}, "mem_gb": 16.01, "mem_gb_teacher": 16.01}
98
+ {"step": 78, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.03847126886160113, "tokens": 120000, "cumulative_loss_tokens": 9360000, "grad_norm": 0.25, "lr": 3e-05, "finish_rate": 0.764, "comp_len": 576.9, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.43, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 87.7, "frames": {"chat": 208}, "mem_gb": 16.04, "mem_gb_teacher": 16.04}
99
+ {"step": 79, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.03187043257508582, "tokens": 120000, "cumulative_loss_tokens": 9480000, "grad_norm": 0.232421875, "lr": 3e-05, "finish_rate": 0.837, "comp_len": 528.6, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.356, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 93.2, "frames": {"chat": 227}, "mem_gb": 15.96, "mem_gb_teacher": 15.96}
100
+ {"step": 80, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.03652716889477645, "tokens": 120000, "cumulative_loss_tokens": 9600000, "grad_norm": 0.251953125, "lr": 3e-05, "finish_rate": 0.824, "comp_len": 543.0, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.387, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 91.2, "frames": {"chat": 221}, "mem_gb": 15.99, "mem_gb_teacher": 15.99}
101
+ [eval step 80] sample: 'To solve the given system of equations for \\(a\\), \\(b\\), \\(k\\), \\(m\\), and \\(p\\), we need to find the values of these variables such that each letter represents a non-zero digit.\n\nThe equations are:\n1'
102
+ {"step": 81, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.03261745297779174, "tokens": 120000, "cumulative_loss_tokens": 9720000, "grad_norm": 0.2138671875, "lr": 3e-05, "finish_rate": 0.815, "comp_len": 517.2, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.457, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 95.5, "frames": {"chat": 232}, "mem_gb": 16.05, "mem_gb_teacher": 16.05}
103
+ {"step": 82, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.03819663266551991, "tokens": 120000, "cumulative_loss_tokens": 9840000, "grad_norm": 0.255859375, "lr": 3e-05, "finish_rate": 0.822, "comp_len": 547.9, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.423, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 90.2, "frames": {"chat": 219}, "mem_gb": 16.05, "mem_gb_teacher": 16.05}
104
+ {"step": 83, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.03785421463410991, "tokens": 120000, "cumulative_loss_tokens": 9960000, "grad_norm": 0.2265625, "lr": 3e-05, "finish_rate": 0.713, "comp_len": 615.4, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.399, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 82.7, "frames": {"chat": 195}, "mem_gb": 16.14, "mem_gb_teacher": 16.14}
105
+ {"step": 84, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.03711687127229913, "tokens": 120000, "cumulative_loss_tokens": 10080000, "grad_norm": 0.244140625, "lr": 3e-05, "finish_rate": 0.833, "comp_len": 555.6, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.503, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 88.9, "frames": {"chat": 216}, "mem_gb": 16.05, "mem_gb_teacher": 16.05}
106
+ {"step": 85, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.042023429202785095, "tokens": 120000, "cumulative_loss_tokens": 10200000, "grad_norm": 0.275390625, "lr": 3e-05, "finish_rate": 0.788, "comp_len": 576.9, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.37, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 87.7, "frames": {"chat": 208}, "mem_gb": 15.93, "mem_gb_teacher": 15.93}
107
+ {"step": 86, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.03288566896258077, "tokens": 120000, "cumulative_loss_tokens": 10320000, "grad_norm": 0.2265625, "lr": 3e-05, "finish_rate": 0.919, "comp_len": 510.6, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.407, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 95.6, "frames": {"chat": 235}, "mem_gb": 15.93, "mem_gb_teacher": 15.93}
108
+ {"step": 87, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.033694713056855834, "tokens": 120000, "cumulative_loss_tokens": 10440000, "grad_norm": 0.2578125, "lr": 3e-05, "finish_rate": 0.853, "comp_len": 533.3, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.391, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 91.6, "frames": {"chat": 225}, "mem_gb": 16.04, "mem_gb_teacher": 16.04}
109
+ {"step": 88, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.04422600561644261, "tokens": 120000, "cumulative_loss_tokens": 10560000, "grad_norm": 0.2451171875, "lr": 3e-05, "finish_rate": 0.77, "comp_len": 563.4, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.483, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 89.2, "frames": {"chat": 213}, "mem_gb": 16.13, "mem_gb_teacher": 16.13}
110
+ {"step": 89, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.029030231156169126, "tokens": 120000, "cumulative_loss_tokens": 10680000, "grad_norm": 0.2255859375, "lr": 3e-05, "finish_rate": 0.922, "comp_len": 466.9, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.372, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 104.0, "frames": {"chat": 257}, "mem_gb": 15.8, "mem_gb_teacher": 15.8}
111
+ {"step": 90, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.043001266495510934, "tokens": 120000, "cumulative_loss_tokens": 10800000, "grad_norm": 0.2578125, "lr": 3e-05, "finish_rate": 0.792, "comp_len": 566.0, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.497, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 89.1, "frames": {"chat": 212}, "mem_gb": 16.07, "mem_gb_teacher": 16.07}
112
+ [eval step 90] sample: 'To solve the given system of equations for \\(a\\), \\(b\\), \\(k\\), \\(m\\), and \\(p\\), we need to find the values of these variables such that each letter represents a non-zero digit.\n\nThe equations are:\n\\'
113
+ {"step": 91, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.03456530287990657, "tokens": 120000, "cumulative_loss_tokens": 10920000, "grad_norm": 0.263671875, "lr": 3e-05, "finish_rate": 0.833, "comp_len": 543.0, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.338, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 90.9, "frames": {"chat": 221}, "mem_gb": 16.05, "mem_gb_teacher": 16.05}
114
+ {"step": 92, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.030914492412268495, "tokens": 120000, "cumulative_loss_tokens": 11040000, "grad_norm": 0.203125, "lr": 3e-05, "finish_rate": 0.868, "comp_len": 495.9, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.442, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 98.2, "frames": {"chat": 242}, "mem_gb": 16.04, "mem_gb_teacher": 16.04}
115
+ {"step": 93, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.0329031198489829, "tokens": 120000, "cumulative_loss_tokens": 11160000, "grad_norm": 0.2216796875, "lr": 3e-05, "finish_rate": 0.836, "comp_len": 545.5, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.352, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 91.5, "frames": {"chat": 220}, "mem_gb": 16.01, "mem_gb_teacher": 16.01}
116
+ {"step": 94, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.029550562638859263, "tokens": 120000, "cumulative_loss_tokens": 11280000, "grad_norm": 0.2197265625, "lr": 3e-05, "finish_rate": 0.896, "comp_len": 500.0, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.291, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 96.4, "frames": {"chat": 240}, "mem_gb": 15.9, "mem_gb_teacher": 15.9}
117
+ {"step": 95, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.0336006456746875, "tokens": 120000, "cumulative_loss_tokens": 11400000, "grad_norm": 0.23046875, "lr": 3e-05, "finish_rate": 0.728, "comp_len": 582.5, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.359, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 87.0, "frames": {"chat": 206}, "mem_gb": 16.03, "mem_gb_teacher": 16.03}
118
+ {"step": 96, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.04226834708025368, "tokens": 120000, "cumulative_loss_tokens": 11520000, "grad_norm": 0.255859375, "lr": 3e-05, "finish_rate": 0.867, "comp_len": 531.0, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.402, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 92.9, "frames": {"chat": 226}, "mem_gb": 16.05, "mem_gb_teacher": 16.05}
119
+ {"step": 97, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.054896473065484314, "tokens": 120000, "cumulative_loss_tokens": 11640000, "grad_norm": 0.3046875, "lr": 3e-05, "finish_rate": 0.877, "comp_len": 491.8, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.382, "t_data_s": 0.1, "t_rollout_s": 0.0, "t_step_s": 99.8, "frames": {"chat": 244}, "mem_gb": 15.83, "mem_gb_teacher": 15.83}
120
+ {"step": 98, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.040829892701484884, "tokens": 120000, "cumulative_loss_tokens": 11760000, "grad_norm": 0.271484375, "lr": 3e-05, "finish_rate": 0.804, "comp_len": 535.7, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.406, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 92.6, "frames": {"chat": 224}, "mem_gb": 16.05, "mem_gb_teacher": 16.05}
121
+ {"step": 99, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.031381215056039705, "tokens": 120000, "cumulative_loss_tokens": 11880000, "grad_norm": 0.240234375, "lr": 3e-05, "finish_rate": 0.923, "comp_len": 442.8, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.319, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 109.0, "frames": {"chat": 271}, "mem_gb": 15.78, "mem_gb_teacher": 15.78}
122
+ {"step": 100, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.036837965581729075, "tokens": 120000, "cumulative_loss_tokens": 12000000, "grad_norm": 0.265625, "lr": 3e-05, "finish_rate": 0.856, "comp_len": 508.5, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.362, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 96.5, "frames": {"chat": 236}, "mem_gb": 16.06, "mem_gb_teacher": 16.06}
123
+ [eval step 100] sample: "To solve the given system of equations for \\(a\\), \\(b\\), \\(k\\), \\(m\\), and \\(p\\), we need to ensure that each letter represents a non-zero digit. Let's break down the problem step-by-step and solve it"
124
+ checkpoint snapshot queued -> outputs/healed/keep50_warmup_fixed_s1224/step0100
125
+ {"step": 101, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.03681235469069021, "tokens": 120000, "cumulative_loss_tokens": 12120000, "grad_norm": 0.244140625, "lr": 3e-05, "finish_rate": 0.841, "comp_len": 517.2, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.306, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 96.0, "frames": {"chat": 232}, "mem_gb": 15.93, "mem_gb_teacher": 15.93}
126
+ {"step": 102, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.03435125927827321, "tokens": 120000, "cumulative_loss_tokens": 12240000, "grad_norm": 0.251953125, "lr": 3e-05, "finish_rate": 0.79, "comp_len": 571.4, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.392, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 87.6, "frames": {"chat": 210}, "mem_gb": 15.98, "mem_gb_teacher": 15.98}
127
+ {"step": 103, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.033812837561887375, "tokens": 120000, "cumulative_loss_tokens": 12360000, "grad_norm": 0.275390625, "lr": 3e-05, "finish_rate": 0.811, "comp_len": 553.0, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.431, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 89.9, "frames": {"chat": 217}, "mem_gb": 15.95, "mem_gb_teacher": 15.95}
128
+ {"step": 104, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.03568466838332825, "tokens": 120000, "cumulative_loss_tokens": 12480000, "grad_norm": 0.25390625, "lr": 3e-05, "finish_rate": 0.839, "comp_len": 535.7, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.38, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 92.7, "frames": {"chat": 224}, "mem_gb": 16.07, "mem_gb_teacher": 16.07}
129
+ {"step": 105, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.04380085541328881, "tokens": 120000, "cumulative_loss_tokens": 12600000, "grad_norm": 0.287109375, "lr": 3e-05, "finish_rate": 0.749, "comp_len": 591.1, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.481, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 86.5, "frames": {"chat": 203}, "mem_gb": 15.92, "mem_gb_teacher": 15.92}
130
+ {"step": 106, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.0331037232719129, "tokens": 120000, "cumulative_loss_tokens": 12720000, "grad_norm": 0.2197265625, "lr": 3e-05, "finish_rate": 0.887, "comp_len": 502.1, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.326, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 97.2, "frames": {"chat": 239}, "mem_gb": 16.02, "mem_gb_teacher": 16.02}
131
+ {"step": 107, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.022136363052297384, "tokens": 120000, "cumulative_loss_tokens": 12840000, "grad_norm": 0.1875, "lr": 3e-05, "finish_rate": 0.902, "comp_len": 472.4, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.301, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 102.8, "frames": {"chat": 254}, "mem_gb": 15.93, "mem_gb_teacher": 15.93}
132
+ {"step": 108, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.02345696447007358, "tokens": 120000, "cumulative_loss_tokens": 12960000, "grad_norm": 0.21875, "lr": 3e-05, "finish_rate": 0.876, "comp_len": 497.9, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.421, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 98.5, "frames": {"chat": 241}, "mem_gb": 16.03, "mem_gb_teacher": 16.03}
133
+ {"step": 109, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.03220615579979494, "tokens": 120000, "cumulative_loss_tokens": 13080000, "grad_norm": 0.2294921875, "lr": 3e-05, "finish_rate": 0.746, "comp_len": 563.4, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.419, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 89.1, "frames": {"chat": 213}, "mem_gb": 16.05, "mem_gb_teacher": 16.05}
134
+ {"step": 110, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.028369275209781095, "tokens": 120000, "cumulative_loss_tokens": 13200000, "grad_norm": 0.205078125, "lr": 3e-05, "finish_rate": 0.864, "comp_len": 543.0, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.641, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 91.4, "frames": {"chat": 221}, "mem_gb": 16.1, "mem_gb_teacher": 16.1}
135
+ [eval step 110] sample: "To solve the given system of equations for \\(a\\), \\(b\\), \\(k\\), \\(m\\), and \\(p\\), we need to ensure that each letter represents a non-zero digit. Let's break down the problem step-by-step and solve it"
136
+ {"step": 111, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.030552633555675855, "tokens": 120000, "cumulative_loss_tokens": 13320000, "grad_norm": 0.201171875, "lr": 3e-05, "finish_rate": 0.745, "comp_len": 612.2, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.334, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 83.1, "frames": {"chat": 196}, "mem_gb": 16.06, "mem_gb_teacher": 16.06}
137
+ {"step": 112, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.02257079966662762, "tokens": 120000, "cumulative_loss_tokens": 13440000, "grad_norm": 0.1689453125, "lr": 3e-05, "finish_rate": 0.926, "comp_len": 444.4, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.427, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 108.7, "frames": {"chat": 270}, "mem_gb": 15.86, "mem_gb_teacher": 15.86}
138
+ {"step": 113, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.024793952927171875, "tokens": 120000, "cumulative_loss_tokens": 13560000, "grad_norm": 0.2041015625, "lr": 3e-05, "finish_rate": 0.815, "comp_len": 555.6, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.327, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 89.5, "frames": {"chat": 216}, "mem_gb": 16.04, "mem_gb_teacher": 16.04}
139
+ {"step": 114, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.02779970950682958, "tokens": 120000, "cumulative_loss_tokens": 13680000, "grad_norm": 0.1923828125, "lr": 3e-05, "finish_rate": 0.775, "comp_len": 600.0, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.314, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 83.9, "frames": {"chat": 200}, "mem_gb": 16.01, "mem_gb_teacher": 16.01}
140
+ {"step": 115, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.025435146312182768, "tokens": 120000, "cumulative_loss_tokens": 13800000, "grad_norm": 0.19921875, "lr": 3e-05, "finish_rate": 0.767, "comp_len": 582.5, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.406, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 86.4, "frames": {"chat": 206}, "mem_gb": 15.96, "mem_gb_teacher": 15.96}
141
+ {"step": 116, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.021958818318624982, "tokens": 120000, "cumulative_loss_tokens": 13920000, "grad_norm": 0.1650390625, "lr": 3e-05, "finish_rate": 0.902, "comp_len": 512.8, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.5, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 94.8, "frames": {"chat": 234}, "mem_gb": 15.99, "mem_gb_teacher": 15.99}
142
+ {"step": 117, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.024690080278979926, "tokens": 120000, "cumulative_loss_tokens": 14040000, "grad_norm": 0.171875, "lr": 3e-05, "finish_rate": 0.823, "comp_len": 558.1, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.325, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 88.4, "frames": {"chat": 215}, "mem_gb": 16.01, "mem_gb_teacher": 16.01}
143
+ {"step": 118, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.02221767870101612, "tokens": 120000, "cumulative_loss_tokens": 14160000, "grad_norm": 0.2021484375, "lr": 3e-05, "finish_rate": 0.922, "comp_len": 470.6, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.447, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 101.5, "frames": {"chat": 255}, "mem_gb": 15.99, "mem_gb_teacher": 15.99}
144
+ {"step": 119, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.021592201069470806, "tokens": 120000, "cumulative_loss_tokens": 14280000, "grad_norm": 0.1708984375, "lr": 3e-05, "finish_rate": 0.892, "comp_len": 480.0, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.377, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 101.8, "frames": {"chat": 250}, "mem_gb": 15.87, "mem_gb_teacher": 15.87}
145
+ {"step": 120, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.022036931684900386, "tokens": 120000, "cumulative_loss_tokens": 14400000, "grad_norm": 0.1826171875, "lr": 3e-05, "finish_rate": 0.884, "comp_len": 495.9, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.525, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 97.8, "frames": {"chat": 242}, "mem_gb": 16.04, "mem_gb_teacher": 16.04}
146
+ [eval step 120] sample: 'To solve the given system of equations for \\(a\\), \\(b\\), \\(k\\), \\(m\\), and \\(p\\), we need to find the values of these variables such that each letter represents a non-zero digit.\n\nThe equations are:\n\\'
147
+ {"step": 121, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.027385966173838823, "tokens": 120000, "cumulative_loss_tokens": 14520000, "grad_norm": 0.177734375, "lr": 3e-05, "finish_rate": 0.729, "comp_len": 603.0, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.517, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 85.1, "frames": {"chat": 199}, "mem_gb": 16.05, "mem_gb_teacher": 16.05}
148
+ {"step": 122, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.037581453007894255, "tokens": 120000, "cumulative_loss_tokens": 14640000, "grad_norm": 0.2138671875, "lr": 3e-05, "finish_rate": 0.784, "comp_len": 576.9, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.386, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 88.8, "frames": {"chat": 208}, "mem_gb": 16.08, "mem_gb_teacher": 16.08}
149
+ {"step": 123, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.02500725610067602, "tokens": 120000, "cumulative_loss_tokens": 14760000, "grad_norm": 0.212890625, "lr": 3e-05, "finish_rate": 0.764, "comp_len": 576.9, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.304, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 86.8, "frames": {"chat": 208}, "mem_gb": 16.02, "mem_gb_teacher": 16.02}
150
+ {"step": 124, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.029994825281358013, "tokens": 120000, "cumulative_loss_tokens": 14880000, "grad_norm": 0.2041015625, "lr": 3e-05, "finish_rate": 0.732, "comp_len": 574.2, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.473, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 89.1, "frames": {"chat": 209}, "mem_gb": 16.17, "mem_gb_teacher": 16.17}
151
+ {"step": 125, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.021332296466493667, "tokens": 120000, "cumulative_loss_tokens": 15000000, "grad_norm": 0.1728515625, "lr": 3e-05, "finish_rate": 0.855, "comp_len": 510.6, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.321, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 97.2, "frames": {"chat": 235}, "mem_gb": 16.0, "mem_gb_teacher": 16.0}
152
+ {"step": 126, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.02423879373523717, "tokens": 120000, "cumulative_loss_tokens": 15120000, "grad_norm": 0.2138671875, "lr": 3e-05, "finish_rate": 0.74, "comp_len": 588.2, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.503, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 85.5, "frames": {"chat": 204}, "mem_gb": 16.0, "mem_gb_teacher": 16.0}
153
+ {"step": 127, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.03143483543585365, "tokens": 120000, "cumulative_loss_tokens": 15240000, "grad_norm": 0.24609375, "lr": 3e-05, "finish_rate": 0.745, "comp_len": 576.9, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.498, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 88.5, "frames": {"chat": 208}, "mem_gb": 16.05, "mem_gb_teacher": 16.05}
154
+ {"step": 128, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.022677327596287555, "tokens": 120000, "cumulative_loss_tokens": 15360000, "grad_norm": 0.1748046875, "lr": 3e-05, "finish_rate": 0.825, "comp_len": 500.0, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.456, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 98.6, "frames": {"chat": 240}, "mem_gb": 16.05, "mem_gb_teacher": 16.05}
155
+ {"step": 129, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.02317165856284555, "tokens": 120000, "cumulative_loss_tokens": 15480000, "grad_norm": 0.18359375, "lr": 3e-05, "finish_rate": 0.89, "comp_len": 487.8, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.425, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 99.3, "frames": {"chat": 246}, "mem_gb": 16.04, "mem_gb_teacher": 16.04}
156
+ {"step": 130, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.02553549518882452, "tokens": 120000, "cumulative_loss_tokens": 15600000, "grad_norm": 0.16796875, "lr": 3e-05, "finish_rate": 0.909, "comp_len": 493.8, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.299, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 98.1, "frames": {"chat": 243}, "mem_gb": 15.86, "mem_gb_teacher": 15.86}
157
+ [eval step 130] sample: "To solve the given system of equations for \\(a\\), \\(b\\), \\(k\\), \\(m\\), and \\(p\\), we need to ensure that each letter represents a non-zero digit. Let's break down the problem step-by-step and solve it"
158
+ {"step": 131, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.03025520235665608, "tokens": 120000, "cumulative_loss_tokens": 15720000, "grad_norm": 0.193359375, "lr": 3e-05, "finish_rate": 0.745, "comp_len": 576.9, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.449, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 87.0, "frames": {"chat": 208}, "mem_gb": 16.06, "mem_gb_teacher": 16.06}
159
+ {"step": 132, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.033440696114464666, "tokens": 120000, "cumulative_loss_tokens": 15840000, "grad_norm": 0.20703125, "lr": 3e-05, "finish_rate": 0.817, "comp_len": 547.9, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.352, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 90.4, "frames": {"chat": 219}, "mem_gb": 16.05, "mem_gb_teacher": 16.05}
160
+ {"step": 133, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.030961406842339785, "tokens": 120000, "cumulative_loss_tokens": 15960000, "grad_norm": 0.205078125, "lr": 3e-05, "finish_rate": 0.782, "comp_len": 568.7, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.447, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 89.4, "frames": {"chat": 211}, "mem_gb": 16.06, "mem_gb_teacher": 16.06}
161
+ {"step": 134, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.030365196120288845, "tokens": 120000, "cumulative_loss_tokens": 16080000, "grad_norm": 0.25, "lr": 3e-05, "finish_rate": 0.862, "comp_len": 517.2, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.283, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 95.3, "frames": {"chat": 232}, "mem_gb": 16.02, "mem_gb_teacher": 16.02}
162
+ {"step": 135, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.03637079153128434, "tokens": 120000, "cumulative_loss_tokens": 16200000, "grad_norm": 0.201171875, "lr": 3e-05, "finish_rate": 0.804, "comp_len": 560.7, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.403, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 90.4, "frames": {"chat": 214}, "mem_gb": 16.05, "mem_gb_teacher": 16.05}
163
+ {"step": 136, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.024788761290698312, "tokens": 120000, "cumulative_loss_tokens": 16320000, "grad_norm": 0.173828125, "lr": 3e-05, "finish_rate": 0.819, "comp_len": 531.0, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.395, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 94.4, "frames": {"chat": 226}, "mem_gb": 15.94, "mem_gb_teacher": 15.94}
164
+ {"step": 137, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.02415202263732596, "tokens": 120000, "cumulative_loss_tokens": 16440000, "grad_norm": 0.171875, "lr": 3e-05, "finish_rate": 0.819, "comp_len": 571.4, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.404, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 88.7, "frames": {"chat": 210}, "mem_gb": 16.06, "mem_gb_teacher": 16.06}
165
+ {"step": 138, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.023906889412474507, "tokens": 120000, "cumulative_loss_tokens": 16560000, "grad_norm": 0.1884765625, "lr": 3e-05, "finish_rate": 0.858, "comp_len": 550.5, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.436, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 91.6, "frames": {"chat": 218}, "mem_gb": 15.88, "mem_gb_teacher": 15.88}
166
+ {"step": 139, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.02281373731412459, "tokens": 120000, "cumulative_loss_tokens": 16680000, "grad_norm": 0.185546875, "lr": 3e-05, "finish_rate": 0.858, "comp_len": 515.0, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.354, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 96.0, "frames": {"chat": 233}, "mem_gb": 16.04, "mem_gb_teacher": 16.04}
167
+ {"step": 140, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.029389094779423128, "tokens": 120000, "cumulative_loss_tokens": 16800000, "grad_norm": 0.1962890625, "lr": 3e-05, "finish_rate": 0.786, "comp_len": 558.1, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.221, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 91.1, "frames": {"chat": 215}, "mem_gb": 16.05, "mem_gb_teacher": 16.05}
168
+ [eval step 140] sample: 'To solve the given system of equations with the constraint that each letter represents a non-zero digit, we need to find the values of \\(a\\), \\(b\\), \\(k\\), \\(m\\), and \\(p\\) that satisfy all the equati'
169
+ {"step": 141, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.02496639485837271, "tokens": 120000, "cumulative_loss_tokens": 16920000, "grad_norm": 0.1826171875, "lr": 3e-05, "finish_rate": 0.845, "comp_len": 515.0, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.428, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 95.0, "frames": {"chat": 233}, "mem_gb": 16.04, "mem_gb_teacher": 16.04}
170
+ {"step": 142, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.024896287856064736, "tokens": 120000, "cumulative_loss_tokens": 17040000, "grad_norm": 0.2138671875, "lr": 3e-05, "finish_rate": 0.766, "comp_len": 574.2, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.353, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 87.5, "frames": {"chat": 209}, "mem_gb": 15.99, "mem_gb_teacher": 15.99}
171
+ {"step": 143, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.0202058710468933, "tokens": 120000, "cumulative_loss_tokens": 17160000, "grad_norm": 0.166015625, "lr": 3e-05, "finish_rate": 0.908, "comp_len": 458.0, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.308, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 105.8, "frames": {"chat": 262}, "mem_gb": 15.92, "mem_gb_teacher": 15.92}
172
+ {"step": 144, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.0228134301078273, "tokens": 120000, "cumulative_loss_tokens": 17280000, "grad_norm": 0.171875, "lr": 3e-05, "finish_rate": 0.9, "comp_len": 481.9, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.241, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 101.0, "frames": {"chat": 249}, "mem_gb": 16.01, "mem_gb_teacher": 16.01}
173
+ {"step": 145, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.030403959429240787, "tokens": 120000, "cumulative_loss_tokens": 17400000, "grad_norm": 0.18359375, "lr": 3e-05, "finish_rate": 0.819, "comp_len": 528.6, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.364, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 94.0, "frames": {"chat": 227}, "mem_gb": 16.04, "mem_gb_teacher": 16.04}
174
+ {"step": 146, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.022969909678919553, "tokens": 120000, "cumulative_loss_tokens": 17520000, "grad_norm": 0.1796875, "lr": 3e-05, "finish_rate": 0.814, "comp_len": 543.0, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.356, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 91.2, "frames": {"chat": 221}, "mem_gb": 16.04, "mem_gb_teacher": 16.04}
175
+ {"step": 147, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.02228280872052225, "tokens": 120000, "cumulative_loss_tokens": 17640000, "grad_norm": 0.171875, "lr": 3e-05, "finish_rate": 0.859, "comp_len": 512.8, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.473, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 95.0, "frames": {"chat": 234}, "mem_gb": 16.06, "mem_gb_teacher": 16.06}
176
+ {"step": 148, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.023852225604599032, "tokens": 120000, "cumulative_loss_tokens": 17760000, "grad_norm": 0.212890625, "lr": 3e-05, "finish_rate": 0.817, "comp_len": 563.4, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.29, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 87.8, "frames": {"chat": 213}, "mem_gb": 16.0, "mem_gb_teacher": 16.0}
177
+ {"step": 149, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.022865247969034438, "tokens": 120000, "cumulative_loss_tokens": 17880000, "grad_norm": 0.1748046875, "lr": 3e-05, "finish_rate": 0.836, "comp_len": 563.4, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.455, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 88.1, "frames": {"chat": 213}, "mem_gb": 15.94, "mem_gb_teacher": 15.94}
178
+ {"step": 150, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.021284172641811892, "tokens": 120000, "cumulative_loss_tokens": 18000000, "grad_norm": 0.1669921875, "lr": 3e-05, "finish_rate": 0.906, "comp_len": 512.8, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.392, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 95.1, "frames": {"chat": 234}, "mem_gb": 15.97, "mem_gb_teacher": 15.97}
179
+ [eval step 150] sample: "To solve the given system of equations for \\(a\\), \\(b\\), \\(k\\), \\(m\\), and \\(p\\), we need to find the values of these variables such that each letter represents a non-zero digit.\n\nLet's break down the"
180
+ checkpoint snapshot queued -> outputs/healed/keep50_warmup_fixed_s1224/step0150
181
+ wandb: updating run metadata
182
+ wandb: uploading output.log; uploading wandb-summary.json; uploading config.yaml
183
+ wandb:
184
+ wandb: Run history:
185
+ wandb: comp_len ▆▃▆▃▄▄▆▃▃▅▆▅▄█▅▄▄▆▅▆▅▅▂▃▄▃▆▅▇▃▁▇▆▄▂▃▅▅▂▅
186
+ wandb: cumulative_loss_tokens ▁▁▁▂▂▂▂▃▃▃▃▄▄▄▄▄▄▄▄▄▅▅▅▆▆▆▆▇▇▇▇▇▇▇▇▇████
187
+ wandb: dropped_truncated ▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁
188
+ wandb: epoch ▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▅▅▅▅▅▅▅▅▅▅▅▅▅██████████
189
+ wandb: finish_rate ▅▄▂▃▆▆▅▃▇▅█▆▆▇█▂▄▄█▆▃▅▆▄▅▃▂▇▇▁▂▇█▂▆▄▆▅█▄
190
+ wandb: forward_topk_kl █▆▄▄▃▃▄▂▂▃▂▂▂▂▂▂▂▁▂▂▁▁▂▁▁▂▂▁▁▁▁▁▁▁▁▁▁▁▁▁
191
+ wandb: grad_norm █▂▂▂▂▂▂▂▂▁▂▁▂▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁
192
+ wandb: lr ▁███████████████████████████████████████
193
+ wandb: mem_gb ▃▆▆▄▂▄▇▆▇▃▆▆▆▅▇▆▅▅▆█▆█▆▅▆▁▄▆▆▅▃▆▅▃▆▆▄▅▆▅
194
+ wandb: mem_gb_teacher ▅▄▆▇▆▂▆▆▆▇▆▅▃█▆▆▆▅▆▃▆▄▂▃▆█▄▄█▁▆▇▂▆▅▅▆▆▆▃
195
+ wandb: +6 ...
196
+ wandb:
197
+ wandb: Run summary:
198
+ wandb: comp_len 512.8
199
+ wandb: cumulative_loss_tokens 18000000
200
+ wandb: dropped_truncated 0
201
+ wandb: epoch 2
202
+ wandb: finish_rate 0.906
203
+ wandb: forward_topk_kl 0.02128
204
+ wandb: grad_norm 0.16699
205
+ wandb: lr 3e-05
206
+ wandb: mem_gb 15.97
207
+ wandb: mem_gb_teacher 15.97
208
+ wandb: +7 ...
209
+ wandb:
210
+ wandb: 🚀 View run warmup-fixed-keep50-s1224 at: https://wandb.ai/hbfreed/glean-heal/runs/sep45w2j
211
+ wandb: ⭐️ View project at: https://wandb.ai/hbfreed/glean-heal
212
+ wandb: Synced 5 W&B file(s), 0 media file(s), 0 artifact file(s) and 0 other file(s)
213
+ wandb: Find logs at: outputs/healed/keep50_warmup_fixed_s1224/wandb/run-20260802_025702-sep45w2j/logs
healed/warmup_gate.log ADDED
@@ -0,0 +1,26 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ expert max|delta| = 1.524e-03
2
+
3
+ Let's break down the problem step by step:
4
+
5
+ 1. Natalia sold 48 clips in April.
6
+ 2. In May, she sold half as many clips as in April. So, the number of clips sold in May is \( \frac{48}{2} = 24 \).
7
+
8
+ To find the total number of clips sold in both months, we add the number of clips sold in April and May:
9
+
10
+ \[ 48 + 24 = 72 \]
11
+
12
+ Thus, Natalia sold a total of 72 clips in April and May.
13
+
14
+ Let's confirm this with Python code.
15
+ ```python
16
+ # Number of clips sold in April
17
+ clips_April = 48
18
+
19
+ # Number of clips sold in May (half of April)
20
+ clips_May = clips_April / 2
21
+
22
+ # Total number of clips sold in April and May
23
+ total_clips = clips_April + clips_May
24
+ print(total_clips)
25
+
26
+ GATE PASSED
healed/warmup_keep50.log ADDED
@@ -0,0 +1,213 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ wandb: [wandb.login()] Loaded credentials for https://api.wandb.ai from /home/henry/.netrc.
2
+ wandb: Currently logged in as: hbfreed to https://api.wandb.ai. Use `wandb login --relogin` to force relogin
3
+ wandb: Tracking run with wandb version 0.28.0
4
+ wandb: Run data is saved locally in outputs/healed/keep50_offpolicy_warmup_s1224/wandb/run-20260801_230924-9td6b5cn
5
+ wandb: Run `wandb offline` to turn off syncing.
6
+ wandb: Syncing run offpolicy-warmup-keep50-s1224
7
+ wandb: ⭐️ View project at https://wandb.ai/hbfreed/glean-heal
8
+ wandb: 🚀 View run at https://wandb.ai/hbfreed/glean-heal/runs/9td6b5cn
9
+
10
+ 12115 cached top-128 chat trajectories / 6,476,634 unique tokens | 53 steps/epoch | 150 total steps | student params 3.70B | teacher overlap=False
11
+ {"step": 1, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.21866693885562322, "tokens": 120000, "cumulative_loss_tokens": 120000, "grad_norm": 0.703125, "lr": 6e-06, "finish_rate": 0.907, "comp_len": 508.5, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.337, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 27.4, "frames": {"chat": 236}, "mem_gb": 9.77, "mem_gb_teacher": 9.77}
12
+ The attention mask is not set and cannot be inferred from input because pad token is same as eos token. As a consequence, you may observe unexpected behavior. Please pass your input's `attention_mask` to obtain reliable results.
13
+ [eval step 1] sample: 'To solve the given system of equations:\n\\[\n\\begin{align*}\na + b &= k, \\\\\nk + m &= p, \\\\\np + a &= r, \\\\\nb + m + r &= 18,\n\\end{align*}\n\\]\nwe need to determine the values'
14
+ {"step": 2, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.27999618121907116, "tokens": 120000, "cumulative_loss_tokens": 240000, "grad_norm": 0.80078125, "lr": 9e-06, "finish_rate": 0.781, "comp_len": 558.1, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.474, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 25.1, "frames": {"chat": 215}, "mem_gb": 9.96, "mem_gb_teacher": 9.96}
15
+ {"step": 3, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.3080657049433639, "tokens": 120000, "cumulative_loss_tokens": 360000, "grad_norm": 0.89453125, "lr": 1.2e-05, "finish_rate": 0.825, "comp_len": 553.0, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.376, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 25.3, "frames": {"chat": 217}, "mem_gb": 9.83, "mem_gb_teacher": 9.83}
16
+ {"step": 4, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.27968517109975216, "tokens": 120000, "cumulative_loss_tokens": 480000, "grad_norm": 0.80078125, "lr": 1.5e-05, "finish_rate": 0.8, "comp_len": 585.4, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.366, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 24.6, "frames": {"chat": 205}, "mem_gb": 9.89, "mem_gb_teacher": 9.89}
17
+ {"step": 5, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.2137425679458926, "tokens": 120000, "cumulative_loss_tokens": 600000, "grad_norm": 0.62890625, "lr": 1.8e-05, "finish_rate": 0.834, "comp_len": 524.0, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.318, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 25.8, "frames": {"chat": 229}, "mem_gb": 9.86, "mem_gb_teacher": 9.86}
18
+ {"step": 6, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.34967619865822297, "tokens": 120000, "cumulative_loss_tokens": 720000, "grad_norm": 1.3125, "lr": 2.1e-05, "finish_rate": 0.812, "comp_len": 538.1, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.336, "t_data_s": 0.1, "t_rollout_s": 0.0, "t_step_s": 25.4, "frames": {"chat": 223}, "mem_gb": 9.93, "mem_gb_teacher": 9.93}
19
+ {"step": 7, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.23986013823635877, "tokens": 120000, "cumulative_loss_tokens": 840000, "grad_norm": 0.5546875, "lr": 2.4e-05, "finish_rate": 0.708, "comp_len": 594.1, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.42, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 25.6, "frames": {"chat": 202}, "mem_gb": 9.97, "mem_gb_teacher": 9.97}
20
+ {"step": 8, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.3001840955584, "tokens": 120000, "cumulative_loss_tokens": 960000, "grad_norm": 0.78515625, "lr": 2.7000000000000002e-05, "finish_rate": 0.77, "comp_len": 574.2, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.453, "t_data_s": 0.1, "t_rollout_s": 0.0, "t_step_s": 24.9, "frames": {"chat": 209}, "mem_gb": 9.94, "mem_gb_teacher": 9.94}
21
+ {"step": 9, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.21187542014177888, "tokens": 120000, "cumulative_loss_tokens": 1080000, "grad_norm": 0.57421875, "lr": 3e-05, "finish_rate": 0.885, "comp_len": 528.6, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.397, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 25.7, "frames": {"chat": 227}, "mem_gb": 9.92, "mem_gb_teacher": 9.92}
22
+ {"step": 10, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.21670206268125525, "tokens": 120000, "cumulative_loss_tokens": 1200000, "grad_norm": 0.6328125, "lr": 3e-05, "finish_rate": 0.848, "comp_len": 521.7, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.546, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 27.0, "frames": {"chat": 230}, "mem_gb": 9.99, "mem_gb_teacher": 9.99}
23
+ [eval step 10] sample: "To solve this problem, we need to find the values of \\(a\\), \\(b\\), \\(k\\), \\(m\\), and \\(p\\) that satisfy the given equations. Let's break down the problem step-by-step:\n\n1. **Understand the Equations:*"
24
+ {"step": 11, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.21374244357372324, "tokens": 120000, "cumulative_loss_tokens": 1320000, "grad_norm": 0.52734375, "lr": 3e-05, "finish_rate": 0.879, "comp_len": 519.5, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.426, "t_data_s": 0.1, "t_rollout_s": 0.0, "t_step_s": 26.0, "frames": {"chat": 231}, "mem_gb": 9.84, "mem_gb_teacher": 9.84}
25
+ {"step": 12, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.23642759951651096, "tokens": 120000, "cumulative_loss_tokens": 1440000, "grad_norm": 0.625, "lr": 3e-05, "finish_rate": 0.882, "comp_len": 489.8, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.246, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 26.3, "frames": {"chat": 245}, "mem_gb": 9.92, "mem_gb_teacher": 9.92}
26
+ {"step": 13, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.2684279385884603, "tokens": 120000, "cumulative_loss_tokens": 1560000, "grad_norm": 0.73828125, "lr": 3e-05, "finish_rate": 0.81, "comp_len": 571.4, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.375, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 24.7, "frames": {"chat": 210}, "mem_gb": 9.92, "mem_gb_teacher": 9.92}
27
+ {"step": 14, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.2720188724226008, "tokens": 120000, "cumulative_loss_tokens": 1680000, "grad_norm": 0.69140625, "lr": 3e-05, "finish_rate": 0.758, "comp_len": 568.7, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.312, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 26.3, "frames": {"chat": 211}, "mem_gb": 9.92, "mem_gb_teacher": 9.92}
28
+ {"step": 15, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.2650659183566769, "tokens": 120000, "cumulative_loss_tokens": 1800000, "grad_norm": 0.56640625, "lr": 3e-05, "finish_rate": 0.819, "comp_len": 543.0, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.413, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 26.1, "frames": {"chat": 221}, "mem_gb": 9.98, "mem_gb_teacher": 9.98}
29
+ {"step": 16, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.22639379921114694, "tokens": 120000, "cumulative_loss_tokens": 1920000, "grad_norm": 0.609375, "lr": 3e-05, "finish_rate": 0.912, "comp_len": 480.0, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.366, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 27.1, "frames": {"chat": 250}, "mem_gb": 9.79, "mem_gb_teacher": 9.79}
30
+ {"step": 17, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.2573891951084137, "tokens": 120000, "cumulative_loss_tokens": 2040000, "grad_norm": 0.796875, "lr": 3e-05, "finish_rate": 0.79, "comp_len": 524.0, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.402, "t_data_s": 0.1, "t_rollout_s": 0.0, "t_step_s": 26.9, "frames": {"chat": 229}, "mem_gb": 9.96, "mem_gb_teacher": 9.96}
31
+ {"step": 18, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.22218937695200244, "tokens": 120000, "cumulative_loss_tokens": 2160000, "grad_norm": 0.77734375, "lr": 3e-05, "finish_rate": 0.888, "comp_len": 480.0, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.479, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 26.7, "frames": {"chat": 250}, "mem_gb": 9.94, "mem_gb_teacher": 9.94}
32
+ {"step": 19, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.33289732446968556, "tokens": 120000, "cumulative_loss_tokens": 2280000, "grad_norm": 2.546875, "lr": 3e-05, "finish_rate": 0.844, "comp_len": 519.5, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.363, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 26.6, "frames": {"chat": 231}, "mem_gb": 9.81, "mem_gb_teacher": 9.81}
33
+ {"step": 20, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.327336692000553, "tokens": 120000, "cumulative_loss_tokens": 2400000, "grad_norm": 1.3984375, "lr": 3e-05, "finish_rate": 0.844, "comp_len": 535.7, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.433, "t_data_s": 0.1, "t_rollout_s": 0.0, "t_step_s": 25.4, "frames": {"chat": 224}, "mem_gb": 9.86, "mem_gb_teacher": 9.86}
34
+ [eval step 20] sample: '```python\ndef f(a, b, c, d, e, f, g, h):\n ax + by + cz + ey + fx + gy + hz = 0\n ```\n\nWe need to find the values of \\(a\\), \\(b\\), \\(c'
35
+ {"step": 21, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.33612352709385257, "tokens": 120000, "cumulative_loss_tokens": 2520000, "grad_norm": 1.640625, "lr": 3e-05, "finish_rate": 0.802, "comp_len": 566.0, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.329, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 24.9, "frames": {"chat": 212}, "mem_gb": 9.9, "mem_gb_teacher": 9.9}
36
+ {"step": 22, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.2883151387684047, "tokens": 120000, "cumulative_loss_tokens": 2640000, "grad_norm": 1.3984375, "lr": 3e-05, "finish_rate": 0.87, "comp_len": 504.2, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.376, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 25.9, "frames": {"chat": 238}, "mem_gb": 9.85, "mem_gb_teacher": 9.85}
37
+ {"step": 23, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.25279294178610046, "tokens": 120000, "cumulative_loss_tokens": 2760000, "grad_norm": 1.3828125, "lr": 3e-05, "finish_rate": 0.903, "comp_len": 466.9, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.445, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 26.9, "frames": {"chat": 257}, "mem_gb": 9.73, "mem_gb_teacher": 9.73}
38
+ {"step": 24, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.2615377183983723, "tokens": 120000, "cumulative_loss_tokens": 2880000, "grad_norm": 1.5625, "lr": 3e-05, "finish_rate": 0.868, "comp_len": 528.6, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.405, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 25.7, "frames": {"chat": 227}, "mem_gb": 9.92, "mem_gb_teacher": 9.92}
39
+ {"step": 25, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.32016109869256615, "tokens": 120000, "cumulative_loss_tokens": 3000000, "grad_norm": 2.546875, "lr": 3e-05, "finish_rate": 0.838, "comp_len": 526.3, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.359, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 25.8, "frames": {"chat": 228}, "mem_gb": 9.95, "mem_gb_teacher": 9.95}
40
+ {"step": 26, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.34886745701755084, "tokens": 120000, "cumulative_loss_tokens": 3120000, "grad_norm": 2.0625, "lr": 3e-05, "finish_rate": 0.803, "comp_len": 515.0, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.542, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 27.1, "frames": {"chat": 233}, "mem_gb": 9.94, "mem_gb_teacher": 9.94}
41
+ {"step": 27, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.3173991178593288, "tokens": 120000, "cumulative_loss_tokens": 3240000, "grad_norm": 2.671875, "lr": 3e-05, "finish_rate": 0.863, "comp_len": 515.0, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.351, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 26.3, "frames": {"chat": 233}, "mem_gb": 9.94, "mem_gb_teacher": 9.94}
42
+ {"step": 28, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.3661775833528489, "tokens": 120000, "cumulative_loss_tokens": 3360000, "grad_norm": 2.984375, "lr": 3e-05, "finish_rate": 0.731, "comp_len": 609.1, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.388, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 26.3, "frames": {"chat": 197}, "mem_gb": 10.03, "mem_gb_teacher": 10.03}
43
+ {"step": 29, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.36429248579877116, "tokens": 120000, "cumulative_loss_tokens": 3480000, "grad_norm": 5.625, "lr": 3e-05, "finish_rate": 0.862, "comp_len": 502.1, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.384, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 27.0, "frames": {"chat": 239}, "mem_gb": 9.79, "mem_gb_teacher": 9.79}
44
+ {"step": 30, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.3706551998923222, "tokens": 120000, "cumulative_loss_tokens": 3600000, "grad_norm": 4.34375, "lr": 3e-05, "finish_rate": 0.83, "comp_len": 535.7, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.302, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 25.1, "frames": {"chat": 224}, "mem_gb": 9.83, "mem_gb_teacher": 9.83}
45
+ [eval step 30] sample: 'To solve the problem, we need to determine the value of \\( p \\) given the equations:\n\n1. \\( a + b = k \\)\n2. \\( k + m = p \\)\n3. \\( p + a = r \\)\n4. \\( b + m + r ='
46
+ {"step": 31, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.35791126018886765, "tokens": 120000, "cumulative_loss_tokens": 3720000, "grad_norm": 2.03125, "lr": 3e-05, "finish_rate": 0.788, "comp_len": 553.0, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.388, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 25.3, "frames": {"chat": 217}, "mem_gb": 9.95, "mem_gb_teacher": 9.95}
47
+ {"step": 32, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.3755841711225609, "tokens": 120000, "cumulative_loss_tokens": 3840000, "grad_norm": 2.3125, "lr": 3e-05, "finish_rate": 0.876, "comp_len": 497.9, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.259, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 26.2, "frames": {"chat": 241}, "mem_gb": 9.95, "mem_gb_teacher": 9.95}
48
+ {"step": 33, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.3561015821622064, "tokens": 120000, "cumulative_loss_tokens": 3960000, "grad_norm": 2.359375, "lr": 3e-05, "finish_rate": 0.835, "comp_len": 550.5, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.407, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 24.9, "frames": {"chat": 218}, "mem_gb": 9.92, "mem_gb_teacher": 9.92}
49
+ {"step": 34, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.38959850126380724, "tokens": 120000, "cumulative_loss_tokens": 4080000, "grad_norm": 2.125, "lr": 3e-05, "finish_rate": 0.767, "comp_len": 582.5, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.439, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 24.9, "frames": {"chat": 206}, "mem_gb": 9.93, "mem_gb_teacher": 9.93}
50
+ {"step": 35, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.3970217802577963, "tokens": 120000, "cumulative_loss_tokens": 4200000, "grad_norm": 2.734375, "lr": 3e-05, "finish_rate": 0.845, "comp_len": 517.2, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.512, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 27.0, "frames": {"chat": 232}, "mem_gb": 9.97, "mem_gb_teacher": 9.97}
51
+ {"step": 36, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.4204790102675557, "tokens": 120000, "cumulative_loss_tokens": 4320000, "grad_norm": 1.796875, "lr": 3e-05, "finish_rate": 0.771, "comp_len": 550.5, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.432, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 27.6, "frames": {"chat": 218}, "mem_gb": 9.99, "mem_gb_teacher": 9.99}
52
+ {"step": 37, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.40933242611338694, "tokens": 120000, "cumulative_loss_tokens": 4440000, "grad_norm": 1.6015625, "lr": 3e-05, "finish_rate": 0.779, "comp_len": 563.4, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.428, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 25.9, "frames": {"chat": 213}, "mem_gb": 9.95, "mem_gb_teacher": 9.95}
53
+ {"step": 38, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.3901712652951479, "tokens": 120000, "cumulative_loss_tokens": 4560000, "grad_norm": 2.640625, "lr": 3e-05, "finish_rate": 0.887, "comp_len": 483.9, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.527, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 27.5, "frames": {"chat": 248}, "mem_gb": 9.92, "mem_gb_teacher": 9.92}
54
+ {"step": 39, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.3685633979354054, "tokens": 120000, "cumulative_loss_tokens": 4680000, "grad_norm": 1.921875, "lr": 3e-05, "finish_rate": 0.803, "comp_len": 550.5, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.333, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 25.2, "frames": {"chat": 218}, "mem_gb": 9.99, "mem_gb_teacher": 9.99}
55
+ {"step": 40, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.39544522463083265, "tokens": 120000, "cumulative_loss_tokens": 4800000, "grad_norm": 2.3125, "lr": 3e-05, "finish_rate": 0.851, "comp_len": 543.0, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.401, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 25.7, "frames": {"chat": 221}, "mem_gb": 9.94, "mem_gb_teacher": 9.94}
56
+ [eval step 40] sample: "To solve the problem, we need to determine the value of \\(p\\) given the constraints. Let's break down the problem step-by-step:\n\n1. **Define Variables:**\n - Let \\(d\\) be the digit \\(0, 1, 2, \\ldots,"
57
+ {"step": 41, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.38834569306795796, "tokens": 120000, "cumulative_loss_tokens": 4920000, "grad_norm": 1.640625, "lr": 3e-05, "finish_rate": 0.894, "comp_len": 508.5, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.396, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 25.5, "frames": {"chat": 236}, "mem_gb": 9.87, "mem_gb_teacher": 9.87}
58
+ {"step": 42, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.40595541520963113, "tokens": 120000, "cumulative_loss_tokens": 5040000, "grad_norm": 2.390625, "lr": 3e-05, "finish_rate": 0.858, "comp_len": 487.8, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.338, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 27.6, "frames": {"chat": 246}, "mem_gb": 9.79, "mem_gb_teacher": 9.79}
59
+ {"step": 43, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.4013322958761205, "tokens": 120000, "cumulative_loss_tokens": 5160000, "grad_norm": 2.703125, "lr": 3e-05, "finish_rate": 0.838, "comp_len": 512.8, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.48, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 26.9, "frames": {"chat": 234}, "mem_gb": 10.05, "mem_gb_teacher": 10.05}
60
+ {"step": 44, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.41921593125065165, "tokens": 120000, "cumulative_loss_tokens": 5280000, "grad_norm": 2.984375, "lr": 3e-05, "finish_rate": 0.748, "comp_len": 594.1, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.474, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 24.7, "frames": {"chat": 202}, "mem_gb": 9.93, "mem_gb_teacher": 9.93}
61
+ {"step": 45, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.45117179917966327, "tokens": 120000, "cumulative_loss_tokens": 5400000, "grad_norm": 1.953125, "lr": 3e-05, "finish_rate": 0.811, "comp_len": 553.0, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.36, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 25.2, "frames": {"chat": 217}, "mem_gb": 9.94, "mem_gb_teacher": 9.94}
62
+ {"step": 46, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.45246317840516564, "tokens": 120000, "cumulative_loss_tokens": 5520000, "grad_norm": 4.0625, "lr": 3e-05, "finish_rate": 0.866, "comp_len": 535.7, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.47, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 25.4, "frames": {"chat": 224}, "mem_gb": 9.94, "mem_gb_teacher": 9.94}
63
+ {"step": 47, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.49074055876086153, "tokens": 120000, "cumulative_loss_tokens": 5640000, "grad_norm": 2.265625, "lr": 3e-05, "finish_rate": 0.753, "comp_len": 558.1, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.285, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 25.2, "frames": {"chat": 215}, "mem_gb": 9.96, "mem_gb_teacher": 9.96}
64
+ {"step": 48, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.502812573158741, "tokens": 120000, "cumulative_loss_tokens": 5760000, "grad_norm": 4.78125, "lr": 3e-05, "finish_rate": 0.884, "comp_len": 463.3, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.356, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 26.6, "frames": {"chat": 259}, "mem_gb": 9.87, "mem_gb_teacher": 9.87}
65
+ {"step": 49, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.4940076053115229, "tokens": 120000, "cumulative_loss_tokens": 5880000, "grad_norm": 2.875, "lr": 3e-05, "finish_rate": 0.829, "comp_len": 571.4, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.475, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 24.5, "frames": {"chat": 210}, "mem_gb": 9.94, "mem_gb_teacher": 9.94}
66
+ {"step": 50, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.47275001460264127, "tokens": 120000, "cumulative_loss_tokens": 6000000, "grad_norm": 3.296875, "lr": 3e-05, "finish_rate": 0.77, "comp_len": 563.4, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.304, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 26.3, "frames": {"chat": 213}, "mem_gb": 9.99, "mem_gb_teacher": 9.99}
67
+ [eval step 50] sample: 'To solve the problem, we need to determine the values of \\(p\\), \\(k\\), \\(r\\), and \\(m\\).\n\nGiven:\n1. \\(a + b = k\\)\n2. \\(k + m = p\\)\n3. \\(p + a = r\\)\n'
68
+ checkpoint snapshot queued -> outputs/healed/keep50_offpolicy_warmup_s1224/step0050
69
+ {"step": 51, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.4224179199380179, "tokens": 120000, "cumulative_loss_tokens": 6120000, "grad_norm": 1.75, "lr": 3e-05, "finish_rate": 0.815, "comp_len": 540.5, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.453, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 25.1, "frames": {"chat": 222}, "mem_gb": 9.9, "mem_gb_teacher": 9.9}
70
+ {"step": 52, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.4378351664955417, "tokens": 120000, "cumulative_loss_tokens": 6240000, "grad_norm": 2.421875, "lr": 3e-05, "finish_rate": 0.889, "comp_len": 510.6, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.439, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 25.7, "frames": {"chat": 235}, "mem_gb": 9.96, "mem_gb_teacher": 9.96}
71
+ {"step": 53, "epoch": 0, "training_mode": "off-policy", "forward_topk_kl": 0.44709739099716145, "tokens": 120000, "cumulative_loss_tokens": 6360000, "grad_norm": 6.625, "lr": 3e-05, "finish_rate": 0.798, "comp_len": 576.9, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.417, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 24.4, "frames": {"chat": 208}, "mem_gb": 9.91, "mem_gb_teacher": 9.91}
72
+ {"step": 54, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.4153756284924845, "tokens": 120000, "cumulative_loss_tokens": 6480000, "grad_norm": 5.4375, "lr": 3e-05, "finish_rate": 0.733, "comp_len": 628.3, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.413, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 23.5, "frames": {"chat": 191}, "mem_gb": 9.95, "mem_gb_teacher": 9.95}
73
+ {"step": 55, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.40591654521947107, "tokens": 120000, "cumulative_loss_tokens": 6600000, "grad_norm": 5.75, "lr": 3e-05, "finish_rate": 0.845, "comp_len": 547.9, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.523, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 25.1, "frames": {"chat": 219}, "mem_gb": 9.95, "mem_gb_teacher": 9.95}
74
+ {"step": 56, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.43598461368133623, "tokens": 120000, "cumulative_loss_tokens": 6720000, "grad_norm": 7.0, "lr": 3e-05, "finish_rate": 0.778, "comp_len": 579.7, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.459, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 24.8, "frames": {"chat": 207}, "mem_gb": 9.95, "mem_gb_teacher": 9.95}
75
+ {"step": 57, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.4643517450052003, "tokens": 120000, "cumulative_loss_tokens": 6840000, "grad_norm": 5.53125, "lr": 3e-05, "finish_rate": 0.755, "comp_len": 576.9, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.449, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 25.4, "frames": {"chat": 208}, "mem_gb": 9.91, "mem_gb_teacher": 9.91}
76
+ {"step": 58, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.41939549018144606, "tokens": 120000, "cumulative_loss_tokens": 6960000, "grad_norm": 26.125, "lr": 3e-05, "finish_rate": 0.799, "comp_len": 547.9, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.288, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 25.7, "frames": {"chat": 219}, "mem_gb": 9.94, "mem_gb_teacher": 9.94}
77
+ {"step": 59, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.44048424391622343, "tokens": 120000, "cumulative_loss_tokens": 7080000, "grad_norm": 58.5, "lr": 3e-05, "finish_rate": 0.915, "comp_len": 487.8, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.329, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 26.8, "frames": {"chat": 246}, "mem_gb": 9.82, "mem_gb_teacher": 9.82}
78
+ {"step": 60, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.4594272028216471, "tokens": 120000, "cumulative_loss_tokens": 7200000, "grad_norm": 46.0, "lr": 3e-05, "finish_rate": 0.704, "comp_len": 582.5, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.437, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 26.3, "frames": {"chat": 206}, "mem_gb": 9.97, "mem_gb_teacher": 9.97}
79
+ [eval step 60] sample: "To solve this problem, we need to determine the values of \\(p\\), \\(k\\), and \\(r\\) based on the given equations. Here's the step-by-step approach:\n\n1. **Understand the Given Equations:**\n \\[\n \\begi"
80
+ {"step": 61, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.4421705829419196, "tokens": 120000, "cumulative_loss_tokens": 7320000, "grad_norm": 5.1875, "lr": 3e-05, "finish_rate": 0.876, "comp_len": 515.0, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.375, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 26.2, "frames": {"chat": 233}, "mem_gb": 9.95, "mem_gb_teacher": 9.95}
81
+ {"step": 62, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.5114135434468587, "tokens": 120000, "cumulative_loss_tokens": 7440000, "grad_norm": 20.375, "lr": 3e-05, "finish_rate": 0.847, "comp_len": 524.0, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.363, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 25.4, "frames": {"chat": 229}, "mem_gb": 9.82, "mem_gb_teacher": 9.82}
82
+ {"step": 63, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.4925492258039614, "tokens": 120000, "cumulative_loss_tokens": 7560000, "grad_norm": 22.0, "lr": 3e-05, "finish_rate": 0.864, "comp_len": 508.5, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.285, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 25.7, "frames": {"chat": 236}, "mem_gb": 9.85, "mem_gb_teacher": 9.85}
83
+ {"step": 64, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.4923259206386904, "tokens": 120000, "cumulative_loss_tokens": 7680000, "grad_norm": 21.875, "lr": 3e-05, "finish_rate": 0.87, "comp_len": 502.1, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.37, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 27.0, "frames": {"chat": 239}, "mem_gb": 9.74, "mem_gb_teacher": 9.74}
84
+ {"step": 65, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.41652837800706427, "tokens": 120000, "cumulative_loss_tokens": 7800000, "grad_norm": 8.875, "lr": 3e-05, "finish_rate": 0.867, "comp_len": 497.9, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.349, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 26.8, "frames": {"chat": 241}, "mem_gb": 9.86, "mem_gb_teacher": 9.86}
85
+ {"step": 66, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.3965861890381823, "tokens": 120000, "cumulative_loss_tokens": 7920000, "grad_norm": 3.484375, "lr": 3e-05, "finish_rate": 0.863, "comp_len": 531.0, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.44, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 26.1, "frames": {"chat": 226}, "mem_gb": 9.82, "mem_gb_teacher": 9.82}
86
+ {"step": 67, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.36972922986969353, "tokens": 120000, "cumulative_loss_tokens": 8040000, "grad_norm": 17.0, "lr": 3e-05, "finish_rate": 0.893, "comp_len": 512.8, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.32, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 26.0, "frames": {"chat": 234}, "mem_gb": 9.95, "mem_gb_teacher": 9.95}
87
+ {"step": 68, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.374694859992216, "tokens": 120000, "cumulative_loss_tokens": 8160000, "grad_norm": 21.125, "lr": 3e-05, "finish_rate": 0.914, "comp_len": 466.9, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.401, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 26.2, "frames": {"chat": 257}, "mem_gb": 9.94, "mem_gb_teacher": 9.94}
88
+ {"step": 69, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.4051088455612461, "tokens": 120000, "cumulative_loss_tokens": 8280000, "grad_norm": 7.3125, "lr": 3e-05, "finish_rate": 0.76, "comp_len": 576.9, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.523, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 27.1, "frames": {"chat": 208}, "mem_gb": 10.0, "mem_gb_teacher": 10.0}
89
+ {"step": 70, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.4079768711109956, "tokens": 120000, "cumulative_loss_tokens": 8400000, "grad_norm": 2.046875, "lr": 3e-05, "finish_rate": 0.763, "comp_len": 568.7, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.522, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 26.0, "frames": {"chat": 211}, "mem_gb": 9.97, "mem_gb_teacher": 9.97}
90
+ [eval step 70] sample: 'To determine the value of \\(p\\), we need to follow the steps outlined in the the problem:\n\n1. **Understand the Problem:**\n - \\(a + b = k\\)\n - \\(k + m = p\\)\n - \\(p + a = r\\)\n -'
91
+ {"step": 71, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.3891906266813477, "tokens": 120000, "cumulative_loss_tokens": 8520000, "grad_norm": 3.484375, "lr": 3e-05, "finish_rate": 0.806, "comp_len": 528.6, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.407, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 26.1, "frames": {"chat": 227}, "mem_gb": 9.95, "mem_gb_teacher": 9.95}
92
+ {"step": 72, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.4033234915149709, "tokens": 120000, "cumulative_loss_tokens": 8640000, "grad_norm": 5.03125, "lr": 3e-05, "finish_rate": 0.796, "comp_len": 568.7, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.434, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 26.2, "frames": {"chat": 211}, "mem_gb": 9.93, "mem_gb_teacher": 9.93}
93
+ {"step": 73, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.3784339854914695, "tokens": 120000, "cumulative_loss_tokens": 8760000, "grad_norm": 4.0, "lr": 3e-05, "finish_rate": 0.861, "comp_len": 504.2, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.403, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 26.6, "frames": {"chat": 238}, "mem_gb": 9.94, "mem_gb_teacher": 9.94}
94
+ {"step": 74, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.37151155275255443, "tokens": 120000, "cumulative_loss_tokens": 8880000, "grad_norm": 1.3515625, "lr": 3e-05, "finish_rate": 0.835, "comp_len": 506.3, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.517, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 27.8, "frames": {"chat": 237}, "mem_gb": 9.99, "mem_gb_teacher": 9.99}
95
+ {"step": 75, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.40700901261599115, "tokens": 120000, "cumulative_loss_tokens": 9000000, "grad_norm": 7.09375, "lr": 3e-05, "finish_rate": 0.721, "comp_len": 576.9, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.339, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 26.5, "frames": {"chat": 208}, "mem_gb": 9.99, "mem_gb_teacher": 9.99}
96
+ {"step": 76, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.3612343382894993, "tokens": 120000, "cumulative_loss_tokens": 9120000, "grad_norm": 6.28125, "lr": 3e-05, "finish_rate": 0.801, "comp_len": 543.0, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.588, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 25.5, "frames": {"chat": 221}, "mem_gb": 10.07, "mem_gb_teacher": 10.07}
97
+ {"step": 77, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.3821119113404304, "tokens": 120000, "cumulative_loss_tokens": 9240000, "grad_norm": 6.90625, "lr": 3e-05, "finish_rate": 0.853, "comp_len": 517.2, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.479, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 26.9, "frames": {"chat": 232}, "mem_gb": 9.91, "mem_gb_teacher": 9.91}
98
+ {"step": 78, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.3988335593829552, "tokens": 120000, "cumulative_loss_tokens": 9360000, "grad_norm": 1.796875, "lr": 3e-05, "finish_rate": 0.764, "comp_len": 576.9, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.43, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 25.7, "frames": {"chat": 208}, "mem_gb": 9.94, "mem_gb_teacher": 9.94}
99
+ {"step": 79, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.36860200558168194, "tokens": 120000, "cumulative_loss_tokens": 9480000, "grad_norm": 3.09375, "lr": 3e-05, "finish_rate": 0.837, "comp_len": 528.6, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.356, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 25.7, "frames": {"chat": 227}, "mem_gb": 9.86, "mem_gb_teacher": 9.86}
100
+ {"step": 80, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.39240696791845064, "tokens": 120000, "cumulative_loss_tokens": 9600000, "grad_norm": 1.9453125, "lr": 3e-05, "finish_rate": 0.824, "comp_len": 543.0, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.387, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 25.5, "frames": {"chat": 221}, "mem_gb": 9.89, "mem_gb_teacher": 9.89}
101
+ [eval step 80] sample: "To solve this problem, we need to determine the values of \\(p\\), \\(r\\), and \\(m\\) based on the given equations. Here's the step-by-step approach:\n\n1. **Understand the Given Equations:**\n \\[\n \\begi"
102
+ {"step": 81, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.38983683857706686, "tokens": 120000, "cumulative_loss_tokens": 9720000, "grad_norm": 1.796875, "lr": 3e-05, "finish_rate": 0.815, "comp_len": 517.2, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.457, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 25.8, "frames": {"chat": 232}, "mem_gb": 9.96, "mem_gb_teacher": 9.96}
103
+ {"step": 82, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.3839597153416524, "tokens": 120000, "cumulative_loss_tokens": 9840000, "grad_norm": 1.390625, "lr": 3e-05, "finish_rate": 0.822, "comp_len": 547.9, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.423, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 25.4, "frames": {"chat": 219}, "mem_gb": 9.96, "mem_gb_teacher": 9.96}
104
+ {"step": 83, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.40005449274579685, "tokens": 120000, "cumulative_loss_tokens": 9960000, "grad_norm": 1.015625, "lr": 3e-05, "finish_rate": 0.713, "comp_len": 615.4, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.399, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 24.8, "frames": {"chat": 195}, "mem_gb": 10.05, "mem_gb_teacher": 10.05}
105
+ {"step": 84, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.37729036068630717, "tokens": 120000, "cumulative_loss_tokens": 10080000, "grad_norm": 2.421875, "lr": 3e-05, "finish_rate": 0.833, "comp_len": 555.6, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.503, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 24.9, "frames": {"chat": 216}, "mem_gb": 9.95, "mem_gb_teacher": 9.95}
106
+ {"step": 85, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.39155043416718643, "tokens": 120000, "cumulative_loss_tokens": 10200000, "grad_norm": 1.640625, "lr": 3e-05, "finish_rate": 0.788, "comp_len": 576.9, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.37, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 24.9, "frames": {"chat": 208}, "mem_gb": 9.84, "mem_gb_teacher": 9.84}
107
+ {"step": 86, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.3371589448125412, "tokens": 120000, "cumulative_loss_tokens": 10320000, "grad_norm": 1.625, "lr": 3e-05, "finish_rate": 0.919, "comp_len": 510.6, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.407, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 25.5, "frames": {"chat": 235}, "mem_gb": 9.83, "mem_gb_teacher": 9.83}
108
+ {"step": 87, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.3228179054065297, "tokens": 120000, "cumulative_loss_tokens": 10440000, "grad_norm": 1.015625, "lr": 3e-05, "finish_rate": 0.853, "comp_len": 533.3, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.391, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 24.9, "frames": {"chat": 225}, "mem_gb": 9.94, "mem_gb_teacher": 9.94}
109
+ {"step": 88, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.3611420020165543, "tokens": 120000, "cumulative_loss_tokens": 10560000, "grad_norm": 2.078125, "lr": 3e-05, "finish_rate": 0.77, "comp_len": 563.4, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.483, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 26.3, "frames": {"chat": 213}, "mem_gb": 10.03, "mem_gb_teacher": 10.03}
110
+ {"step": 89, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.3108938364227613, "tokens": 120000, "cumulative_loss_tokens": 10680000, "grad_norm": 1.25, "lr": 3e-05, "finish_rate": 0.922, "comp_len": 466.9, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.372, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 27.4, "frames": {"chat": 257}, "mem_gb": 9.71, "mem_gb_teacher": 9.71}
111
+ {"step": 90, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.3567947539317111, "tokens": 120000, "cumulative_loss_tokens": 10800000, "grad_norm": 2.515625, "lr": 3e-05, "finish_rate": 0.792, "comp_len": 566.0, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.497, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 25.9, "frames": {"chat": 212}, "mem_gb": 9.98, "mem_gb_teacher": 9.98}
112
+ [eval step 90] sample: 'To determine the value of \\( p \\), we need to follow the steps outlined in thepy:\n\n1. **Understand the Problem:**\n - \\( p \\) is the value of \\( k \\) when \\( k + m = 18 \\).\n - \\( k \\) is the value'
113
+ {"step": 91, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.35618535781440636, "tokens": 120000, "cumulative_loss_tokens": 10920000, "grad_norm": 2.65625, "lr": 3e-05, "finish_rate": 0.833, "comp_len": 543.0, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.338, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 25.2, "frames": {"chat": 221}, "mem_gb": 9.95, "mem_gb_teacher": 9.95}
114
+ {"step": 92, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.3354658261674146, "tokens": 120000, "cumulative_loss_tokens": 11040000, "grad_norm": 2.390625, "lr": 3e-05, "finish_rate": 0.868, "comp_len": 495.9, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.442, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 26.0, "frames": {"chat": 242}, "mem_gb": 9.95, "mem_gb_teacher": 9.95}
115
+ {"step": 93, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.3363398057249685, "tokens": 120000, "cumulative_loss_tokens": 11160000, "grad_norm": 1.53125, "lr": 3e-05, "finish_rate": 0.836, "comp_len": 545.5, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.352, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 25.6, "frames": {"chat": 220}, "mem_gb": 9.91, "mem_gb_teacher": 9.91}
116
+ {"step": 94, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.30306674740935363, "tokens": 120000, "cumulative_loss_tokens": 11280000, "grad_norm": 0.7734375, "lr": 3e-05, "finish_rate": 0.896, "comp_len": 500.0, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.291, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 25.5, "frames": {"chat": 240}, "mem_gb": 9.81, "mem_gb_teacher": 9.81}
117
+ {"step": 95, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.29799054561704397, "tokens": 120000, "cumulative_loss_tokens": 11400000, "grad_norm": 1.40625, "lr": 3e-05, "finish_rate": 0.728, "comp_len": 582.5, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.359, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 24.8, "frames": {"chat": 206}, "mem_gb": 9.94, "mem_gb_teacher": 9.94}
118
+ {"step": 96, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.3275977665552249, "tokens": 120000, "cumulative_loss_tokens": 11520000, "grad_norm": 1.4921875, "lr": 3e-05, "finish_rate": 0.867, "comp_len": 531.0, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.402, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 25.6, "frames": {"chat": 226}, "mem_gb": 9.95, "mem_gb_teacher": 9.95}
119
+ {"step": 97, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.34486291259291274, "tokens": 120000, "cumulative_loss_tokens": 11640000, "grad_norm": 1.25, "lr": 3e-05, "finish_rate": 0.877, "comp_len": 491.8, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.382, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 26.8, "frames": {"chat": 244}, "mem_gb": 9.74, "mem_gb_teacher": 9.74}
120
+ {"step": 98, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.31863660610305766, "tokens": 120000, "cumulative_loss_tokens": 11760000, "grad_norm": 1.09375, "lr": 3e-05, "finish_rate": 0.804, "comp_len": 535.7, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.406, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 25.9, "frames": {"chat": 224}, "mem_gb": 9.95, "mem_gb_teacher": 9.95}
121
+ {"step": 99, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.313539165522034, "tokens": 120000, "cumulative_loss_tokens": 11880000, "grad_norm": 0.99609375, "lr": 3e-05, "finish_rate": 0.923, "comp_len": 442.8, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.319, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 27.9, "frames": {"chat": 271}, "mem_gb": 9.68, "mem_gb_teacher": 9.68}
122
+ {"step": 100, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.30199915543012323, "tokens": 120000, "cumulative_loss_tokens": 12000000, "grad_norm": 0.80078125, "lr": 3e-05, "finish_rate": 0.856, "comp_len": 508.5, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.362, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 26.5, "frames": {"chat": 236}, "mem_gb": 9.96, "mem_gb_teacher": 9.96}
123
+ [eval step 100] sample: "To solve the problem, we need to determine the values of \\(p\\) and \\(k\\) given the constraints. Let's break down the problem step-by-step:\n\n1. **Understand the Constraints:**\n - Each letter represen"
124
+ checkpoint snapshot queued -> outputs/healed/keep50_offpolicy_warmup_s1224/step0100
125
+ {"step": 101, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.30484250679599745, "tokens": 120000, "cumulative_loss_tokens": 12120000, "grad_norm": 0.75, "lr": 3e-05, "finish_rate": 0.841, "comp_len": 517.2, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.306, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 26.5, "frames": {"chat": 232}, "mem_gb": 9.83, "mem_gb_teacher": 9.83}
126
+ {"step": 102, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.30012307230867447, "tokens": 120000, "cumulative_loss_tokens": 12240000, "grad_norm": 1.1640625, "lr": 3e-05, "finish_rate": 0.79, "comp_len": 571.4, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.392, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 25.3, "frames": {"chat": 210}, "mem_gb": 9.89, "mem_gb_teacher": 9.89}
127
+ {"step": 103, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.2981501909478257, "tokens": 120000, "cumulative_loss_tokens": 12360000, "grad_norm": 1.2421875, "lr": 3e-05, "finish_rate": 0.811, "comp_len": 553.0, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.431, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 25.0, "frames": {"chat": 217}, "mem_gb": 9.85, "mem_gb_teacher": 9.85}
128
+ {"step": 104, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.29460206581093373, "tokens": 120000, "cumulative_loss_tokens": 12480000, "grad_norm": 0.83203125, "lr": 3e-05, "finish_rate": 0.839, "comp_len": 535.7, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.38, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 26.1, "frames": {"chat": 224}, "mem_gb": 9.97, "mem_gb_teacher": 9.97}
129
+ {"step": 105, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.3155513667286684, "tokens": 120000, "cumulative_loss_tokens": 12600000, "grad_norm": 0.83203125, "lr": 3e-05, "finish_rate": 0.749, "comp_len": 591.1, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.481, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 26.3, "frames": {"chat": 203}, "mem_gb": 9.82, "mem_gb_teacher": 9.82}
130
+ {"step": 106, "epoch": 1, "training_mode": "off-policy", "forward_topk_kl": 0.2852651763110111, "tokens": 120000, "cumulative_loss_tokens": 12720000, "grad_norm": 0.75, "lr": 3e-05, "finish_rate": 0.887, "comp_len": 502.1, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.326, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 25.8, "frames": {"chat": 239}, "mem_gb": 9.92, "mem_gb_teacher": 9.92}
131
+ {"step": 107, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.2750922793724885, "tokens": 120000, "cumulative_loss_tokens": 12840000, "grad_norm": 0.93359375, "lr": 3e-05, "finish_rate": 0.902, "comp_len": 472.4, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.301, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 26.9, "frames": {"chat": 254}, "mem_gb": 9.83, "mem_gb_teacher": 9.83}
132
+ {"step": 108, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.27038282493477067, "tokens": 120000, "cumulative_loss_tokens": 12960000, "grad_norm": 0.91796875, "lr": 3e-05, "finish_rate": 0.876, "comp_len": 497.9, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.421, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 26.4, "frames": {"chat": 241}, "mem_gb": 9.93, "mem_gb_teacher": 9.93}
133
+ {"step": 109, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.30530283329064645, "tokens": 120000, "cumulative_loss_tokens": 13080000, "grad_norm": 0.8671875, "lr": 3e-05, "finish_rate": 0.746, "comp_len": 563.4, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.419, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 26.2, "frames": {"chat": 213}, "mem_gb": 9.95, "mem_gb_teacher": 9.95}
134
+ {"step": 110, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.2827595801195751, "tokens": 120000, "cumulative_loss_tokens": 13200000, "grad_norm": 0.78125, "lr": 3e-05, "finish_rate": 0.864, "comp_len": 543.0, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.641, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 25.6, "frames": {"chat": 221}, "mem_gb": 10.0, "mem_gb_teacher": 10.0}
135
+ [eval step 110] sample: "To solve the problem, we need to determine the digits \\(p\\), \\(a\\), and \\(b\\) that satisfy the given conditions. Let's break down the problem step-by-step:\n\n1. **Understand the Constraints:**\n - \\(p"
136
+ {"step": 111, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.3139219396378845, "tokens": 120000, "cumulative_loss_tokens": 13320000, "grad_norm": 0.9375, "lr": 3e-05, "finish_rate": 0.745, "comp_len": 612.2, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.334, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 24.6, "frames": {"chat": 196}, "mem_gb": 9.96, "mem_gb_teacher": 9.96}
137
+ {"step": 112, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.29290350563563405, "tokens": 120000, "cumulative_loss_tokens": 13440000, "grad_norm": 1.2109375, "lr": 3e-05, "finish_rate": 0.926, "comp_len": 444.4, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.427, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 28.5, "frames": {"chat": 270}, "mem_gb": 9.77, "mem_gb_teacher": 9.77}
138
+ {"step": 113, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.2911341037095835, "tokens": 120000, "cumulative_loss_tokens": 13560000, "grad_norm": 1.2578125, "lr": 3e-05, "finish_rate": 0.815, "comp_len": 555.6, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.327, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 25.3, "frames": {"chat": 216}, "mem_gb": 9.94, "mem_gb_teacher": 9.94}
139
+ {"step": 114, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.30757556294202804, "tokens": 120000, "cumulative_loss_tokens": 13680000, "grad_norm": 0.97265625, "lr": 3e-05, "finish_rate": 0.775, "comp_len": 600.0, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.314, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 24.2, "frames": {"chat": 200}, "mem_gb": 9.91, "mem_gb_teacher": 9.91}
140
+ {"step": 115, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.27265539040267467, "tokens": 120000, "cumulative_loss_tokens": 13800000, "grad_norm": 1.046875, "lr": 3e-05, "finish_rate": 0.767, "comp_len": 582.5, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.406, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 24.9, "frames": {"chat": 206}, "mem_gb": 9.86, "mem_gb_teacher": 9.86}
141
+ {"step": 116, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.25661204309028884, "tokens": 120000, "cumulative_loss_tokens": 13920000, "grad_norm": 1.0703125, "lr": 3e-05, "finish_rate": 0.902, "comp_len": 512.8, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.5, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 25.4, "frames": {"chat": 234}, "mem_gb": 9.9, "mem_gb_teacher": 9.9}
142
+ {"step": 117, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.28587267751296364, "tokens": 120000, "cumulative_loss_tokens": 14040000, "grad_norm": 1.078125, "lr": 3e-05, "finish_rate": 0.823, "comp_len": 558.1, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.325, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 24.5, "frames": {"chat": 215}, "mem_gb": 9.91, "mem_gb_teacher": 9.91}
143
+ {"step": 118, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.24581487802788615, "tokens": 120000, "cumulative_loss_tokens": 14160000, "grad_norm": 0.8515625, "lr": 3e-05, "finish_rate": 0.922, "comp_len": 470.6, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.447, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 26.0, "frames": {"chat": 255}, "mem_gb": 9.89, "mem_gb_teacher": 9.89}
144
+ {"step": 119, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.2630173706655701, "tokens": 120000, "cumulative_loss_tokens": 14280000, "grad_norm": 1.0390625, "lr": 3e-05, "finish_rate": 0.892, "comp_len": 480.0, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.377, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 26.9, "frames": {"chat": 250}, "mem_gb": 9.77, "mem_gb_teacher": 9.77}
145
+ {"step": 120, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.26195900368392466, "tokens": 120000, "cumulative_loss_tokens": 14400000, "grad_norm": 0.85546875, "lr": 3e-05, "finish_rate": 0.884, "comp_len": 495.9, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.525, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 25.8, "frames": {"chat": 242}, "mem_gb": 9.94, "mem_gb_teacher": 9.94}
146
+ [eval step 120] sample: "To solve this problem, we need to determine the values of \\(p\\) and \\(k\\) based on the given equations. Let's break down the problem step-by-step:\n\n1. **Understand the Equations:**\n - \\(a + b = k\\)\n"
147
+ {"step": 121, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.27887726591676476, "tokens": 120000, "cumulative_loss_tokens": 14520000, "grad_norm": 0.98828125, "lr": 3e-05, "finish_rate": 0.729, "comp_len": 603.0, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.517, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 25.5, "frames": {"chat": 199}, "mem_gb": 9.95, "mem_gb_teacher": 9.95}
148
+ {"step": 122, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.31573768441453576, "tokens": 120000, "cumulative_loss_tokens": 14640000, "grad_norm": 1.15625, "lr": 3e-05, "finish_rate": 0.784, "comp_len": 576.9, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.386, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 26.7, "frames": {"chat": 208}, "mem_gb": 9.99, "mem_gb_teacher": 9.99}
149
+ {"step": 123, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.2738492341738194, "tokens": 120000, "cumulative_loss_tokens": 14760000, "grad_norm": 0.859375, "lr": 3e-05, "finish_rate": 0.764, "comp_len": 576.9, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.304, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 24.6, "frames": {"chat": 208}, "mem_gb": 9.92, "mem_gb_teacher": 9.92}
150
+ {"step": 124, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.3032253828023871, "tokens": 120000, "cumulative_loss_tokens": 14880000, "grad_norm": 1.21875, "lr": 3e-05, "finish_rate": 0.732, "comp_len": 574.2, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.473, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 26.8, "frames": {"chat": 209}, "mem_gb": 10.07, "mem_gb_teacher": 10.07}
151
+ {"step": 125, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.27386389810865125, "tokens": 120000, "cumulative_loss_tokens": 15000000, "grad_norm": 1.5859375, "lr": 3e-05, "finish_rate": 0.855, "comp_len": 510.6, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.321, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 26.6, "frames": {"chat": 235}, "mem_gb": 9.91, "mem_gb_teacher": 9.91}
152
+ {"step": 126, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.2811740072357158, "tokens": 120000, "cumulative_loss_tokens": 15120000, "grad_norm": 1.4453125, "lr": 3e-05, "finish_rate": 0.74, "comp_len": 588.2, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.503, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 24.8, "frames": {"chat": 204}, "mem_gb": 9.9, "mem_gb_teacher": 9.9}
153
+ {"step": 127, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.32506899852765103, "tokens": 120000, "cumulative_loss_tokens": 15240000, "grad_norm": 2.453125, "lr": 3e-05, "finish_rate": 0.745, "comp_len": 576.9, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.498, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 26.8, "frames": {"chat": 208}, "mem_gb": 9.95, "mem_gb_teacher": 9.95}
154
+ {"step": 128, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.27416869887411593, "tokens": 120000, "cumulative_loss_tokens": 15360000, "grad_norm": 1.375, "lr": 3e-05, "finish_rate": 0.825, "comp_len": 500.0, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.456, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 27.2, "frames": {"chat": 240}, "mem_gb": 9.95, "mem_gb_teacher": 9.95}
155
+ {"step": 129, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.27367745394359033, "tokens": 120000, "cumulative_loss_tokens": 15480000, "grad_norm": 1.390625, "lr": 3e-05, "finish_rate": 0.89, "comp_len": 487.8, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.425, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 26.4, "frames": {"chat": 246}, "mem_gb": 9.94, "mem_gb_teacher": 9.94}
156
+ {"step": 130, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.27279284177869556, "tokens": 120000, "cumulative_loss_tokens": 15600000, "grad_norm": 1.1171875, "lr": 3e-05, "finish_rate": 0.909, "comp_len": 493.8, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.299, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 26.1, "frames": {"chat": 243}, "mem_gb": 9.76, "mem_gb_teacher": 9.76}
157
+ [eval step 130] sample: "To solve the problem, we need to determine the values of \\(p\\), \\(r\\), and \\(k\\) given the constraints. Let's break down the problem step-by-step:\n\n1. **Understand the Constraints:**\n - \\(a + b = k\\"
158
+ {"step": 131, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.2835182323958725, "tokens": 120000, "cumulative_loss_tokens": 15720000, "grad_norm": 0.953125, "lr": 3e-05, "finish_rate": 0.745, "comp_len": 576.9, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.449, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 24.8, "frames": {"chat": 208}, "mem_gb": 9.96, "mem_gb_teacher": 9.96}
159
+ {"step": 132, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.2755484388658156, "tokens": 120000, "cumulative_loss_tokens": 15840000, "grad_norm": 0.8203125, "lr": 3e-05, "finish_rate": 0.817, "comp_len": 547.9, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.352, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 25.3, "frames": {"chat": 219}, "mem_gb": 9.95, "mem_gb_teacher": 9.95}
160
+ {"step": 133, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.27247595444793504, "tokens": 120000, "cumulative_loss_tokens": 15960000, "grad_norm": 0.796875, "lr": 3e-05, "finish_rate": 0.782, "comp_len": 568.7, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.447, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 26.3, "frames": {"chat": 211}, "mem_gb": 9.96, "mem_gb_teacher": 9.96}
161
+ {"step": 134, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.252491025553147, "tokens": 120000, "cumulative_loss_tokens": 16080000, "grad_norm": 0.67578125, "lr": 3e-05, "finish_rate": 0.862, "comp_len": 517.2, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.283, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 26.0, "frames": {"chat": 232}, "mem_gb": 9.92, "mem_gb_teacher": 9.92}
162
+ {"step": 135, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.2650780773670723, "tokens": 120000, "cumulative_loss_tokens": 16200000, "grad_norm": 0.71875, "lr": 3e-05, "finish_rate": 0.804, "comp_len": 560.7, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.403, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 25.3, "frames": {"chat": 214}, "mem_gb": 9.96, "mem_gb_teacher": 9.96}
163
+ {"step": 136, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.2518176361516118, "tokens": 120000, "cumulative_loss_tokens": 16320000, "grad_norm": 0.890625, "lr": 3e-05, "finish_rate": 0.819, "comp_len": 531.0, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.395, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 25.8, "frames": {"chat": 226}, "mem_gb": 9.85, "mem_gb_teacher": 9.85}
164
+ {"step": 137, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.23503943474429348, "tokens": 120000, "cumulative_loss_tokens": 16440000, "grad_norm": 0.75, "lr": 3e-05, "finish_rate": 0.819, "comp_len": 571.4, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.404, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 25.4, "frames": {"chat": 210}, "mem_gb": 9.96, "mem_gb_teacher": 9.96}
165
+ {"step": 138, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.2334536761138588, "tokens": 120000, "cumulative_loss_tokens": 16560000, "grad_norm": 0.6796875, "lr": 3e-05, "finish_rate": 0.858, "comp_len": 550.5, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.436, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 25.4, "frames": {"chat": 218}, "mem_gb": 9.78, "mem_gb_teacher": 9.78}
166
+ {"step": 139, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.2328599476976941, "tokens": 120000, "cumulative_loss_tokens": 16680000, "grad_norm": 0.67578125, "lr": 3e-05, "finish_rate": 0.858, "comp_len": 515.0, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.354, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 26.5, "frames": {"chat": 233}, "mem_gb": 9.94, "mem_gb_teacher": 9.94}
167
+ {"step": 140, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.26084051485359666, "tokens": 120000, "cumulative_loss_tokens": 16800000, "grad_norm": 0.70703125, "lr": 3e-05, "finish_rate": 0.786, "comp_len": 558.1, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.221, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 27.0, "frames": {"chat": 215}, "mem_gb": 9.95, "mem_gb_teacher": 9.95}
168
+ [eval step 140] sample: "To solve this problem, we need to determine the values of \\(p\\), \\(a\\), \\(b\\), \\(m\\), and \\(r\\) that satisfy the given equations. Let's break down the problem step-by-step:\n\n1. **Understand the Equati"
169
+ {"step": 141, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.26040056609710055, "tokens": 120000, "cumulative_loss_tokens": 16920000, "grad_norm": 0.83203125, "lr": 3e-05, "finish_rate": 0.845, "comp_len": 515.0, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.428, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 25.3, "frames": {"chat": 233}, "mem_gb": 9.94, "mem_gb_teacher": 9.94}
170
+ {"step": 142, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.2547849820467333, "tokens": 120000, "cumulative_loss_tokens": 17040000, "grad_norm": 0.91015625, "lr": 3e-05, "finish_rate": 0.766, "comp_len": 574.2, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.353, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 25.2, "frames": {"chat": 209}, "mem_gb": 9.89, "mem_gb_teacher": 9.89}
171
+ {"step": 143, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.2485178017048786, "tokens": 120000, "cumulative_loss_tokens": 17160000, "grad_norm": 0.83203125, "lr": 3e-05, "finish_rate": 0.908, "comp_len": 458.0, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.308, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 27.5, "frames": {"chat": 262}, "mem_gb": 9.82, "mem_gb_teacher": 9.82}
172
+ {"step": 144, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.25363729545498886, "tokens": 120000, "cumulative_loss_tokens": 17280000, "grad_norm": 0.7578125, "lr": 3e-05, "finish_rate": 0.9, "comp_len": 481.9, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.241, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 26.7, "frames": {"chat": 249}, "mem_gb": 9.91, "mem_gb_teacher": 9.91}
173
+ {"step": 145, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.29562477170241375, "tokens": 120000, "cumulative_loss_tokens": 17400000, "grad_norm": 0.9453125, "lr": 3e-05, "finish_rate": 0.819, "comp_len": 528.6, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.364, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 26.4, "frames": {"chat": 227}, "mem_gb": 9.95, "mem_gb_teacher": 9.95}
174
+ {"step": 146, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.25183481702382365, "tokens": 120000, "cumulative_loss_tokens": 17520000, "grad_norm": 0.98828125, "lr": 3e-05, "finish_rate": 0.814, "comp_len": 543.0, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.356, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 25.5, "frames": {"chat": 221}, "mem_gb": 9.94, "mem_gb_teacher": 9.94}
175
+ {"step": 147, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.2600625433813781, "tokens": 120000, "cumulative_loss_tokens": 17640000, "grad_norm": 0.89453125, "lr": 3e-05, "finish_rate": 0.859, "comp_len": 512.8, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.473, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 25.4, "frames": {"chat": 234}, "mem_gb": 9.96, "mem_gb_teacher": 9.96}
176
+ {"step": 148, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.24899872875362636, "tokens": 120000, "cumulative_loss_tokens": 17760000, "grad_norm": 0.7265625, "lr": 3e-05, "finish_rate": 0.817, "comp_len": 563.4, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.29, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 24.5, "frames": {"chat": 213}, "mem_gb": 9.9, "mem_gb_teacher": 9.9}
177
+ {"step": 149, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.22898227033279836, "tokens": 120000, "cumulative_loss_tokens": 17880000, "grad_norm": 0.87890625, "lr": 3e-05, "finish_rate": 0.836, "comp_len": 563.4, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.455, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 24.7, "frames": {"chat": 213}, "mem_gb": 9.84, "mem_gb_teacher": 9.84}
178
+ {"step": 150, "epoch": 2, "training_mode": "off-policy", "forward_topk_kl": 0.23123409348068139, "tokens": 120000, "cumulative_loss_tokens": 18000000, "grad_norm": 0.82421875, "lr": 3e-05, "finish_rate": 0.906, "comp_len": 512.8, "dropped_truncated": 0, "gold_loss": null, "gold_lambda": null, "rep_ratio": 2.392, "t_data_s": 0.0, "t_rollout_s": 0.0, "t_step_s": 25.8, "frames": {"chat": 234}, "mem_gb": 9.87, "mem_gb_teacher": 9.87}
179
+ [eval step 150] sample: "To solve this problem, we need to determine the values of \\(p\\), \\(a\\), \\(b\\), \\(m\\), and \\(r\\) that satisfy the given equations. Let's break down the problem step-by-step:\n\n1. **Understand the Constr"
180
+ checkpoint snapshot queued -> outputs/healed/keep50_offpolicy_warmup_s1224/step0150
181
+ wandb: updating run metadata
182
+ wandb: uploading summary
183
+ wandb:
184
+ wandb: Run history:
185
+ wandb: comp_len ▆▅▅▆▄█▆▇▆▅▇▄▅▄▂▇▄▆▃▃▁▄▄▇▅█▇▄▆█▄▇▇▇▄▇▆▃▅▆
186
+ wandb: cumulative_loss_tokens ▁▁▁▂▂▂▂▃▃▃▃▃▃▃▃▄▄▄▄▄▄▄▅▅▅▅▆▆▆▆▇▇▇▇▇█████
187
+ wandb: dropped_truncated ▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁
188
+ wandb: epoch ▁▁▁▁▁▁▁▁▁▁▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅██████████████
189
+ wandb: finish_rate ▁▆▇▆▄▂▆▇▆▃▇▃▁▆▇▃▃▅▅▆▅▅▁▅▃▇▆▄▆▃▃█▇▂▂▆▃█▅█
190
+ wandb: forward_topk_kl ▃▃▁▂▁▅▄▆▅▅▅▆█▇▆▇▇█▅▅▅▅▄▅▃▄▃▄▃▃▃▃▂▃▂▂▂▃▂▁
191
+ wandb: grad_norm ▁▁▁▁▁▁▁▁▁▂▁▁▁▁▁▁▂▂▂▂█▂▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁
192
+ wandb: lr ▁▃▅█████████████████████████████████████
193
+ wandb: mem_gb ▅▆▇▆▇█▆▆▇▃▆▆▆▆▄▆▆▆▆▅█▄▆▁▅▄▆▄▃▆▅▆▅▆▅▆▇▄▆▇
194
+ wandb: mem_gb_teacher ▆▇▆▂▆▆▃▆▆▅█▅▆▆▆▁▃▆▆▇▅▄▆█▃▆▆▆▃▅▄▆▆▅▅▆▆▅▆▆
195
+ wandb: +6 ...
196
+ wandb:
197
+ wandb: Run summary:
198
+ wandb: comp_len 512.8
199
+ wandb: cumulative_loss_tokens 18000000
200
+ wandb: dropped_truncated 0
201
+ wandb: epoch 2
202
+ wandb: finish_rate 0.906
203
+ wandb: forward_topk_kl 0.23123
204
+ wandb: grad_norm 0.82422
205
+ wandb: lr 3e-05
206
+ wandb: mem_gb 9.87
207
+ wandb: mem_gb_teacher 9.87
208
+ wandb: +7 ...
209
+ wandb:
210
+ wandb: 🚀 View run offpolicy-warmup-keep50-s1224 at: https://wandb.ai/hbfreed/glean-heal/runs/9td6b5cn
211
+ wandb: ⭐️ View project at: https://wandb.ai/hbfreed/glean-heal
212
+ wandb: Synced 5 W&B file(s), 0 media file(s), 0 artifact file(s) and 0 other file(s)
213
+ wandb: Find logs at: outputs/healed/keep50_offpolicy_warmup_s1224/wandb/run-20260801_230924-9td6b5cn/logs
pruned/knee0924_keep40_save.log ADDED
@@ -0,0 +1,2 @@
 
 
 
1
+
2
+ saved 3.04B params (233/1024 experts deleted) -> outputs/pruned/knee0924/keep40 (6.1 GB safetensors)
pruned/uniform_keep2575.materialize.log ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+
2
+ baseline c4=4.3505
3
+
4
+ uniform_keep25 c4=5.0687 [dead 0/1024, kept 262144, params 2.09B]
5
+
6
+ uniform_keep75 c4=4.6639 [dead 0/1024, kept 786432, params 5.31B]
7
+
8
+ decision gate (variant NLL - glean NLL; positive = glean better):
9
+ saved -> outputs/uniform_control_materialize/results.json
pruned/uniform_keep50.materialize.log ADDED
@@ -0,0 +1,7 @@
 
 
 
 
 
 
 
 
1
+
2
+ baseline c4=4.3505
3
+
4
+ uniform_keep50 c4=5.1110 [dead 0/1024, kept 524288, params 3.70B]
5
+
6
+ decision gate (variant NLL - glean NLL; positive = glean better):
7
+ saved -> outputs/uniform_control_materialize/results.json
quant_ab/bf16.json ADDED
@@ -0,0 +1,13 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "quant": "bf16",
3
+ "shards": 3,
4
+ "tokens": 221592,
5
+ "kl_mean_nats": 1.2805973847606442e-08,
6
+ "kl_median_nats": 5.440404030054857e-11,
7
+ "kl_p95_nats": 1.516273187007755e-07,
8
+ "kl_max_nats": 6.564368959516287e-07,
9
+ "top128_jaccard_mean": 1.0,
10
+ "top1_agreement": 1.0,
11
+ "captured_mass_bf16": 0.9997188871511677,
12
+ "captured_mass_quant_at_bf16_support": 0.9997189014740139
13
+ }
quant_ab/int8.json ADDED
@@ -0,0 +1,13 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "quant": "int8",
3
+ "shards": 3,
4
+ "tokens": 221592,
5
+ "kl_mean_nats": 0.005379186076306062,
6
+ "kl_median_nats": 1.4354211089084856e-05,
7
+ "kl_p95_nats": 0.02423230931162834,
8
+ "kl_max_nats": 2.3649048805236816,
9
+ "top128_jaccard_mean": 0.8820737534866655,
10
+ "top1_agreement": 0.983817105310661,
11
+ "captured_mass_bf16": 0.9997188871511677,
12
+ "captured_mass_quant_at_bf16_support": 0.9997208774759044
13
+ }
quant_ab/nf4.json ADDED
@@ -0,0 +1,13 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "quant": "nf4",
3
+ "shards": 3,
4
+ "tokens": 221592,
5
+ "kl_mean_nats": 0.03224208885256773,
6
+ "kl_median_nats": 0.00010718336125137284,
7
+ "kl_p95_nats": 0.15037551522254944,
8
+ "kl_max_nats": 14.360836029052734,
9
+ "top128_jaccard_mean": 0.7352590498594933,
10
+ "top1_agreement": 0.9612215242427524,
11
+ "captured_mass_bf16": 0.9997188871511677,
12
+ "captured_mass_quant_at_bf16_support": 0.9996136682180469
13
+ }
quant_ab/run_bf16.log ADDED
@@ -0,0 +1,17 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+
2
+ shard shard_00000.pt: cumulative 73031 tokens
3
+ shard shard_00001.pt: cumulative 142706 tokens
4
+ shard shard_00002.pt: cumulative 221592 tokens
5
+ {
6
+ "quant": "bf16",
7
+ "shards": 3,
8
+ "tokens": 221592,
9
+ "kl_mean_nats": 1.2805973847606442e-08,
10
+ "kl_median_nats": 5.440404030054857e-11,
11
+ "kl_p95_nats": 1.516273187007755e-07,
12
+ "kl_max_nats": 6.564368959516287e-07,
13
+ "top128_jaccard_mean": 1.0,
14
+ "top1_agreement": 1.0,
15
+ "captured_mass_bf16": 0.9997188871511677,
16
+ "captured_mass_quant_at_bf16_support": 0.9997189014740139
17
+ }
quant_ab/run_int8.log ADDED
@@ -0,0 +1,17 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+
2
+ shard shard_00000.pt: cumulative 73031 tokens
3
+ shard shard_00001.pt: cumulative 142706 tokens
4
+ shard shard_00002.pt: cumulative 221592 tokens
5
+ {
6
+ "quant": "int8",
7
+ "shards": 3,
8
+ "tokens": 221592,
9
+ "kl_mean_nats": 0.005379186076306062,
10
+ "kl_median_nats": 1.4354211089084856e-05,
11
+ "kl_p95_nats": 0.02423230931162834,
12
+ "kl_max_nats": 2.3649048805236816,
13
+ "top128_jaccard_mean": 0.8820737534866655,
14
+ "top1_agreement": 0.983817105310661,
15
+ "captured_mass_bf16": 0.9997188871511677,
16
+ "captured_mass_quant_at_bf16_support": 0.9997208774759044
17
+ }
quant_ab/run_nf4.log ADDED
@@ -0,0 +1,21 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+
2
+ torch._check_is_size(blocksize)
3
+
4
+ /home/henry/Documents/PythonProjects/variable-reap/.venv/lib/python3.12/site-packages/bitsandbytes/backends/cuda/ops.py:468: FutureWarning: _check_is_size will be removed in a future PyTorch release along with guard_size_oblivious. Use _check(i >= 0) instead.
5
+ torch._check_is_size(blocksize)
6
+ shard shard_00000.pt: cumulative 73031 tokens
7
+ shard shard_00001.pt: cumulative 142706 tokens
8
+ shard shard_00002.pt: cumulative 221592 tokens
9
+ {
10
+ "quant": "nf4",
11
+ "shards": 3,
12
+ "tokens": 221592,
13
+ "kl_mean_nats": 0.03224208885256773,
14
+ "kl_median_nats": 0.00010718336125137284,
15
+ "kl_p95_nats": 0.15037551522254944,
16
+ "kl_max_nats": 14.360836029052734,
17
+ "top128_jaccard_mean": 0.7352590498594933,
18
+ "top1_agreement": 0.9612215242427524,
19
+ "captured_mass_bf16": 0.9997188871511677,
20
+ "captured_mass_quant_at_bf16_support": 0.9996136682180469
21
+ }
quant_ab/run_w4a16.log ADDED
@@ -0,0 +1,21 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ /home/henry/.cache/uv/archive-v0/fRk9qAw51B-A84KDUWQdY/lib/python3.12/site-packages/transformers/quantizers/auto.py:239: UserWarning: You passed `quantization_config` or equivalent parameters to `from_pretrained` but the model you're loading already has a `quantization_config` attribute. The `quantization_config` from the model will be used.However, loading attributes (e.g. ['run_compressed']) will be overwritten with the one you passed to `from_pretrained`. The rest will be ignored.
2
+ warnings.warn(warning_msg)
3
+
4
+
5
+ shard shard_00000.pt: cumulative 73031 tokens
6
+ shard shard_00001.pt: cumulative 142706 tokens
7
+ [W720 13:34:51.273718740 CUDACachingAllocator.cpp:3933] memory allocation failed with OOM on device 0 while trying to allocate 618659840 bytes (free: 165085184, total: 25292898304).
8
+ shard shard_00002.pt: cumulative 221592 tokens
9
+ {
10
+ "quant": "w4a16",
11
+ "shards": 3,
12
+ "tokens": 221592,
13
+ "kl_mean_nats": 0.03984629407661169,
14
+ "kl_median_nats": 0.00012320266978349537,
15
+ "kl_p95_nats": 0.18111330270767212,
16
+ "kl_max_nats": 15.414010047912598,
17
+ "top128_jaccard_mean": 0.7149058585433935,
18
+ "top1_agreement": 0.9574262608758439,
19
+ "captured_mass_bf16": 0.9997188871511677,
20
+ "captured_mass_quant_at_bf16_support": 0.9995829765622533
21
+ }
quant_ab/w4a16.json ADDED
@@ -0,0 +1,13 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "quant": "w4a16",
3
+ "shards": 3,
4
+ "tokens": 221592,
5
+ "kl_mean_nats": 0.03984629407661169,
6
+ "kl_median_nats": 0.00012320266978349537,
7
+ "kl_p95_nats": 0.18111330270767212,
8
+ "kl_max_nats": 15.414010047912598,
9
+ "top128_jaccard_mean": 0.7149058585433935,
10
+ "top1_agreement": 0.9574262608758439,
11
+ "captured_mass_bf16": 0.9997188871511677,
12
+ "captured_mass_quant_at_bf16_support": 0.9995829765622533
13
+ }
qwen35_reap_keep25/chat_template.jinja ADDED
@@ -0,0 +1,154 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {%- set image_count = namespace(value=0) %}
2
+ {%- set video_count = namespace(value=0) %}
3
+ {%- macro render_content(content, do_vision_count, is_system_content=false) %}
4
+ {%- if content is string %}
5
+ {{- content }}
6
+ {%- elif content is iterable and content is not mapping %}
7
+ {%- for item in content %}
8
+ {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
9
+ {%- if is_system_content %}
10
+ {{- raise_exception('System message cannot contain images.') }}
11
+ {%- endif %}
12
+ {%- if do_vision_count %}
13
+ {%- set image_count.value = image_count.value + 1 %}
14
+ {%- endif %}
15
+ {%- if add_vision_id %}
16
+ {{- 'Picture ' ~ image_count.value ~ ': ' }}
17
+ {%- endif %}
18
+ {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
19
+ {%- elif 'video' in item or item.type == 'video' %}
20
+ {%- if is_system_content %}
21
+ {{- raise_exception('System message cannot contain videos.') }}
22
+ {%- endif %}
23
+ {%- if do_vision_count %}
24
+ {%- set video_count.value = video_count.value + 1 %}
25
+ {%- endif %}
26
+ {%- if add_vision_id %}
27
+ {{- 'Video ' ~ video_count.value ~ ': ' }}
28
+ {%- endif %}
29
+ {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
30
+ {%- elif 'text' in item %}
31
+ {{- item.text }}
32
+ {%- else %}
33
+ {{- raise_exception('Unexpected item type in content.') }}
34
+ {%- endif %}
35
+ {%- endfor %}
36
+ {%- elif content is none or content is undefined %}
37
+ {{- '' }}
38
+ {%- else %}
39
+ {{- raise_exception('Unexpected content type.') }}
40
+ {%- endif %}
41
+ {%- endmacro %}
42
+ {%- if not messages %}
43
+ {{- raise_exception('No messages provided.') }}
44
+ {%- endif %}
45
+ {%- if tools and tools is iterable and tools is not mapping %}
46
+ {{- '<|im_start|>system\n' }}
47
+ {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
48
+ {%- for tool in tools %}
49
+ {{- "\n" }}
50
+ {{- tool | tojson }}
51
+ {%- endfor %}
52
+ {{- "\n</tools>" }}
53
+ {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
54
+ {%- if messages[0].role == 'system' %}
55
+ {%- set content = render_content(messages[0].content, false, true)|trim %}
56
+ {%- if content %}
57
+ {{- '\n\n' + content }}
58
+ {%- endif %}
59
+ {%- endif %}
60
+ {{- '<|im_end|>\n' }}
61
+ {%- else %}
62
+ {%- if messages[0].role == 'system' %}
63
+ {%- set content = render_content(messages[0].content, false, true)|trim %}
64
+ {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
65
+ {%- endif %}
66
+ {%- endif %}
67
+ {%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
68
+ {%- for message in messages[::-1] %}
69
+ {%- set index = (messages|length - 1) - loop.index0 %}
70
+ {%- if ns.multi_step_tool and message.role == "user" %}
71
+ {%- set content = render_content(message.content, false)|trim %}
72
+ {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
73
+ {%- set ns.multi_step_tool = false %}
74
+ {%- set ns.last_query_index = index %}
75
+ {%- endif %}
76
+ {%- endif %}
77
+ {%- endfor %}
78
+ {%- if ns.multi_step_tool %}
79
+ {{- raise_exception('No user query found in messages.') }}
80
+ {%- endif %}
81
+ {%- for message in messages %}
82
+ {%- set content = render_content(message.content, true)|trim %}
83
+ {%- if message.role == "system" %}
84
+ {%- if not loop.first %}
85
+ {{- raise_exception('System message must be at the beginning.') }}
86
+ {%- endif %}
87
+ {%- elif message.role == "user" %}
88
+ {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
89
+ {%- elif message.role == "assistant" %}
90
+ {%- set reasoning_content = '' %}
91
+ {%- if message.reasoning_content is string %}
92
+ {%- set reasoning_content = message.reasoning_content %}
93
+ {%- else %}
94
+ {%- if '</think>' in content %}
95
+ {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
96
+ {%- set content = content.split('</think>')[-1].lstrip('\n') %}
97
+ {%- endif %}
98
+ {%- endif %}
99
+ {%- set reasoning_content = reasoning_content|trim %}
100
+ {%- if (preserve_thinking is defined and preserve_thinking is true) or (loop.index0 > ns.last_query_index) %}
101
+ {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
102
+ {%- else %}
103
+ {{- '<|im_start|>' + message.role + '\n' + content }}
104
+ {%- endif %}
105
+ {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
106
+ {%- for tool_call in message.tool_calls %}
107
+ {%- if tool_call.function is defined %}
108
+ {%- set tool_call = tool_call.function %}
109
+ {%- endif %}
110
+ {%- if loop.first %}
111
+ {%- if content|trim %}
112
+ {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
113
+ {%- else %}
114
+ {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
115
+ {%- endif %}
116
+ {%- else %}
117
+ {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
118
+ {%- endif %}
119
+ {%- if tool_call.arguments is defined %}
120
+ {%- for args_name, args_value in tool_call.arguments|items %}
121
+ {{- '<parameter=' + args_name + '>\n' }}
122
+ {%- set args_value = args_value | string if args_value is string else args_value | tojson | safe %}
123
+ {{- args_value }}
124
+ {{- '\n</parameter>\n' }}
125
+ {%- endfor %}
126
+ {%- endif %}
127
+ {{- '</function>\n</tool_call>' }}
128
+ {%- endfor %}
129
+ {%- endif %}
130
+ {{- '<|im_end|>\n' }}
131
+ {%- elif message.role == "tool" %}
132
+ {%- if loop.previtem and loop.previtem.role != "tool" %}
133
+ {{- '<|im_start|>user' }}
134
+ {%- endif %}
135
+ {{- '\n<tool_response>\n' }}
136
+ {{- content }}
137
+ {{- '\n</tool_response>' }}
138
+ {%- if not loop.last and loop.nextitem.role != "tool" %}
139
+ {{- '<|im_end|>\n' }}
140
+ {%- elif loop.last %}
141
+ {{- '<|im_end|>\n' }}
142
+ {%- endif %}
143
+ {%- else %}
144
+ {{- raise_exception('Unexpected message role.') }}
145
+ {%- endif %}
146
+ {%- endfor %}
147
+ {%- if add_generation_prompt %}
148
+ {{- '<|im_start|>assistant\n' }}
149
+ {%- if enable_thinking is defined and enable_thinking is false %}
150
+ {{- '<think>\n\n</think>\n\n' }}
151
+ {%- else %}
152
+ {{- '<think>\n' }}
153
+ {%- endif %}
154
+ {%- endif %}
qwen35_reap_keep25/config.json ADDED
@@ -0,0 +1,2752 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "PrunedQwen3_5MoeForCausalLM"
4
+ ],
5
+ "attention_bias": false,
6
+ "attention_dropout": 0.0,
7
+ "attn_output_gate": true,
8
+ "auto_map": {
9
+ "AutoConfig": "configuration_pruned_qwen3_5_moe.PrunedQwen3_5MoeTextConfig",
10
+ "AutoModelForCausalLM": "modeling_pruned_qwen3_5_moe.PrunedQwen3_5MoeForCausalLM"
11
+ },
12
+ "bos_token_id": 248044,
13
+ "dtype": "bfloat16",
14
+ "eos_token_id": 248044,
15
+ "expert_widths": [
16
+ [
17
+ 512,
18
+ 512,
19
+ 512,
20
+ 512,
21
+ 512,
22
+ 512,
23
+ 512,
24
+ 512,
25
+ 512,
26
+ 512,
27
+ 512,
28
+ 512,
29
+ 512,
30
+ 512,
31
+ 512,
32
+ 512,
33
+ 512,
34
+ 512,
35
+ 512,
36
+ 512,
37
+ 512,
38
+ 512,
39
+ 512,
40
+ 512,
41
+ 512,
42
+ 512,
43
+ 512,
44
+ 512,
45
+ 512,
46
+ 512,
47
+ 512,
48
+ 512,
49
+ 512,
50
+ 512,
51
+ 512,
52
+ 512,
53
+ 512,
54
+ 512,
55
+ 512,
56
+ 512,
57
+ 512,
58
+ 512,
59
+ 512,
60
+ 512,
61
+ 512,
62
+ 512,
63
+ 512,
64
+ 512,
65
+ 512,
66
+ 512,
67
+ 512,
68
+ 512,
69
+ 512,
70
+ 512,
71
+ 512,
72
+ 512,
73
+ 512,
74
+ 512,
75
+ 512,
76
+ 512,
77
+ 512,
78
+ 512,
79
+ 512,
80
+ 512
81
+ ],
82
+ [
83
+ 512,
84
+ 512,
85
+ 512,
86
+ 512,
87
+ 512,
88
+ 512,
89
+ 512,
90
+ 512,
91
+ 512,
92
+ 512,
93
+ 512,
94
+ 512,
95
+ 512,
96
+ 512,
97
+ 512,
98
+ 512,
99
+ 512,
100
+ 512,
101
+ 512,
102
+ 512,
103
+ 512,
104
+ 512,
105
+ 512,
106
+ 512,
107
+ 512,
108
+ 512,
109
+ 512,
110
+ 512,
111
+ 512,
112
+ 512,
113
+ 512,
114
+ 512,
115
+ 512,
116
+ 512,
117
+ 512,
118
+ 512,
119
+ 512,
120
+ 512,
121
+ 512,
122
+ 512,
123
+ 512,
124
+ 512,
125
+ 512,
126
+ 512,
127
+ 512,
128
+ 512,
129
+ 512,
130
+ 512,
131
+ 512,
132
+ 512,
133
+ 512,
134
+ 512,
135
+ 512,
136
+ 512,
137
+ 512,
138
+ 512,
139
+ 512,
140
+ 512,
141
+ 512,
142
+ 512,
143
+ 512,
144
+ 512,
145
+ 512,
146
+ 512
147
+ ],
148
+ [
149
+ 512,
150
+ 512,
151
+ 512,
152
+ 512,
153
+ 512,
154
+ 512,
155
+ 512,
156
+ 512,
157
+ 512,
158
+ 512,
159
+ 512,
160
+ 512,
161
+ 512,
162
+ 512,
163
+ 512,
164
+ 512,
165
+ 512,
166
+ 512,
167
+ 512,
168
+ 512,
169
+ 512,
170
+ 512,
171
+ 512,
172
+ 512,
173
+ 512,
174
+ 512,
175
+ 512,
176
+ 512,
177
+ 512,
178
+ 512,
179
+ 512,
180
+ 512,
181
+ 512,
182
+ 512,
183
+ 512,
184
+ 512,
185
+ 512,
186
+ 512,
187
+ 512,
188
+ 512,
189
+ 512,
190
+ 512,
191
+ 512,
192
+ 512,
193
+ 512,
194
+ 512,
195
+ 512,
196
+ 512,
197
+ 512,
198
+ 512,
199
+ 512,
200
+ 512,
201
+ 512,
202
+ 512,
203
+ 512,
204
+ 512,
205
+ 512,
206
+ 512,
207
+ 512,
208
+ 512,
209
+ 512,
210
+ 512,
211
+ 512,
212
+ 512
213
+ ],
214
+ [
215
+ 512,
216
+ 512,
217
+ 512,
218
+ 512,
219
+ 512,
220
+ 512,
221
+ 512,
222
+ 512,
223
+ 512,
224
+ 512,
225
+ 512,
226
+ 512,
227
+ 512,
228
+ 512,
229
+ 512,
230
+ 512,
231
+ 512,
232
+ 512,
233
+ 512,
234
+ 512,
235
+ 512,
236
+ 512,
237
+ 512,
238
+ 512,
239
+ 512,
240
+ 512,
241
+ 512,
242
+ 512,
243
+ 512,
244
+ 512,
245
+ 512,
246
+ 512,
247
+ 512,
248
+ 512,
249
+ 512,
250
+ 512,
251
+ 512,
252
+ 512,
253
+ 512,
254
+ 512,
255
+ 512,
256
+ 512,
257
+ 512,
258
+ 512,
259
+ 512,
260
+ 512,
261
+ 512,
262
+ 512,
263
+ 512,
264
+ 512,
265
+ 512,
266
+ 512,
267
+ 512,
268
+ 512,
269
+ 512,
270
+ 512,
271
+ 512,
272
+ 512,
273
+ 512,
274
+ 512,
275
+ 512,
276
+ 512,
277
+ 512,
278
+ 512
279
+ ],
280
+ [
281
+ 512,
282
+ 512,
283
+ 512,
284
+ 512,
285
+ 512,
286
+ 512,
287
+ 512,
288
+ 512,
289
+ 512,
290
+ 512,
291
+ 512,
292
+ 512,
293
+ 512,
294
+ 512,
295
+ 512,
296
+ 512,
297
+ 512,
298
+ 512,
299
+ 512,
300
+ 512,
301
+ 512,
302
+ 512,
303
+ 512,
304
+ 512,
305
+ 512,
306
+ 512,
307
+ 512,
308
+ 512,
309
+ 512,
310
+ 512,
311
+ 512,
312
+ 512,
313
+ 512,
314
+ 512,
315
+ 512,
316
+ 512,
317
+ 512,
318
+ 512,
319
+ 512,
320
+ 512,
321
+ 512,
322
+ 512,
323
+ 512,
324
+ 512,
325
+ 512,
326
+ 512,
327
+ 512,
328
+ 512,
329
+ 512,
330
+ 512,
331
+ 512,
332
+ 512,
333
+ 512,
334
+ 512,
335
+ 512,
336
+ 512,
337
+ 512,
338
+ 512,
339
+ 512,
340
+ 512,
341
+ 512,
342
+ 512,
343
+ 512,
344
+ 512
345
+ ],
346
+ [
347
+ 512,
348
+ 512,
349
+ 512,
350
+ 512,
351
+ 512,
352
+ 512,
353
+ 512,
354
+ 512,
355
+ 512,
356
+ 512,
357
+ 512,
358
+ 512,
359
+ 512,
360
+ 512,
361
+ 512,
362
+ 512,
363
+ 512,
364
+ 512,
365
+ 512,
366
+ 512,
367
+ 512,
368
+ 512,
369
+ 512,
370
+ 512,
371
+ 512,
372
+ 512,
373
+ 512,
374
+ 512,
375
+ 512,
376
+ 512,
377
+ 512,
378
+ 512,
379
+ 512,
380
+ 512,
381
+ 512,
382
+ 512,
383
+ 512,
384
+ 512,
385
+ 512,
386
+ 512,
387
+ 512,
388
+ 512,
389
+ 512,
390
+ 512,
391
+ 512,
392
+ 512,
393
+ 512,
394
+ 512,
395
+ 512,
396
+ 512,
397
+ 512,
398
+ 512,
399
+ 512,
400
+ 512,
401
+ 512,
402
+ 512,
403
+ 512,
404
+ 512,
405
+ 512,
406
+ 512,
407
+ 512,
408
+ 512,
409
+ 512,
410
+ 512
411
+ ],
412
+ [
413
+ 512,
414
+ 512,
415
+ 512,
416
+ 512,
417
+ 512,
418
+ 512,
419
+ 512,
420
+ 512,
421
+ 512,
422
+ 512,
423
+ 512,
424
+ 512,
425
+ 512,
426
+ 512,
427
+ 512,
428
+ 512,
429
+ 512,
430
+ 512,
431
+ 512,
432
+ 512,
433
+ 512,
434
+ 512,
435
+ 512,
436
+ 512,
437
+ 512,
438
+ 512,
439
+ 512,
440
+ 512,
441
+ 512,
442
+ 512,
443
+ 512,
444
+ 512,
445
+ 512,
446
+ 512,
447
+ 512,
448
+ 512,
449
+ 512,
450
+ 512,
451
+ 512,
452
+ 512,
453
+ 512,
454
+ 512,
455
+ 512,
456
+ 512,
457
+ 512,
458
+ 512,
459
+ 512,
460
+ 512,
461
+ 512,
462
+ 512,
463
+ 512,
464
+ 512,
465
+ 512,
466
+ 512,
467
+ 512,
468
+ 512,
469
+ 512,
470
+ 512,
471
+ 512,
472
+ 512,
473
+ 512,
474
+ 512,
475
+ 512,
476
+ 512
477
+ ],
478
+ [
479
+ 512,
480
+ 512,
481
+ 512,
482
+ 512,
483
+ 512,
484
+ 512,
485
+ 512,
486
+ 512,
487
+ 512,
488
+ 512,
489
+ 512,
490
+ 512,
491
+ 512,
492
+ 512,
493
+ 512,
494
+ 512,
495
+ 512,
496
+ 512,
497
+ 512,
498
+ 512,
499
+ 512,
500
+ 512,
501
+ 512,
502
+ 512,
503
+ 512,
504
+ 512,
505
+ 512,
506
+ 512,
507
+ 512,
508
+ 512,
509
+ 512,
510
+ 512,
511
+ 512,
512
+ 512,
513
+ 512,
514
+ 512,
515
+ 512,
516
+ 512,
517
+ 512,
518
+ 512,
519
+ 512,
520
+ 512,
521
+ 512,
522
+ 512,
523
+ 512,
524
+ 512,
525
+ 512,
526
+ 512,
527
+ 512,
528
+ 512,
529
+ 512,
530
+ 512,
531
+ 512,
532
+ 512,
533
+ 512,
534
+ 512,
535
+ 512,
536
+ 512,
537
+ 512,
538
+ 512,
539
+ 512,
540
+ 512,
541
+ 512,
542
+ 512
543
+ ],
544
+ [
545
+ 512,
546
+ 512,
547
+ 512,
548
+ 512,
549
+ 512,
550
+ 512,
551
+ 512,
552
+ 512,
553
+ 512,
554
+ 512,
555
+ 512,
556
+ 512,
557
+ 512,
558
+ 512,
559
+ 512,
560
+ 512,
561
+ 512,
562
+ 512,
563
+ 512,
564
+ 512,
565
+ 512,
566
+ 512,
567
+ 512,
568
+ 512,
569
+ 512,
570
+ 512,
571
+ 512,
572
+ 512,
573
+ 512,
574
+ 512,
575
+ 512,
576
+ 512,
577
+ 512,
578
+ 512,
579
+ 512,
580
+ 512,
581
+ 512,
582
+ 512,
583
+ 512,
584
+ 512,
585
+ 512,
586
+ 512,
587
+ 512,
588
+ 512,
589
+ 512,
590
+ 512,
591
+ 512,
592
+ 512,
593
+ 512,
594
+ 512,
595
+ 512,
596
+ 512,
597
+ 512,
598
+ 512,
599
+ 512,
600
+ 512,
601
+ 512,
602
+ 512,
603
+ 512,
604
+ 512,
605
+ 512,
606
+ 512,
607
+ 512,
608
+ 512
609
+ ],
610
+ [
611
+ 512,
612
+ 512,
613
+ 512,
614
+ 512,
615
+ 512,
616
+ 512,
617
+ 512,
618
+ 512,
619
+ 512,
620
+ 512,
621
+ 512,
622
+ 512,
623
+ 512,
624
+ 512,
625
+ 512,
626
+ 512,
627
+ 512,
628
+ 512,
629
+ 512,
630
+ 512,
631
+ 512,
632
+ 512,
633
+ 512,
634
+ 512,
635
+ 512,
636
+ 512,
637
+ 512,
638
+ 512,
639
+ 512,
640
+ 512,
641
+ 512,
642
+ 512,
643
+ 512,
644
+ 512,
645
+ 512,
646
+ 512,
647
+ 512,
648
+ 512,
649
+ 512,
650
+ 512,
651
+ 512,
652
+ 512,
653
+ 512,
654
+ 512,
655
+ 512,
656
+ 512,
657
+ 512,
658
+ 512,
659
+ 512,
660
+ 512,
661
+ 512,
662
+ 512,
663
+ 512,
664
+ 512,
665
+ 512,
666
+ 512,
667
+ 512,
668
+ 512,
669
+ 512,
670
+ 512,
671
+ 512,
672
+ 512,
673
+ 512,
674
+ 512
675
+ ],
676
+ [
677
+ 512,
678
+ 512,
679
+ 512,
680
+ 512,
681
+ 512,
682
+ 512,
683
+ 512,
684
+ 512,
685
+ 512,
686
+ 512,
687
+ 512,
688
+ 512,
689
+ 512,
690
+ 512,
691
+ 512,
692
+ 512,
693
+ 512,
694
+ 512,
695
+ 512,
696
+ 512,
697
+ 512,
698
+ 512,
699
+ 512,
700
+ 512,
701
+ 512,
702
+ 512,
703
+ 512,
704
+ 512,
705
+ 512,
706
+ 512,
707
+ 512,
708
+ 512,
709
+ 512,
710
+ 512,
711
+ 512,
712
+ 512,
713
+ 512,
714
+ 512,
715
+ 512,
716
+ 512,
717
+ 512,
718
+ 512,
719
+ 512,
720
+ 512,
721
+ 512,
722
+ 512,
723
+ 512,
724
+ 512,
725
+ 512,
726
+ 512,
727
+ 512,
728
+ 512,
729
+ 512,
730
+ 512,
731
+ 512,
732
+ 512,
733
+ 512,
734
+ 512,
735
+ 512,
736
+ 512,
737
+ 512,
738
+ 512,
739
+ 512,
740
+ 512
741
+ ],
742
+ [
743
+ 512,
744
+ 512,
745
+ 512,
746
+ 512,
747
+ 512,
748
+ 512,
749
+ 512,
750
+ 512,
751
+ 512,
752
+ 512,
753
+ 512,
754
+ 512,
755
+ 512,
756
+ 512,
757
+ 512,
758
+ 512,
759
+ 512,
760
+ 512,
761
+ 512,
762
+ 512,
763
+ 512,
764
+ 512,
765
+ 512,
766
+ 512,
767
+ 512,
768
+ 512,
769
+ 512,
770
+ 512,
771
+ 512,
772
+ 512,
773
+ 512,
774
+ 512,
775
+ 512,
776
+ 512,
777
+ 512,
778
+ 512,
779
+ 512,
780
+ 512,
781
+ 512,
782
+ 512,
783
+ 512,
784
+ 512,
785
+ 512,
786
+ 512,
787
+ 512,
788
+ 512,
789
+ 512,
790
+ 512,
791
+ 512,
792
+ 512,
793
+ 512,
794
+ 512,
795
+ 512,
796
+ 512,
797
+ 512,
798
+ 512,
799
+ 512,
800
+ 512,
801
+ 512,
802
+ 512,
803
+ 512,
804
+ 512,
805
+ 512,
806
+ 512
807
+ ],
808
+ [
809
+ 512,
810
+ 512,
811
+ 512,
812
+ 512,
813
+ 512,
814
+ 512,
815
+ 512,
816
+ 512,
817
+ 512,
818
+ 512,
819
+ 512,
820
+ 512,
821
+ 512,
822
+ 512,
823
+ 512,
824
+ 512,
825
+ 512,
826
+ 512,
827
+ 512,
828
+ 512,
829
+ 512,
830
+ 512,
831
+ 512,
832
+ 512,
833
+ 512,
834
+ 512,
835
+ 512,
836
+ 512,
837
+ 512,
838
+ 512,
839
+ 512,
840
+ 512,
841
+ 512,
842
+ 512,
843
+ 512,
844
+ 512,
845
+ 512,
846
+ 512,
847
+ 512,
848
+ 512,
849
+ 512,
850
+ 512,
851
+ 512,
852
+ 512,
853
+ 512,
854
+ 512,
855
+ 512,
856
+ 512,
857
+ 512,
858
+ 512,
859
+ 512,
860
+ 512,
861
+ 512,
862
+ 512,
863
+ 512,
864
+ 512,
865
+ 512,
866
+ 512,
867
+ 512,
868
+ 512,
869
+ 512,
870
+ 512,
871
+ 512,
872
+ 512
873
+ ],
874
+ [
875
+ 512,
876
+ 512,
877
+ 512,
878
+ 512,
879
+ 512,
880
+ 512,
881
+ 512,
882
+ 512,
883
+ 512,
884
+ 512,
885
+ 512,
886
+ 512,
887
+ 512,
888
+ 512,
889
+ 512,
890
+ 512,
891
+ 512,
892
+ 512,
893
+ 512,
894
+ 512,
895
+ 512,
896
+ 512,
897
+ 512,
898
+ 512,
899
+ 512,
900
+ 512,
901
+ 512,
902
+ 512,
903
+ 512,
904
+ 512,
905
+ 512,
906
+ 512,
907
+ 512,
908
+ 512,
909
+ 512,
910
+ 512,
911
+ 512,
912
+ 512,
913
+ 512,
914
+ 512,
915
+ 512,
916
+ 512,
917
+ 512,
918
+ 512,
919
+ 512,
920
+ 512,
921
+ 512,
922
+ 512,
923
+ 512,
924
+ 512,
925
+ 512,
926
+ 512,
927
+ 512,
928
+ 512,
929
+ 512,
930
+ 512,
931
+ 512,
932
+ 512,
933
+ 512,
934
+ 512,
935
+ 512,
936
+ 512,
937
+ 512,
938
+ 512
939
+ ],
940
+ [
941
+ 512,
942
+ 512,
943
+ 512,
944
+ 512,
945
+ 512,
946
+ 512,
947
+ 512,
948
+ 512,
949
+ 512,
950
+ 512,
951
+ 512,
952
+ 512,
953
+ 512,
954
+ 512,
955
+ 512,
956
+ 512,
957
+ 512,
958
+ 512,
959
+ 512,
960
+ 512,
961
+ 512,
962
+ 512,
963
+ 512,
964
+ 512,
965
+ 512,
966
+ 512,
967
+ 512,
968
+ 512,
969
+ 512,
970
+ 512,
971
+ 512,
972
+ 512,
973
+ 512,
974
+ 512,
975
+ 512,
976
+ 512,
977
+ 512,
978
+ 512,
979
+ 512,
980
+ 512,
981
+ 512,
982
+ 512,
983
+ 512,
984
+ 512,
985
+ 512,
986
+ 512,
987
+ 512,
988
+ 512,
989
+ 512,
990
+ 512,
991
+ 512,
992
+ 512,
993
+ 512,
994
+ 512,
995
+ 512,
996
+ 512,
997
+ 512,
998
+ 512,
999
+ 512,
1000
+ 512,
1001
+ 512,
1002
+ 512,
1003
+ 512,
1004
+ 512
1005
+ ],
1006
+ [
1007
+ 512,
1008
+ 512,
1009
+ 512,
1010
+ 512,
1011
+ 512,
1012
+ 512,
1013
+ 512,
1014
+ 512,
1015
+ 512,
1016
+ 512,
1017
+ 512,
1018
+ 512,
1019
+ 512,
1020
+ 512,
1021
+ 512,
1022
+ 512,
1023
+ 512,
1024
+ 512,
1025
+ 512,
1026
+ 512,
1027
+ 512,
1028
+ 512,
1029
+ 512,
1030
+ 512,
1031
+ 512,
1032
+ 512,
1033
+ 512,
1034
+ 512,
1035
+ 512,
1036
+ 512,
1037
+ 512,
1038
+ 512,
1039
+ 512,
1040
+ 512,
1041
+ 512,
1042
+ 512,
1043
+ 512,
1044
+ 512,
1045
+ 512,
1046
+ 512,
1047
+ 512,
1048
+ 512,
1049
+ 512,
1050
+ 512,
1051
+ 512,
1052
+ 512,
1053
+ 512,
1054
+ 512,
1055
+ 512,
1056
+ 512,
1057
+ 512,
1058
+ 512,
1059
+ 512,
1060
+ 512,
1061
+ 512,
1062
+ 512,
1063
+ 512,
1064
+ 512,
1065
+ 512,
1066
+ 512,
1067
+ 512,
1068
+ 512,
1069
+ 512,
1070
+ 512
1071
+ ],
1072
+ [
1073
+ 512,
1074
+ 512,
1075
+ 512,
1076
+ 512,
1077
+ 512,
1078
+ 512,
1079
+ 512,
1080
+ 512,
1081
+ 512,
1082
+ 512,
1083
+ 512,
1084
+ 512,
1085
+ 512,
1086
+ 512,
1087
+ 512,
1088
+ 512,
1089
+ 512,
1090
+ 512,
1091
+ 512,
1092
+ 512,
1093
+ 512,
1094
+ 512,
1095
+ 512,
1096
+ 512,
1097
+ 512,
1098
+ 512,
1099
+ 512,
1100
+ 512,
1101
+ 512,
1102
+ 512,
1103
+ 512,
1104
+ 512,
1105
+ 512,
1106
+ 512,
1107
+ 512,
1108
+ 512,
1109
+ 512,
1110
+ 512,
1111
+ 512,
1112
+ 512,
1113
+ 512,
1114
+ 512,
1115
+ 512,
1116
+ 512,
1117
+ 512,
1118
+ 512,
1119
+ 512,
1120
+ 512,
1121
+ 512,
1122
+ 512,
1123
+ 512,
1124
+ 512,
1125
+ 512,
1126
+ 512,
1127
+ 512,
1128
+ 512,
1129
+ 512,
1130
+ 512,
1131
+ 512,
1132
+ 512,
1133
+ 512,
1134
+ 512,
1135
+ 512,
1136
+ 512
1137
+ ],
1138
+ [
1139
+ 512,
1140
+ 512,
1141
+ 512,
1142
+ 512,
1143
+ 512,
1144
+ 512,
1145
+ 512,
1146
+ 512,
1147
+ 512,
1148
+ 512,
1149
+ 512,
1150
+ 512,
1151
+ 512,
1152
+ 512,
1153
+ 512,
1154
+ 512,
1155
+ 512,
1156
+ 512,
1157
+ 512,
1158
+ 512,
1159
+ 512,
1160
+ 512,
1161
+ 512,
1162
+ 512,
1163
+ 512,
1164
+ 512,
1165
+ 512,
1166
+ 512,
1167
+ 512,
1168
+ 512,
1169
+ 512,
1170
+ 512,
1171
+ 512,
1172
+ 512,
1173
+ 512,
1174
+ 512,
1175
+ 512,
1176
+ 512,
1177
+ 512,
1178
+ 512,
1179
+ 512,
1180
+ 512,
1181
+ 512,
1182
+ 512,
1183
+ 512,
1184
+ 512,
1185
+ 512,
1186
+ 512,
1187
+ 512,
1188
+ 512,
1189
+ 512,
1190
+ 512,
1191
+ 512,
1192
+ 512,
1193
+ 512,
1194
+ 512,
1195
+ 512,
1196
+ 512,
1197
+ 512,
1198
+ 512,
1199
+ 512,
1200
+ 512,
1201
+ 512,
1202
+ 512
1203
+ ],
1204
+ [
1205
+ 512,
1206
+ 512,
1207
+ 512,
1208
+ 512,
1209
+ 512,
1210
+ 512,
1211
+ 512,
1212
+ 512,
1213
+ 512,
1214
+ 512,
1215
+ 512,
1216
+ 512,
1217
+ 512,
1218
+ 512,
1219
+ 512,
1220
+ 512,
1221
+ 512,
1222
+ 512,
1223
+ 512,
1224
+ 512,
1225
+ 512,
1226
+ 512,
1227
+ 512,
1228
+ 512,
1229
+ 512,
1230
+ 512,
1231
+ 512,
1232
+ 512,
1233
+ 512,
1234
+ 512,
1235
+ 512,
1236
+ 512,
1237
+ 512,
1238
+ 512,
1239
+ 512,
1240
+ 512,
1241
+ 512,
1242
+ 512,
1243
+ 512,
1244
+ 512,
1245
+ 512,
1246
+ 512,
1247
+ 512,
1248
+ 512,
1249
+ 512,
1250
+ 512,
1251
+ 512,
1252
+ 512,
1253
+ 512,
1254
+ 512,
1255
+ 512,
1256
+ 512,
1257
+ 512,
1258
+ 512,
1259
+ 512,
1260
+ 512,
1261
+ 512,
1262
+ 512,
1263
+ 512,
1264
+ 512,
1265
+ 512,
1266
+ 512,
1267
+ 512,
1268
+ 512
1269
+ ],
1270
+ [
1271
+ 512,
1272
+ 512,
1273
+ 512,
1274
+ 512,
1275
+ 512,
1276
+ 512,
1277
+ 512,
1278
+ 512,
1279
+ 512,
1280
+ 512,
1281
+ 512,
1282
+ 512,
1283
+ 512,
1284
+ 512,
1285
+ 512,
1286
+ 512,
1287
+ 512,
1288
+ 512,
1289
+ 512,
1290
+ 512,
1291
+ 512,
1292
+ 512,
1293
+ 512,
1294
+ 512,
1295
+ 512,
1296
+ 512,
1297
+ 512,
1298
+ 512,
1299
+ 512,
1300
+ 512,
1301
+ 512,
1302
+ 512,
1303
+ 512,
1304
+ 512,
1305
+ 512,
1306
+ 512,
1307
+ 512,
1308
+ 512,
1309
+ 512,
1310
+ 512,
1311
+ 512,
1312
+ 512,
1313
+ 512,
1314
+ 512,
1315
+ 512,
1316
+ 512,
1317
+ 512,
1318
+ 512,
1319
+ 512,
1320
+ 512,
1321
+ 512,
1322
+ 512,
1323
+ 512,
1324
+ 512,
1325
+ 512,
1326
+ 512,
1327
+ 512,
1328
+ 512,
1329
+ 512,
1330
+ 512,
1331
+ 512,
1332
+ 512,
1333
+ 512,
1334
+ 512
1335
+ ],
1336
+ [
1337
+ 512,
1338
+ 512,
1339
+ 512,
1340
+ 512,
1341
+ 512,
1342
+ 512,
1343
+ 512,
1344
+ 512,
1345
+ 512,
1346
+ 512,
1347
+ 512,
1348
+ 512,
1349
+ 512,
1350
+ 512,
1351
+ 512,
1352
+ 512,
1353
+ 512,
1354
+ 512,
1355
+ 512,
1356
+ 512,
1357
+ 512,
1358
+ 512,
1359
+ 512,
1360
+ 512,
1361
+ 512,
1362
+ 512,
1363
+ 512,
1364
+ 512,
1365
+ 512,
1366
+ 512,
1367
+ 512,
1368
+ 512,
1369
+ 512,
1370
+ 512,
1371
+ 512,
1372
+ 512,
1373
+ 512,
1374
+ 512,
1375
+ 512,
1376
+ 512,
1377
+ 512,
1378
+ 512,
1379
+ 512,
1380
+ 512,
1381
+ 512,
1382
+ 512,
1383
+ 512,
1384
+ 512,
1385
+ 512,
1386
+ 512,
1387
+ 512,
1388
+ 512,
1389
+ 512,
1390
+ 512,
1391
+ 512,
1392
+ 512,
1393
+ 512,
1394
+ 512,
1395
+ 512,
1396
+ 512,
1397
+ 512,
1398
+ 512,
1399
+ 512,
1400
+ 512
1401
+ ],
1402
+ [
1403
+ 512,
1404
+ 512,
1405
+ 512,
1406
+ 512,
1407
+ 512,
1408
+ 512,
1409
+ 512,
1410
+ 512,
1411
+ 512,
1412
+ 512,
1413
+ 512,
1414
+ 512,
1415
+ 512,
1416
+ 512,
1417
+ 512,
1418
+ 512,
1419
+ 512,
1420
+ 512,
1421
+ 512,
1422
+ 512,
1423
+ 512,
1424
+ 512,
1425
+ 512,
1426
+ 512,
1427
+ 512,
1428
+ 512,
1429
+ 512,
1430
+ 512,
1431
+ 512,
1432
+ 512,
1433
+ 512,
1434
+ 512,
1435
+ 512,
1436
+ 512,
1437
+ 512,
1438
+ 512,
1439
+ 512,
1440
+ 512,
1441
+ 512,
1442
+ 512,
1443
+ 512,
1444
+ 512,
1445
+ 512,
1446
+ 512,
1447
+ 512,
1448
+ 512,
1449
+ 512,
1450
+ 512,
1451
+ 512,
1452
+ 512,
1453
+ 512,
1454
+ 512,
1455
+ 512,
1456
+ 512,
1457
+ 512,
1458
+ 512,
1459
+ 512,
1460
+ 512,
1461
+ 512,
1462
+ 512,
1463
+ 512,
1464
+ 512,
1465
+ 512,
1466
+ 512
1467
+ ],
1468
+ [
1469
+ 512,
1470
+ 512,
1471
+ 512,
1472
+ 512,
1473
+ 512,
1474
+ 512,
1475
+ 512,
1476
+ 512,
1477
+ 512,
1478
+ 512,
1479
+ 512,
1480
+ 512,
1481
+ 512,
1482
+ 512,
1483
+ 512,
1484
+ 512,
1485
+ 512,
1486
+ 512,
1487
+ 512,
1488
+ 512,
1489
+ 512,
1490
+ 512,
1491
+ 512,
1492
+ 512,
1493
+ 512,
1494
+ 512,
1495
+ 512,
1496
+ 512,
1497
+ 512,
1498
+ 512,
1499
+ 512,
1500
+ 512,
1501
+ 512,
1502
+ 512,
1503
+ 512,
1504
+ 512,
1505
+ 512,
1506
+ 512,
1507
+ 512,
1508
+ 512,
1509
+ 512,
1510
+ 512,
1511
+ 512,
1512
+ 512,
1513
+ 512,
1514
+ 512,
1515
+ 512,
1516
+ 512,
1517
+ 512,
1518
+ 512,
1519
+ 512,
1520
+ 512,
1521
+ 512,
1522
+ 512,
1523
+ 512,
1524
+ 512,
1525
+ 512,
1526
+ 512,
1527
+ 512,
1528
+ 512,
1529
+ 512,
1530
+ 512,
1531
+ 512,
1532
+ 512
1533
+ ],
1534
+ [
1535
+ 512,
1536
+ 512,
1537
+ 512,
1538
+ 512,
1539
+ 512,
1540
+ 512,
1541
+ 512,
1542
+ 512,
1543
+ 512,
1544
+ 512,
1545
+ 512,
1546
+ 512,
1547
+ 512,
1548
+ 512,
1549
+ 512,
1550
+ 512,
1551
+ 512,
1552
+ 512,
1553
+ 512,
1554
+ 512,
1555
+ 512,
1556
+ 512,
1557
+ 512,
1558
+ 512,
1559
+ 512,
1560
+ 512,
1561
+ 512,
1562
+ 512,
1563
+ 512,
1564
+ 512,
1565
+ 512,
1566
+ 512,
1567
+ 512,
1568
+ 512,
1569
+ 512,
1570
+ 512,
1571
+ 512,
1572
+ 512,
1573
+ 512,
1574
+ 512,
1575
+ 512,
1576
+ 512,
1577
+ 512,
1578
+ 512,
1579
+ 512,
1580
+ 512,
1581
+ 512,
1582
+ 512,
1583
+ 512,
1584
+ 512,
1585
+ 512,
1586
+ 512,
1587
+ 512,
1588
+ 512,
1589
+ 512,
1590
+ 512,
1591
+ 512,
1592
+ 512,
1593
+ 512,
1594
+ 512,
1595
+ 512,
1596
+ 512,
1597
+ 512,
1598
+ 512
1599
+ ],
1600
+ [
1601
+ 512,
1602
+ 512,
1603
+ 512,
1604
+ 512,
1605
+ 512,
1606
+ 512,
1607
+ 512,
1608
+ 512,
1609
+ 512,
1610
+ 512,
1611
+ 512,
1612
+ 512,
1613
+ 512,
1614
+ 512,
1615
+ 512,
1616
+ 512,
1617
+ 512,
1618
+ 512,
1619
+ 512,
1620
+ 512,
1621
+ 512,
1622
+ 512,
1623
+ 512,
1624
+ 512,
1625
+ 512,
1626
+ 512,
1627
+ 512,
1628
+ 512,
1629
+ 512,
1630
+ 512,
1631
+ 512,
1632
+ 512,
1633
+ 512,
1634
+ 512,
1635
+ 512,
1636
+ 512,
1637
+ 512,
1638
+ 512,
1639
+ 512,
1640
+ 512,
1641
+ 512,
1642
+ 512,
1643
+ 512,
1644
+ 512,
1645
+ 512,
1646
+ 512,
1647
+ 512,
1648
+ 512,
1649
+ 512,
1650
+ 512,
1651
+ 512,
1652
+ 512,
1653
+ 512,
1654
+ 512,
1655
+ 512,
1656
+ 512,
1657
+ 512,
1658
+ 512,
1659
+ 512,
1660
+ 512,
1661
+ 512,
1662
+ 512,
1663
+ 512,
1664
+ 512
1665
+ ],
1666
+ [
1667
+ 512,
1668
+ 512,
1669
+ 512,
1670
+ 512,
1671
+ 512,
1672
+ 512,
1673
+ 512,
1674
+ 512,
1675
+ 512,
1676
+ 512,
1677
+ 512,
1678
+ 512,
1679
+ 512,
1680
+ 512,
1681
+ 512,
1682
+ 512,
1683
+ 512,
1684
+ 512,
1685
+ 512,
1686
+ 512,
1687
+ 512,
1688
+ 512,
1689
+ 512,
1690
+ 512,
1691
+ 512,
1692
+ 512,
1693
+ 512,
1694
+ 512,
1695
+ 512,
1696
+ 512,
1697
+ 512,
1698
+ 512,
1699
+ 512,
1700
+ 512,
1701
+ 512,
1702
+ 512,
1703
+ 512,
1704
+ 512,
1705
+ 512,
1706
+ 512,
1707
+ 512,
1708
+ 512,
1709
+ 512,
1710
+ 512,
1711
+ 512,
1712
+ 512,
1713
+ 512,
1714
+ 512,
1715
+ 512,
1716
+ 512,
1717
+ 512,
1718
+ 512,
1719
+ 512,
1720
+ 512,
1721
+ 512,
1722
+ 512,
1723
+ 512,
1724
+ 512,
1725
+ 512,
1726
+ 512,
1727
+ 512,
1728
+ 512,
1729
+ 512,
1730
+ 512
1731
+ ],
1732
+ [
1733
+ 512,
1734
+ 512,
1735
+ 512,
1736
+ 512,
1737
+ 512,
1738
+ 512,
1739
+ 512,
1740
+ 512,
1741
+ 512,
1742
+ 512,
1743
+ 512,
1744
+ 512,
1745
+ 512,
1746
+ 512,
1747
+ 512,
1748
+ 512,
1749
+ 512,
1750
+ 512,
1751
+ 512,
1752
+ 512,
1753
+ 512,
1754
+ 512,
1755
+ 512,
1756
+ 512,
1757
+ 512,
1758
+ 512,
1759
+ 512,
1760
+ 512,
1761
+ 512,
1762
+ 512,
1763
+ 512,
1764
+ 512,
1765
+ 512,
1766
+ 512,
1767
+ 512,
1768
+ 512,
1769
+ 512,
1770
+ 512,
1771
+ 512,
1772
+ 512,
1773
+ 512,
1774
+ 512,
1775
+ 512,
1776
+ 512,
1777
+ 512,
1778
+ 512,
1779
+ 512,
1780
+ 512,
1781
+ 512,
1782
+ 512,
1783
+ 512,
1784
+ 512,
1785
+ 512,
1786
+ 512,
1787
+ 512,
1788
+ 512,
1789
+ 512,
1790
+ 512,
1791
+ 512,
1792
+ 512,
1793
+ 512,
1794
+ 512,
1795
+ 512,
1796
+ 512
1797
+ ],
1798
+ [
1799
+ 512,
1800
+ 512,
1801
+ 512,
1802
+ 512,
1803
+ 512,
1804
+ 512,
1805
+ 512,
1806
+ 512,
1807
+ 512,
1808
+ 512,
1809
+ 512,
1810
+ 512,
1811
+ 512,
1812
+ 512,
1813
+ 512,
1814
+ 512,
1815
+ 512,
1816
+ 512,
1817
+ 512,
1818
+ 512,
1819
+ 512,
1820
+ 512,
1821
+ 512,
1822
+ 512,
1823
+ 512,
1824
+ 512,
1825
+ 512,
1826
+ 512,
1827
+ 512,
1828
+ 512,
1829
+ 512,
1830
+ 512,
1831
+ 512,
1832
+ 512,
1833
+ 512,
1834
+ 512,
1835
+ 512,
1836
+ 512,
1837
+ 512,
1838
+ 512,
1839
+ 512,
1840
+ 512,
1841
+ 512,
1842
+ 512,
1843
+ 512,
1844
+ 512,
1845
+ 512,
1846
+ 512,
1847
+ 512,
1848
+ 512,
1849
+ 512,
1850
+ 512,
1851
+ 512,
1852
+ 512,
1853
+ 512,
1854
+ 512,
1855
+ 512,
1856
+ 512,
1857
+ 512,
1858
+ 512,
1859
+ 512,
1860
+ 512,
1861
+ 512,
1862
+ 512
1863
+ ],
1864
+ [
1865
+ 512,
1866
+ 512,
1867
+ 512,
1868
+ 512,
1869
+ 512,
1870
+ 512,
1871
+ 512,
1872
+ 512,
1873
+ 512,
1874
+ 512,
1875
+ 512,
1876
+ 512,
1877
+ 512,
1878
+ 512,
1879
+ 512,
1880
+ 512,
1881
+ 512,
1882
+ 512,
1883
+ 512,
1884
+ 512,
1885
+ 512,
1886
+ 512,
1887
+ 512,
1888
+ 512,
1889
+ 512,
1890
+ 512,
1891
+ 512,
1892
+ 512,
1893
+ 512,
1894
+ 512,
1895
+ 512,
1896
+ 512,
1897
+ 512,
1898
+ 512,
1899
+ 512,
1900
+ 512,
1901
+ 512,
1902
+ 512,
1903
+ 512,
1904
+ 512,
1905
+ 512,
1906
+ 512,
1907
+ 512,
1908
+ 512,
1909
+ 512,
1910
+ 512,
1911
+ 512,
1912
+ 512,
1913
+ 512,
1914
+ 512,
1915
+ 512,
1916
+ 512,
1917
+ 512,
1918
+ 512,
1919
+ 512,
1920
+ 512,
1921
+ 512,
1922
+ 512,
1923
+ 512,
1924
+ 512,
1925
+ 512,
1926
+ 512,
1927
+ 512,
1928
+ 512
1929
+ ],
1930
+ [
1931
+ 512,
1932
+ 512,
1933
+ 512,
1934
+ 512,
1935
+ 512,
1936
+ 512,
1937
+ 512,
1938
+ 512,
1939
+ 512,
1940
+ 512,
1941
+ 512,
1942
+ 512,
1943
+ 512,
1944
+ 512,
1945
+ 512,
1946
+ 512,
1947
+ 512,
1948
+ 512,
1949
+ 512,
1950
+ 512,
1951
+ 512,
1952
+ 512,
1953
+ 512,
1954
+ 512,
1955
+ 512,
1956
+ 512,
1957
+ 512,
1958
+ 512,
1959
+ 512,
1960
+ 512,
1961
+ 512,
1962
+ 512,
1963
+ 512,
1964
+ 512,
1965
+ 512,
1966
+ 512,
1967
+ 512,
1968
+ 512,
1969
+ 512,
1970
+ 512,
1971
+ 512,
1972
+ 512,
1973
+ 512,
1974
+ 512,
1975
+ 512,
1976
+ 512,
1977
+ 512,
1978
+ 512,
1979
+ 512,
1980
+ 512,
1981
+ 512,
1982
+ 512,
1983
+ 512,
1984
+ 512,
1985
+ 512,
1986
+ 512,
1987
+ 512,
1988
+ 512,
1989
+ 512,
1990
+ 512,
1991
+ 512,
1992
+ 512,
1993
+ 512,
1994
+ 512
1995
+ ],
1996
+ [
1997
+ 512,
1998
+ 512,
1999
+ 512,
2000
+ 512,
2001
+ 512,
2002
+ 512,
2003
+ 512,
2004
+ 512,
2005
+ 512,
2006
+ 512,
2007
+ 512,
2008
+ 512,
2009
+ 512,
2010
+ 512,
2011
+ 512,
2012
+ 512,
2013
+ 512,
2014
+ 512,
2015
+ 512,
2016
+ 512,
2017
+ 512,
2018
+ 512,
2019
+ 512,
2020
+ 512,
2021
+ 512,
2022
+ 512,
2023
+ 512,
2024
+ 512,
2025
+ 512,
2026
+ 512,
2027
+ 512,
2028
+ 512,
2029
+ 512,
2030
+ 512,
2031
+ 512,
2032
+ 512,
2033
+ 512,
2034
+ 512,
2035
+ 512,
2036
+ 512,
2037
+ 512,
2038
+ 512,
2039
+ 512,
2040
+ 512,
2041
+ 512,
2042
+ 512,
2043
+ 512,
2044
+ 512,
2045
+ 512,
2046
+ 512,
2047
+ 512,
2048
+ 512,
2049
+ 512,
2050
+ 512,
2051
+ 512,
2052
+ 512,
2053
+ 512,
2054
+ 512,
2055
+ 512,
2056
+ 512,
2057
+ 512,
2058
+ 512,
2059
+ 512,
2060
+ 512
2061
+ ],
2062
+ [
2063
+ 512,
2064
+ 512,
2065
+ 512,
2066
+ 512,
2067
+ 512,
2068
+ 512,
2069
+ 512,
2070
+ 512,
2071
+ 512,
2072
+ 512,
2073
+ 512,
2074
+ 512,
2075
+ 512,
2076
+ 512,
2077
+ 512,
2078
+ 512,
2079
+ 512,
2080
+ 512,
2081
+ 512,
2082
+ 512,
2083
+ 512,
2084
+ 512,
2085
+ 512,
2086
+ 512,
2087
+ 512,
2088
+ 512,
2089
+ 512,
2090
+ 512,
2091
+ 512,
2092
+ 512,
2093
+ 512,
2094
+ 512,
2095
+ 512,
2096
+ 512,
2097
+ 512,
2098
+ 512,
2099
+ 512,
2100
+ 512,
2101
+ 512,
2102
+ 512,
2103
+ 512,
2104
+ 512,
2105
+ 512,
2106
+ 512,
2107
+ 512,
2108
+ 512,
2109
+ 512,
2110
+ 512,
2111
+ 512,
2112
+ 512,
2113
+ 512,
2114
+ 512,
2115
+ 512,
2116
+ 512,
2117
+ 512,
2118
+ 512,
2119
+ 512,
2120
+ 512,
2121
+ 512,
2122
+ 512,
2123
+ 512,
2124
+ 512,
2125
+ 512,
2126
+ 512
2127
+ ],
2128
+ [
2129
+ 512,
2130
+ 512,
2131
+ 512,
2132
+ 512,
2133
+ 512,
2134
+ 512,
2135
+ 512,
2136
+ 512,
2137
+ 512,
2138
+ 512,
2139
+ 512,
2140
+ 512,
2141
+ 512,
2142
+ 512,
2143
+ 512,
2144
+ 512,
2145
+ 512,
2146
+ 512,
2147
+ 512,
2148
+ 512,
2149
+ 512,
2150
+ 512,
2151
+ 512,
2152
+ 512,
2153
+ 512,
2154
+ 512,
2155
+ 512,
2156
+ 512,
2157
+ 512,
2158
+ 512,
2159
+ 512,
2160
+ 512,
2161
+ 512,
2162
+ 512,
2163
+ 512,
2164
+ 512,
2165
+ 512,
2166
+ 512,
2167
+ 512,
2168
+ 512,
2169
+ 512,
2170
+ 512,
2171
+ 512,
2172
+ 512,
2173
+ 512,
2174
+ 512,
2175
+ 512,
2176
+ 512,
2177
+ 512,
2178
+ 512,
2179
+ 512,
2180
+ 512,
2181
+ 512,
2182
+ 512,
2183
+ 512,
2184
+ 512,
2185
+ 512,
2186
+ 512,
2187
+ 512,
2188
+ 512,
2189
+ 512,
2190
+ 512,
2191
+ 512,
2192
+ 512
2193
+ ],
2194
+ [
2195
+ 512,
2196
+ 512,
2197
+ 512,
2198
+ 512,
2199
+ 512,
2200
+ 512,
2201
+ 512,
2202
+ 512,
2203
+ 512,
2204
+ 512,
2205
+ 512,
2206
+ 512,
2207
+ 512,
2208
+ 512,
2209
+ 512,
2210
+ 512,
2211
+ 512,
2212
+ 512,
2213
+ 512,
2214
+ 512,
2215
+ 512,
2216
+ 512,
2217
+ 512,
2218
+ 512,
2219
+ 512,
2220
+ 512,
2221
+ 512,
2222
+ 512,
2223
+ 512,
2224
+ 512,
2225
+ 512,
2226
+ 512,
2227
+ 512,
2228
+ 512,
2229
+ 512,
2230
+ 512,
2231
+ 512,
2232
+ 512,
2233
+ 512,
2234
+ 512,
2235
+ 512,
2236
+ 512,
2237
+ 512,
2238
+ 512,
2239
+ 512,
2240
+ 512,
2241
+ 512,
2242
+ 512,
2243
+ 512,
2244
+ 512,
2245
+ 512,
2246
+ 512,
2247
+ 512,
2248
+ 512,
2249
+ 512,
2250
+ 512,
2251
+ 512,
2252
+ 512,
2253
+ 512,
2254
+ 512,
2255
+ 512,
2256
+ 512,
2257
+ 512,
2258
+ 512
2259
+ ],
2260
+ [
2261
+ 512,
2262
+ 512,
2263
+ 512,
2264
+ 512,
2265
+ 512,
2266
+ 512,
2267
+ 512,
2268
+ 512,
2269
+ 512,
2270
+ 512,
2271
+ 512,
2272
+ 512,
2273
+ 512,
2274
+ 512,
2275
+ 512,
2276
+ 512,
2277
+ 512,
2278
+ 512,
2279
+ 512,
2280
+ 512,
2281
+ 512,
2282
+ 512,
2283
+ 512,
2284
+ 512,
2285
+ 512,
2286
+ 512,
2287
+ 512,
2288
+ 512,
2289
+ 512,
2290
+ 512,
2291
+ 512,
2292
+ 512,
2293
+ 512,
2294
+ 512,
2295
+ 512,
2296
+ 512,
2297
+ 512,
2298
+ 512,
2299
+ 512,
2300
+ 512,
2301
+ 512,
2302
+ 512,
2303
+ 512,
2304
+ 512,
2305
+ 512,
2306
+ 512,
2307
+ 512,
2308
+ 512,
2309
+ 512,
2310
+ 512,
2311
+ 512,
2312
+ 512,
2313
+ 512,
2314
+ 512,
2315
+ 512,
2316
+ 512,
2317
+ 512,
2318
+ 512,
2319
+ 512,
2320
+ 512,
2321
+ 512,
2322
+ 512,
2323
+ 512,
2324
+ 512
2325
+ ],
2326
+ [
2327
+ 512,
2328
+ 512,
2329
+ 512,
2330
+ 512,
2331
+ 512,
2332
+ 512,
2333
+ 512,
2334
+ 512,
2335
+ 512,
2336
+ 512,
2337
+ 512,
2338
+ 512,
2339
+ 512,
2340
+ 512,
2341
+ 512,
2342
+ 512,
2343
+ 512,
2344
+ 512,
2345
+ 512,
2346
+ 512,
2347
+ 512,
2348
+ 512,
2349
+ 512,
2350
+ 512,
2351
+ 512,
2352
+ 512,
2353
+ 512,
2354
+ 512,
2355
+ 512,
2356
+ 512,
2357
+ 512,
2358
+ 512,
2359
+ 512,
2360
+ 512,
2361
+ 512,
2362
+ 512,
2363
+ 512,
2364
+ 512,
2365
+ 512,
2366
+ 512,
2367
+ 512,
2368
+ 512,
2369
+ 512,
2370
+ 512,
2371
+ 512,
2372
+ 512,
2373
+ 512,
2374
+ 512,
2375
+ 512,
2376
+ 512,
2377
+ 512,
2378
+ 512,
2379
+ 512,
2380
+ 512,
2381
+ 512,
2382
+ 512,
2383
+ 512,
2384
+ 512,
2385
+ 512,
2386
+ 512,
2387
+ 512,
2388
+ 512,
2389
+ 512,
2390
+ 512
2391
+ ],
2392
+ [
2393
+ 512,
2394
+ 512,
2395
+ 512,
2396
+ 512,
2397
+ 512,
2398
+ 512,
2399
+ 512,
2400
+ 512,
2401
+ 512,
2402
+ 512,
2403
+ 512,
2404
+ 512,
2405
+ 512,
2406
+ 512,
2407
+ 512,
2408
+ 512,
2409
+ 512,
2410
+ 512,
2411
+ 512,
2412
+ 512,
2413
+ 512,
2414
+ 512,
2415
+ 512,
2416
+ 512,
2417
+ 512,
2418
+ 512,
2419
+ 512,
2420
+ 512,
2421
+ 512,
2422
+ 512,
2423
+ 512,
2424
+ 512,
2425
+ 512,
2426
+ 512,
2427
+ 512,
2428
+ 512,
2429
+ 512,
2430
+ 512,
2431
+ 512,
2432
+ 512,
2433
+ 512,
2434
+ 512,
2435
+ 512,
2436
+ 512,
2437
+ 512,
2438
+ 512,
2439
+ 512,
2440
+ 512,
2441
+ 512,
2442
+ 512,
2443
+ 512,
2444
+ 512,
2445
+ 512,
2446
+ 512,
2447
+ 512,
2448
+ 512,
2449
+ 512,
2450
+ 512,
2451
+ 512,
2452
+ 512,
2453
+ 512,
2454
+ 512,
2455
+ 512,
2456
+ 512
2457
+ ],
2458
+ [
2459
+ 512,
2460
+ 512,
2461
+ 512,
2462
+ 512,
2463
+ 512,
2464
+ 512,
2465
+ 512,
2466
+ 512,
2467
+ 512,
2468
+ 512,
2469
+ 512,
2470
+ 512,
2471
+ 512,
2472
+ 512,
2473
+ 512,
2474
+ 512,
2475
+ 512,
2476
+ 512,
2477
+ 512,
2478
+ 512,
2479
+ 512,
2480
+ 512,
2481
+ 512,
2482
+ 512,
2483
+ 512,
2484
+ 512,
2485
+ 512,
2486
+ 512,
2487
+ 512,
2488
+ 512,
2489
+ 512,
2490
+ 512,
2491
+ 512,
2492
+ 512,
2493
+ 512,
2494
+ 512,
2495
+ 512,
2496
+ 512,
2497
+ 512,
2498
+ 512,
2499
+ 512,
2500
+ 512,
2501
+ 512,
2502
+ 512,
2503
+ 512,
2504
+ 512,
2505
+ 512,
2506
+ 512,
2507
+ 512,
2508
+ 512,
2509
+ 512,
2510
+ 512,
2511
+ 512,
2512
+ 512,
2513
+ 512,
2514
+ 512,
2515
+ 512,
2516
+ 512,
2517
+ 512,
2518
+ 512,
2519
+ 512,
2520
+ 512,
2521
+ 512,
2522
+ 512
2523
+ ],
2524
+ [
2525
+ 512,
2526
+ 512,
2527
+ 512,
2528
+ 512,
2529
+ 512,
2530
+ 512,
2531
+ 512,
2532
+ 512,
2533
+ 512,
2534
+ 512,
2535
+ 512,
2536
+ 512,
2537
+ 512,
2538
+ 512,
2539
+ 512,
2540
+ 512,
2541
+ 512,
2542
+ 512,
2543
+ 512,
2544
+ 512,
2545
+ 512,
2546
+ 512,
2547
+ 512,
2548
+ 512,
2549
+ 512,
2550
+ 512,
2551
+ 512,
2552
+ 512,
2553
+ 512,
2554
+ 512,
2555
+ 512,
2556
+ 512,
2557
+ 512,
2558
+ 512,
2559
+ 512,
2560
+ 512,
2561
+ 512,
2562
+ 512,
2563
+ 512,
2564
+ 512,
2565
+ 512,
2566
+ 512,
2567
+ 512,
2568
+ 512,
2569
+ 512,
2570
+ 512,
2571
+ 512,
2572
+ 512,
2573
+ 512,
2574
+ 512,
2575
+ 512,
2576
+ 512,
2577
+ 512,
2578
+ 512,
2579
+ 512,
2580
+ 512,
2581
+ 512,
2582
+ 512,
2583
+ 512,
2584
+ 512,
2585
+ 512,
2586
+ 512,
2587
+ 512,
2588
+ 512
2589
+ ],
2590
+ [
2591
+ 512,
2592
+ 512,
2593
+ 512,
2594
+ 512,
2595
+ 512,
2596
+ 512,
2597
+ 512,
2598
+ 512,
2599
+ 512,
2600
+ 512,
2601
+ 512,
2602
+ 512,
2603
+ 512,
2604
+ 512,
2605
+ 512,
2606
+ 512,
2607
+ 512,
2608
+ 512,
2609
+ 512,
2610
+ 512,
2611
+ 512,
2612
+ 512,
2613
+ 512,
2614
+ 512,
2615
+ 512,
2616
+ 512,
2617
+ 512,
2618
+ 512,
2619
+ 512,
2620
+ 512,
2621
+ 512,
2622
+ 512,
2623
+ 512,
2624
+ 512,
2625
+ 512,
2626
+ 512,
2627
+ 512,
2628
+ 512,
2629
+ 512,
2630
+ 512,
2631
+ 512,
2632
+ 512,
2633
+ 512,
2634
+ 512,
2635
+ 512,
2636
+ 512,
2637
+ 512,
2638
+ 512,
2639
+ 512,
2640
+ 512,
2641
+ 512,
2642
+ 512,
2643
+ 512,
2644
+ 512,
2645
+ 512,
2646
+ 512,
2647
+ 512,
2648
+ 512,
2649
+ 512,
2650
+ 512,
2651
+ 512,
2652
+ 512,
2653
+ 512,
2654
+ 512
2655
+ ]
2656
+ ],
2657
+ "full_attention_interval": 4,
2658
+ "glean_metadata": {
2659
+ "keep_fraction": 0.25,
2660
+ "method": "REAP whole-expert prune (reap), uniform per-layer keep",
2661
+ "score": {
2662
+ "alpha": 1,
2663
+ "b": 1.0,
2664
+ "beta": 1
2665
+ },
2666
+ "source_model": "Qwen/Qwen3.6-35B-A3B",
2667
+ "stats_path": "outputs/qwen35_stats1024.pt"
2668
+ },
2669
+ "head_dim": 256,
2670
+ "hidden_act": "silu",
2671
+ "hidden_size": 2048,
2672
+ "initializer_range": 0.02,
2673
+ "layer_types": [
2674
+ "linear_attention",
2675
+ "linear_attention",
2676
+ "linear_attention",
2677
+ "full_attention",
2678
+ "linear_attention",
2679
+ "linear_attention",
2680
+ "linear_attention",
2681
+ "full_attention",
2682
+ "linear_attention",
2683
+ "linear_attention",
2684
+ "linear_attention",
2685
+ "full_attention",
2686
+ "linear_attention",
2687
+ "linear_attention",
2688
+ "linear_attention",
2689
+ "full_attention",
2690
+ "linear_attention",
2691
+ "linear_attention",
2692
+ "linear_attention",
2693
+ "full_attention",
2694
+ "linear_attention",
2695
+ "linear_attention",
2696
+ "linear_attention",
2697
+ "full_attention",
2698
+ "linear_attention",
2699
+ "linear_attention",
2700
+ "linear_attention",
2701
+ "full_attention",
2702
+ "linear_attention",
2703
+ "linear_attention",
2704
+ "linear_attention",
2705
+ "full_attention",
2706
+ "linear_attention",
2707
+ "linear_attention",
2708
+ "linear_attention",
2709
+ "full_attention",
2710
+ "linear_attention",
2711
+ "linear_attention",
2712
+ "linear_attention",
2713
+ "full_attention"
2714
+ ],
2715
+ "linear_conv_kernel_dim": 4,
2716
+ "linear_key_head_dim": 128,
2717
+ "linear_num_key_heads": 16,
2718
+ "linear_num_value_heads": 32,
2719
+ "linear_value_head_dim": 128,
2720
+ "mamba_ssm_dtype": "float32",
2721
+ "max_position_embeddings": 262144,
2722
+ "model_type": "pruned_qwen3_5_moe",
2723
+ "moe_intermediate_size": 512,
2724
+ "mtp_num_hidden_layers": 1,
2725
+ "mtp_use_dedicated_embeddings": false,
2726
+ "num_attention_heads": 16,
2727
+ "num_experts": 256,
2728
+ "num_experts_per_tok": 8,
2729
+ "num_hidden_layers": 40,
2730
+ "num_key_value_heads": 2,
2731
+ "output_router_logits": false,
2732
+ "pad_token_id": null,
2733
+ "partial_rotary_factor": 0.25,
2734
+ "rms_norm_eps": 1e-06,
2735
+ "rope_parameters": {
2736
+ "mrope_interleaved": true,
2737
+ "mrope_section": [
2738
+ 11,
2739
+ 11,
2740
+ 10
2741
+ ],
2742
+ "partial_rotary_factor": 0.25,
2743
+ "rope_theta": 10000000,
2744
+ "rope_type": "default"
2745
+ },
2746
+ "router_aux_loss_coef": 0.001,
2747
+ "shared_expert_intermediate_size": 512,
2748
+ "tie_word_embeddings": false,
2749
+ "transformers_version": "5.13.1",
2750
+ "use_cache": true,
2751
+ "vocab_size": 248320
2752
+ }
qwen35_reap_keep25/configuration_pruned_qwen3_5_moe.py ADDED
@@ -0,0 +1,31 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ """Configuration for GLEAN-pruned Qwen3.5-MoE: variable-width, variable-count experts.
2
+
3
+ Requires the ``qwen3_5_moe`` architecture (transformers >= 5.x); checkpoints
4
+ carrying this module import the upstream classes rather than vendoring the
5
+ ~1.5k-line hybrid (GatedDeltaNet + attention) stack.
6
+ """
7
+
8
+ from transformers.models.qwen3_5_moe.configuration_qwen3_5_moe import (
9
+ Qwen3_5MoeTextConfig,
10
+ )
11
+
12
+
13
+ class PrunedQwen3_5MoeTextConfig(Qwen3_5MoeTextConfig):
14
+ """Qwen3_5MoeTextConfig plus a per-(layer, expert) width table.
15
+
16
+ ``expert_widths[l]`` lists the SwiGLU intermediate width of each surviving
17
+ routed expert in decoder layer ``l``, in original expert order. Lists are
18
+ ragged: layers may keep different numbers of experts (deleted experts
19
+ simply don't appear — the router in layer ``l`` has
20
+ ``len(expert_widths[l])`` rows), and each width may differ. ``None`` means
21
+ an unpruned model (uniform ``num_experts`` x ``moe_intermediate_size``).
22
+
23
+ The inherited ``num_experts`` / ``moe_intermediate_size`` keep their
24
+ ORIGINAL (pre-pruning) values for provenance; the width table is
25
+ authoritative for the built architecture. The shared expert and its
26
+ sigmoid gate are untouched by pruning and keep their stock config fields.
27
+ """
28
+
29
+ model_type = "pruned_qwen3_5_moe"
30
+
31
+ expert_widths: list[list[int]] | None = None
qwen35_reap_keep25/generation_config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_from_model_config": true,
3
+ "bos_token_id": 248044,
4
+ "eos_token_id": 248044,
5
+ "output_attentions": false,
6
+ "output_hidden_states": false,
7
+ "transformers_version": "5.13.1",
8
+ "use_cache": true
9
+ }
qwen35_reap_keep25/modeling_pruned_qwen3_5_moe.py ADDED
@@ -0,0 +1,122 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ """GLEAN-pruned Qwen3.5-MoE: HF-loadable model with ragged (variable-width) experts.
2
+
3
+ Same pattern as ``modeling_pruned_olmoe``: ``super().__init__`` builds the
4
+ uniform architecture from the config (free on the meta device during
5
+ ``from_pretrained``), then every MoE block's router and fused experts are
6
+ rebuilt to their pruned shape — surviving experts only, each at its own width,
7
+ router rows sliced to match. The token mixers (GatedDeltaNet / attention), the
8
+ shared expert, and its sigmoid gate are stock and untouched.
9
+
10
+ Unlike OLMoE's per-expert ``nn.Linear`` modules, Qwen3.5-MoE fuses experts
11
+ into 3-D parameters, which cannot hold ragged widths — so the pruned experts
12
+ module stores per-expert 2-D parameters in ``nn.ParameterList``s
13
+ (``gate_up_projs.{j}`` ``[2*w_j, H]``, ``down_projs.{j}`` ``[H, w_j]``) and
14
+ runs the same routed per-expert loop as upstream ``Qwen3_5MoeExperts.forward``
15
+ (verified against transformers 5.13.1; re-verify after upgrades).
16
+
17
+ Caveats: ``output_router_logits=True`` (the load-balancing aux loss) assumes a
18
+ uniform ``config.num_experts`` and is unsupported on ragged models; the eager
19
+ per-expert loop bypasses the fused ``_experts_implementation`` kernels.
20
+ """
21
+
22
+ import torch
23
+ import torch.nn.functional as F
24
+ from torch import nn
25
+ from transformers.activations import ACT2FN
26
+ from transformers.models.qwen3_5_moe.modeling_qwen3_5_moe import (
27
+ Qwen3_5MoeForCausalLM,
28
+ )
29
+
30
+ from .configuration_pruned_qwen3_5_moe import PrunedQwen3_5MoeTextConfig
31
+
32
+
33
+ class PrunedQwen3_5MoeExperts(nn.Module):
34
+ """Ragged replacement for the fused ``Qwen3_5MoeExperts``.
35
+
36
+ Forward replicates the upstream routed per-expert loop exactly, indexing
37
+ per-expert parameters instead of slices of a stacked 3-D tensor.
38
+ """
39
+
40
+ def __init__(self, config: PrunedQwen3_5MoeTextConfig, widths: list[int]):
41
+ super().__init__()
42
+ self.num_experts = len(widths)
43
+ self.hidden_dim = config.hidden_size
44
+ self.gate_up_projs = nn.ParameterList(
45
+ nn.Parameter(torch.empty(2 * w, config.hidden_size)) for w in widths
46
+ )
47
+ self.down_projs = nn.ParameterList(
48
+ nn.Parameter(torch.empty(config.hidden_size, w)) for w in widths
49
+ )
50
+ self.act_fn = ACT2FN[config.hidden_act]
51
+
52
+ def forward(
53
+ self,
54
+ hidden_states: torch.Tensor,
55
+ top_k_index: torch.Tensor,
56
+ top_k_weights: torch.Tensor,
57
+ ) -> torch.Tensor:
58
+ final_hidden_states = torch.zeros_like(hidden_states)
59
+ with torch.no_grad():
60
+ expert_mask = F.one_hot(top_k_index, num_classes=self.num_experts)
61
+ expert_mask = expert_mask.permute(2, 1, 0)
62
+ expert_hit = torch.greater(expert_mask.sum(dim=(-1, -2)), 0).nonzero()
63
+
64
+ for expert_idx in expert_hit:
65
+ expert_idx = int(expert_idx[0])
66
+ top_k_pos, token_idx = torch.where(expert_mask[expert_idx])
67
+ current_state = hidden_states[token_idx]
68
+ gate, up = F.linear(current_state, self.gate_up_projs[expert_idx]).chunk(2, dim=-1)
69
+ current_hidden_states = self.act_fn(gate) * up
70
+ current_hidden_states = F.linear(current_hidden_states, self.down_projs[expert_idx])
71
+ current_hidden_states = current_hidden_states * top_k_weights[token_idx, top_k_pos, None]
72
+ final_hidden_states.index_add_(
73
+ 0, token_idx, current_hidden_states.to(final_hidden_states.dtype)
74
+ )
75
+
76
+ return final_hidden_states
77
+
78
+
79
+ class PrunedQwen3_5MoeForCausalLM(Qwen3_5MoeForCausalLM):
80
+ """Qwen3.5-MoE with per-layer surviving-expert lists at per-expert widths."""
81
+
82
+ config_class = PrunedQwen3_5MoeTextConfig
83
+
84
+ def __init__(self, config: PrunedQwen3_5MoeTextConfig):
85
+ super().__init__(config)
86
+ widths_table = getattr(config, "expert_widths", None)
87
+ if widths_table is None:
88
+ return # unpruned: plain Qwen3.5-MoE
89
+ if len(widths_table) != len(self.model.layers):
90
+ raise ValueError(
91
+ f"expert_widths has {len(widths_table)} rows but the model has "
92
+ f"{len(self.model.layers)} decoder layers"
93
+ )
94
+ for layer, widths in zip(self.model.layers, widths_table):
95
+ if any(w <= 0 for w in widths):
96
+ raise ValueError("expert_widths must list surviving experts only (>0)")
97
+ block = layer.mlp
98
+ if len(widths) < block.gate.top_k:
99
+ raise ValueError(
100
+ f"a layer keeps {len(widths)} experts < top_k={block.gate.top_k}"
101
+ )
102
+ # keep the stock router class (OutputRecorder isinstance, forward
103
+ # unchanged) but shrink it to the surviving experts' rows
104
+ block.gate.num_experts = len(widths)
105
+ block.gate.weight = nn.Parameter(
106
+ torch.empty(
107
+ len(widths),
108
+ config.hidden_size,
109
+ dtype=block.gate.weight.dtype,
110
+ device=block.gate.weight.device,
111
+ )
112
+ )
113
+ block.experts = PrunedQwen3_5MoeExperts(config, list(widths))
114
+
115
+ @torch.no_grad()
116
+ def _init_weights(self, module):
117
+ super()._init_weights(module)
118
+ # the base isinstance(Qwen3_5MoeExperts) branch never sees our ragged
119
+ # module; raw nn.Parameters get no default init otherwise
120
+ if isinstance(module, PrunedQwen3_5MoeExperts):
121
+ for p in list(module.gate_up_projs) + list(module.down_projs):
122
+ nn.init.normal_(p, mean=0.0, std=self.config.initializer_range)
qwen35_reap_keep25/reap_verify.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "keep": 0.25,
3
+ "criterion": "reap",
4
+ "ppl": 22.945579528808594,
5
+ "n_params_b": 10.485691008,
6
+ "eval_seq": 32,
7
+ "seq_len": 2048,
8
+ "dataset": "c4"
9
+ }
qwen35_reap_keep25/tokenizer_config.json ADDED
@@ -0,0 +1,32 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_prefix_space": false,
3
+ "audio_bos_token": "<|audio_start|>",
4
+ "audio_eos_token": "<|audio_end|>",
5
+ "audio_token": "<|audio_pad|>",
6
+ "backend": "tokenizers",
7
+ "bos_token": null,
8
+ "clean_up_tokenization_spaces": false,
9
+ "eos_token": "<|im_end|>",
10
+ "errors": "replace",
11
+ "image_token": "<|image_pad|>",
12
+ "is_local": false,
13
+ "local_files_only": false,
14
+ "model_max_length": 262144,
15
+ "model_specific_special_tokens": {
16
+ "audio_bos_token": "<|audio_start|>",
17
+ "audio_eos_token": "<|audio_end|>",
18
+ "audio_token": "<|audio_pad|>",
19
+ "image_token": "<|image_pad|>",
20
+ "video_token": "<|video_pad|>",
21
+ "vision_bos_token": "<|vision_start|>",
22
+ "vision_eos_token": "<|vision_end|>"
23
+ },
24
+ "pad_token": "<|endoftext|>",
25
+ "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
26
+ "split_special_tokens": false,
27
+ "tokenizer_class": "Qwen2Tokenizer",
28
+ "unk_token": null,
29
+ "video_token": "<|video_pad|>",
30
+ "vision_bos_token": "<|vision_start|>",
31
+ "vision_eos_token": "<|vision_end|>"
32
+ }
qwen35_reap_keep50/chat_template.jinja ADDED
@@ -0,0 +1,154 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {%- set image_count = namespace(value=0) %}
2
+ {%- set video_count = namespace(value=0) %}
3
+ {%- macro render_content(content, do_vision_count, is_system_content=false) %}
4
+ {%- if content is string %}
5
+ {{- content }}
6
+ {%- elif content is iterable and content is not mapping %}
7
+ {%- for item in content %}
8
+ {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
9
+ {%- if is_system_content %}
10
+ {{- raise_exception('System message cannot contain images.') }}
11
+ {%- endif %}
12
+ {%- if do_vision_count %}
13
+ {%- set image_count.value = image_count.value + 1 %}
14
+ {%- endif %}
15
+ {%- if add_vision_id %}
16
+ {{- 'Picture ' ~ image_count.value ~ ': ' }}
17
+ {%- endif %}
18
+ {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
19
+ {%- elif 'video' in item or item.type == 'video' %}
20
+ {%- if is_system_content %}
21
+ {{- raise_exception('System message cannot contain videos.') }}
22
+ {%- endif %}
23
+ {%- if do_vision_count %}
24
+ {%- set video_count.value = video_count.value + 1 %}
25
+ {%- endif %}
26
+ {%- if add_vision_id %}
27
+ {{- 'Video ' ~ video_count.value ~ ': ' }}
28
+ {%- endif %}
29
+ {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
30
+ {%- elif 'text' in item %}
31
+ {{- item.text }}
32
+ {%- else %}
33
+ {{- raise_exception('Unexpected item type in content.') }}
34
+ {%- endif %}
35
+ {%- endfor %}
36
+ {%- elif content is none or content is undefined %}
37
+ {{- '' }}
38
+ {%- else %}
39
+ {{- raise_exception('Unexpected content type.') }}
40
+ {%- endif %}
41
+ {%- endmacro %}
42
+ {%- if not messages %}
43
+ {{- raise_exception('No messages provided.') }}
44
+ {%- endif %}
45
+ {%- if tools and tools is iterable and tools is not mapping %}
46
+ {{- '<|im_start|>system\n' }}
47
+ {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
48
+ {%- for tool in tools %}
49
+ {{- "\n" }}
50
+ {{- tool | tojson }}
51
+ {%- endfor %}
52
+ {{- "\n</tools>" }}
53
+ {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
54
+ {%- if messages[0].role == 'system' %}
55
+ {%- set content = render_content(messages[0].content, false, true)|trim %}
56
+ {%- if content %}
57
+ {{- '\n\n' + content }}
58
+ {%- endif %}
59
+ {%- endif %}
60
+ {{- '<|im_end|>\n' }}
61
+ {%- else %}
62
+ {%- if messages[0].role == 'system' %}
63
+ {%- set content = render_content(messages[0].content, false, true)|trim %}
64
+ {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
65
+ {%- endif %}
66
+ {%- endif %}
67
+ {%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
68
+ {%- for message in messages[::-1] %}
69
+ {%- set index = (messages|length - 1) - loop.index0 %}
70
+ {%- if ns.multi_step_tool and message.role == "user" %}
71
+ {%- set content = render_content(message.content, false)|trim %}
72
+ {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
73
+ {%- set ns.multi_step_tool = false %}
74
+ {%- set ns.last_query_index = index %}
75
+ {%- endif %}
76
+ {%- endif %}
77
+ {%- endfor %}
78
+ {%- if ns.multi_step_tool %}
79
+ {{- raise_exception('No user query found in messages.') }}
80
+ {%- endif %}
81
+ {%- for message in messages %}
82
+ {%- set content = render_content(message.content, true)|trim %}
83
+ {%- if message.role == "system" %}
84
+ {%- if not loop.first %}
85
+ {{- raise_exception('System message must be at the beginning.') }}
86
+ {%- endif %}
87
+ {%- elif message.role == "user" %}
88
+ {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
89
+ {%- elif message.role == "assistant" %}
90
+ {%- set reasoning_content = '' %}
91
+ {%- if message.reasoning_content is string %}
92
+ {%- set reasoning_content = message.reasoning_content %}
93
+ {%- else %}
94
+ {%- if '</think>' in content %}
95
+ {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
96
+ {%- set content = content.split('</think>')[-1].lstrip('\n') %}
97
+ {%- endif %}
98
+ {%- endif %}
99
+ {%- set reasoning_content = reasoning_content|trim %}
100
+ {%- if (preserve_thinking is defined and preserve_thinking is true) or (loop.index0 > ns.last_query_index) %}
101
+ {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
102
+ {%- else %}
103
+ {{- '<|im_start|>' + message.role + '\n' + content }}
104
+ {%- endif %}
105
+ {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
106
+ {%- for tool_call in message.tool_calls %}
107
+ {%- if tool_call.function is defined %}
108
+ {%- set tool_call = tool_call.function %}
109
+ {%- endif %}
110
+ {%- if loop.first %}
111
+ {%- if content|trim %}
112
+ {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
113
+ {%- else %}
114
+ {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
115
+ {%- endif %}
116
+ {%- else %}
117
+ {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
118
+ {%- endif %}
119
+ {%- if tool_call.arguments is defined %}
120
+ {%- for args_name, args_value in tool_call.arguments|items %}
121
+ {{- '<parameter=' + args_name + '>\n' }}
122
+ {%- set args_value = args_value | string if args_value is string else args_value | tojson | safe %}
123
+ {{- args_value }}
124
+ {{- '\n</parameter>\n' }}
125
+ {%- endfor %}
126
+ {%- endif %}
127
+ {{- '</function>\n</tool_call>' }}
128
+ {%- endfor %}
129
+ {%- endif %}
130
+ {{- '<|im_end|>\n' }}
131
+ {%- elif message.role == "tool" %}
132
+ {%- if loop.previtem and loop.previtem.role != "tool" %}
133
+ {{- '<|im_start|>user' }}
134
+ {%- endif %}
135
+ {{- '\n<tool_response>\n' }}
136
+ {{- content }}
137
+ {{- '\n</tool_response>' }}
138
+ {%- if not loop.last and loop.nextitem.role != "tool" %}
139
+ {{- '<|im_end|>\n' }}
140
+ {%- elif loop.last %}
141
+ {{- '<|im_end|>\n' }}
142
+ {%- endif %}
143
+ {%- else %}
144
+ {{- raise_exception('Unexpected message role.') }}
145
+ {%- endif %}
146
+ {%- endfor %}
147
+ {%- if add_generation_prompt %}
148
+ {{- '<|im_start|>assistant\n' }}
149
+ {%- if enable_thinking is defined and enable_thinking is false %}
150
+ {{- '<think>\n\n</think>\n\n' }}
151
+ {%- else %}
152
+ {{- '<think>\n' }}
153
+ {%- endif %}
154
+ {%- endif %}
qwen35_reap_keep50/config.json ADDED
@@ -0,0 +1,5312 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "PrunedQwen3_5MoeForCausalLM"
4
+ ],
5
+ "attention_bias": false,
6
+ "attention_dropout": 0.0,
7
+ "attn_output_gate": true,
8
+ "auto_map": {
9
+ "AutoConfig": "configuration_pruned_qwen3_5_moe.PrunedQwen3_5MoeTextConfig",
10
+ "AutoModelForCausalLM": "modeling_pruned_qwen3_5_moe.PrunedQwen3_5MoeForCausalLM"
11
+ },
12
+ "bos_token_id": 248044,
13
+ "dtype": "bfloat16",
14
+ "eos_token_id": 248044,
15
+ "expert_widths": [
16
+ [
17
+ 512,
18
+ 512,
19
+ 512,
20
+ 512,
21
+ 512,
22
+ 512,
23
+ 512,
24
+ 512,
25
+ 512,
26
+ 512,
27
+ 512,
28
+ 512,
29
+ 512,
30
+ 512,
31
+ 512,
32
+ 512,
33
+ 512,
34
+ 512,
35
+ 512,
36
+ 512,
37
+ 512,
38
+ 512,
39
+ 512,
40
+ 512,
41
+ 512,
42
+ 512,
43
+ 512,
44
+ 512,
45
+ 512,
46
+ 512,
47
+ 512,
48
+ 512,
49
+ 512,
50
+ 512,
51
+ 512,
52
+ 512,
53
+ 512,
54
+ 512,
55
+ 512,
56
+ 512,
57
+ 512,
58
+ 512,
59
+ 512,
60
+ 512,
61
+ 512,
62
+ 512,
63
+ 512,
64
+ 512,
65
+ 512,
66
+ 512,
67
+ 512,
68
+ 512,
69
+ 512,
70
+ 512,
71
+ 512,
72
+ 512,
73
+ 512,
74
+ 512,
75
+ 512,
76
+ 512,
77
+ 512,
78
+ 512,
79
+ 512,
80
+ 512,
81
+ 512,
82
+ 512,
83
+ 512,
84
+ 512,
85
+ 512,
86
+ 512,
87
+ 512,
88
+ 512,
89
+ 512,
90
+ 512,
91
+ 512,
92
+ 512,
93
+ 512,
94
+ 512,
95
+ 512,
96
+ 512,
97
+ 512,
98
+ 512,
99
+ 512,
100
+ 512,
101
+ 512,
102
+ 512,
103
+ 512,
104
+ 512,
105
+ 512,
106
+ 512,
107
+ 512,
108
+ 512,
109
+ 512,
110
+ 512,
111
+ 512,
112
+ 512,
113
+ 512,
114
+ 512,
115
+ 512,
116
+ 512,
117
+ 512,
118
+ 512,
119
+ 512,
120
+ 512,
121
+ 512,
122
+ 512,
123
+ 512,
124
+ 512,
125
+ 512,
126
+ 512,
127
+ 512,
128
+ 512,
129
+ 512,
130
+ 512,
131
+ 512,
132
+ 512,
133
+ 512,
134
+ 512,
135
+ 512,
136
+ 512,
137
+ 512,
138
+ 512,
139
+ 512,
140
+ 512,
141
+ 512,
142
+ 512,
143
+ 512,
144
+ 512
145
+ ],
146
+ [
147
+ 512,
148
+ 512,
149
+ 512,
150
+ 512,
151
+ 512,
152
+ 512,
153
+ 512,
154
+ 512,
155
+ 512,
156
+ 512,
157
+ 512,
158
+ 512,
159
+ 512,
160
+ 512,
161
+ 512,
162
+ 512,
163
+ 512,
164
+ 512,
165
+ 512,
166
+ 512,
167
+ 512,
168
+ 512,
169
+ 512,
170
+ 512,
171
+ 512,
172
+ 512,
173
+ 512,
174
+ 512,
175
+ 512,
176
+ 512,
177
+ 512,
178
+ 512,
179
+ 512,
180
+ 512,
181
+ 512,
182
+ 512,
183
+ 512,
184
+ 512,
185
+ 512,
186
+ 512,
187
+ 512,
188
+ 512,
189
+ 512,
190
+ 512,
191
+ 512,
192
+ 512,
193
+ 512,
194
+ 512,
195
+ 512,
196
+ 512,
197
+ 512,
198
+ 512,
199
+ 512,
200
+ 512,
201
+ 512,
202
+ 512,
203
+ 512,
204
+ 512,
205
+ 512,
206
+ 512,
207
+ 512,
208
+ 512,
209
+ 512,
210
+ 512,
211
+ 512,
212
+ 512,
213
+ 512,
214
+ 512,
215
+ 512,
216
+ 512,
217
+ 512,
218
+ 512,
219
+ 512,
220
+ 512,
221
+ 512,
222
+ 512,
223
+ 512,
224
+ 512,
225
+ 512,
226
+ 512,
227
+ 512,
228
+ 512,
229
+ 512,
230
+ 512,
231
+ 512,
232
+ 512,
233
+ 512,
234
+ 512,
235
+ 512,
236
+ 512,
237
+ 512,
238
+ 512,
239
+ 512,
240
+ 512,
241
+ 512,
242
+ 512,
243
+ 512,
244
+ 512,
245
+ 512,
246
+ 512,
247
+ 512,
248
+ 512,
249
+ 512,
250
+ 512,
251
+ 512,
252
+ 512,
253
+ 512,
254
+ 512,
255
+ 512,
256
+ 512,
257
+ 512,
258
+ 512,
259
+ 512,
260
+ 512,
261
+ 512,
262
+ 512,
263
+ 512,
264
+ 512,
265
+ 512,
266
+ 512,
267
+ 512,
268
+ 512,
269
+ 512,
270
+ 512,
271
+ 512,
272
+ 512,
273
+ 512,
274
+ 512
275
+ ],
276
+ [
277
+ 512,
278
+ 512,
279
+ 512,
280
+ 512,
281
+ 512,
282
+ 512,
283
+ 512,
284
+ 512,
285
+ 512,
286
+ 512,
287
+ 512,
288
+ 512,
289
+ 512,
290
+ 512,
291
+ 512,
292
+ 512,
293
+ 512,
294
+ 512,
295
+ 512,
296
+ 512,
297
+ 512,
298
+ 512,
299
+ 512,
300
+ 512,
301
+ 512,
302
+ 512,
303
+ 512,
304
+ 512,
305
+ 512,
306
+ 512,
307
+ 512,
308
+ 512,
309
+ 512,
310
+ 512,
311
+ 512,
312
+ 512,
313
+ 512,
314
+ 512,
315
+ 512,
316
+ 512,
317
+ 512,
318
+ 512,
319
+ 512,
320
+ 512,
321
+ 512,
322
+ 512,
323
+ 512,
324
+ 512,
325
+ 512,
326
+ 512,
327
+ 512,
328
+ 512,
329
+ 512,
330
+ 512,
331
+ 512,
332
+ 512,
333
+ 512,
334
+ 512,
335
+ 512,
336
+ 512,
337
+ 512,
338
+ 512,
339
+ 512,
340
+ 512,
341
+ 512,
342
+ 512,
343
+ 512,
344
+ 512,
345
+ 512,
346
+ 512,
347
+ 512,
348
+ 512,
349
+ 512,
350
+ 512,
351
+ 512,
352
+ 512,
353
+ 512,
354
+ 512,
355
+ 512,
356
+ 512,
357
+ 512,
358
+ 512,
359
+ 512,
360
+ 512,
361
+ 512,
362
+ 512,
363
+ 512,
364
+ 512,
365
+ 512,
366
+ 512,
367
+ 512,
368
+ 512,
369
+ 512,
370
+ 512,
371
+ 512,
372
+ 512,
373
+ 512,
374
+ 512,
375
+ 512,
376
+ 512,
377
+ 512,
378
+ 512,
379
+ 512,
380
+ 512,
381
+ 512,
382
+ 512,
383
+ 512,
384
+ 512,
385
+ 512,
386
+ 512,
387
+ 512,
388
+ 512,
389
+ 512,
390
+ 512,
391
+ 512,
392
+ 512,
393
+ 512,
394
+ 512,
395
+ 512,
396
+ 512,
397
+ 512,
398
+ 512,
399
+ 512,
400
+ 512,
401
+ 512,
402
+ 512,
403
+ 512,
404
+ 512
405
+ ],
406
+ [
407
+ 512,
408
+ 512,
409
+ 512,
410
+ 512,
411
+ 512,
412
+ 512,
413
+ 512,
414
+ 512,
415
+ 512,
416
+ 512,
417
+ 512,
418
+ 512,
419
+ 512,
420
+ 512,
421
+ 512,
422
+ 512,
423
+ 512,
424
+ 512,
425
+ 512,
426
+ 512,
427
+ 512,
428
+ 512,
429
+ 512,
430
+ 512,
431
+ 512,
432
+ 512,
433
+ 512,
434
+ 512,
435
+ 512,
436
+ 512,
437
+ 512,
438
+ 512,
439
+ 512,
440
+ 512,
441
+ 512,
442
+ 512,
443
+ 512,
444
+ 512,
445
+ 512,
446
+ 512,
447
+ 512,
448
+ 512,
449
+ 512,
450
+ 512,
451
+ 512,
452
+ 512,
453
+ 512,
454
+ 512,
455
+ 512,
456
+ 512,
457
+ 512,
458
+ 512,
459
+ 512,
460
+ 512,
461
+ 512,
462
+ 512,
463
+ 512,
464
+ 512,
465
+ 512,
466
+ 512,
467
+ 512,
468
+ 512,
469
+ 512,
470
+ 512,
471
+ 512,
472
+ 512,
473
+ 512,
474
+ 512,
475
+ 512,
476
+ 512,
477
+ 512,
478
+ 512,
479
+ 512,
480
+ 512,
481
+ 512,
482
+ 512,
483
+ 512,
484
+ 512,
485
+ 512,
486
+ 512,
487
+ 512,
488
+ 512,
489
+ 512,
490
+ 512,
491
+ 512,
492
+ 512,
493
+ 512,
494
+ 512,
495
+ 512,
496
+ 512,
497
+ 512,
498
+ 512,
499
+ 512,
500
+ 512,
501
+ 512,
502
+ 512,
503
+ 512,
504
+ 512,
505
+ 512,
506
+ 512,
507
+ 512,
508
+ 512,
509
+ 512,
510
+ 512,
511
+ 512,
512
+ 512,
513
+ 512,
514
+ 512,
515
+ 512,
516
+ 512,
517
+ 512,
518
+ 512,
519
+ 512,
520
+ 512,
521
+ 512,
522
+ 512,
523
+ 512,
524
+ 512,
525
+ 512,
526
+ 512,
527
+ 512,
528
+ 512,
529
+ 512,
530
+ 512,
531
+ 512,
532
+ 512,
533
+ 512,
534
+ 512
535
+ ],
536
+ [
537
+ 512,
538
+ 512,
539
+ 512,
540
+ 512,
541
+ 512,
542
+ 512,
543
+ 512,
544
+ 512,
545
+ 512,
546
+ 512,
547
+ 512,
548
+ 512,
549
+ 512,
550
+ 512,
551
+ 512,
552
+ 512,
553
+ 512,
554
+ 512,
555
+ 512,
556
+ 512,
557
+ 512,
558
+ 512,
559
+ 512,
560
+ 512,
561
+ 512,
562
+ 512,
563
+ 512,
564
+ 512,
565
+ 512,
566
+ 512,
567
+ 512,
568
+ 512,
569
+ 512,
570
+ 512,
571
+ 512,
572
+ 512,
573
+ 512,
574
+ 512,
575
+ 512,
576
+ 512,
577
+ 512,
578
+ 512,
579
+ 512,
580
+ 512,
581
+ 512,
582
+ 512,
583
+ 512,
584
+ 512,
585
+ 512,
586
+ 512,
587
+ 512,
588
+ 512,
589
+ 512,
590
+ 512,
591
+ 512,
592
+ 512,
593
+ 512,
594
+ 512,
595
+ 512,
596
+ 512,
597
+ 512,
598
+ 512,
599
+ 512,
600
+ 512,
601
+ 512,
602
+ 512,
603
+ 512,
604
+ 512,
605
+ 512,
606
+ 512,
607
+ 512,
608
+ 512,
609
+ 512,
610
+ 512,
611
+ 512,
612
+ 512,
613
+ 512,
614
+ 512,
615
+ 512,
616
+ 512,
617
+ 512,
618
+ 512,
619
+ 512,
620
+ 512,
621
+ 512,
622
+ 512,
623
+ 512,
624
+ 512,
625
+ 512,
626
+ 512,
627
+ 512,
628
+ 512,
629
+ 512,
630
+ 512,
631
+ 512,
632
+ 512,
633
+ 512,
634
+ 512,
635
+ 512,
636
+ 512,
637
+ 512,
638
+ 512,
639
+ 512,
640
+ 512,
641
+ 512,
642
+ 512,
643
+ 512,
644
+ 512,
645
+ 512,
646
+ 512,
647
+ 512,
648
+ 512,
649
+ 512,
650
+ 512,
651
+ 512,
652
+ 512,
653
+ 512,
654
+ 512,
655
+ 512,
656
+ 512,
657
+ 512,
658
+ 512,
659
+ 512,
660
+ 512,
661
+ 512,
662
+ 512,
663
+ 512,
664
+ 512
665
+ ],
666
+ [
667
+ 512,
668
+ 512,
669
+ 512,
670
+ 512,
671
+ 512,
672
+ 512,
673
+ 512,
674
+ 512,
675
+ 512,
676
+ 512,
677
+ 512,
678
+ 512,
679
+ 512,
680
+ 512,
681
+ 512,
682
+ 512,
683
+ 512,
684
+ 512,
685
+ 512,
686
+ 512,
687
+ 512,
688
+ 512,
689
+ 512,
690
+ 512,
691
+ 512,
692
+ 512,
693
+ 512,
694
+ 512,
695
+ 512,
696
+ 512,
697
+ 512,
698
+ 512,
699
+ 512,
700
+ 512,
701
+ 512,
702
+ 512,
703
+ 512,
704
+ 512,
705
+ 512,
706
+ 512,
707
+ 512,
708
+ 512,
709
+ 512,
710
+ 512,
711
+ 512,
712
+ 512,
713
+ 512,
714
+ 512,
715
+ 512,
716
+ 512,
717
+ 512,
718
+ 512,
719
+ 512,
720
+ 512,
721
+ 512,
722
+ 512,
723
+ 512,
724
+ 512,
725
+ 512,
726
+ 512,
727
+ 512,
728
+ 512,
729
+ 512,
730
+ 512,
731
+ 512,
732
+ 512,
733
+ 512,
734
+ 512,
735
+ 512,
736
+ 512,
737
+ 512,
738
+ 512,
739
+ 512,
740
+ 512,
741
+ 512,
742
+ 512,
743
+ 512,
744
+ 512,
745
+ 512,
746
+ 512,
747
+ 512,
748
+ 512,
749
+ 512,
750
+ 512,
751
+ 512,
752
+ 512,
753
+ 512,
754
+ 512,
755
+ 512,
756
+ 512,
757
+ 512,
758
+ 512,
759
+ 512,
760
+ 512,
761
+ 512,
762
+ 512,
763
+ 512,
764
+ 512,
765
+ 512,
766
+ 512,
767
+ 512,
768
+ 512,
769
+ 512,
770
+ 512,
771
+ 512,
772
+ 512,
773
+ 512,
774
+ 512,
775
+ 512,
776
+ 512,
777
+ 512,
778
+ 512,
779
+ 512,
780
+ 512,
781
+ 512,
782
+ 512,
783
+ 512,
784
+ 512,
785
+ 512,
786
+ 512,
787
+ 512,
788
+ 512,
789
+ 512,
790
+ 512,
791
+ 512,
792
+ 512,
793
+ 512,
794
+ 512
795
+ ],
796
+ [
797
+ 512,
798
+ 512,
799
+ 512,
800
+ 512,
801
+ 512,
802
+ 512,
803
+ 512,
804
+ 512,
805
+ 512,
806
+ 512,
807
+ 512,
808
+ 512,
809
+ 512,
810
+ 512,
811
+ 512,
812
+ 512,
813
+ 512,
814
+ 512,
815
+ 512,
816
+ 512,
817
+ 512,
818
+ 512,
819
+ 512,
820
+ 512,
821
+ 512,
822
+ 512,
823
+ 512,
824
+ 512,
825
+ 512,
826
+ 512,
827
+ 512,
828
+ 512,
829
+ 512,
830
+ 512,
831
+ 512,
832
+ 512,
833
+ 512,
834
+ 512,
835
+ 512,
836
+ 512,
837
+ 512,
838
+ 512,
839
+ 512,
840
+ 512,
841
+ 512,
842
+ 512,
843
+ 512,
844
+ 512,
845
+ 512,
846
+ 512,
847
+ 512,
848
+ 512,
849
+ 512,
850
+ 512,
851
+ 512,
852
+ 512,
853
+ 512,
854
+ 512,
855
+ 512,
856
+ 512,
857
+ 512,
858
+ 512,
859
+ 512,
860
+ 512,
861
+ 512,
862
+ 512,
863
+ 512,
864
+ 512,
865
+ 512,
866
+ 512,
867
+ 512,
868
+ 512,
869
+ 512,
870
+ 512,
871
+ 512,
872
+ 512,
873
+ 512,
874
+ 512,
875
+ 512,
876
+ 512,
877
+ 512,
878
+ 512,
879
+ 512,
880
+ 512,
881
+ 512,
882
+ 512,
883
+ 512,
884
+ 512,
885
+ 512,
886
+ 512,
887
+ 512,
888
+ 512,
889
+ 512,
890
+ 512,
891
+ 512,
892
+ 512,
893
+ 512,
894
+ 512,
895
+ 512,
896
+ 512,
897
+ 512,
898
+ 512,
899
+ 512,
900
+ 512,
901
+ 512,
902
+ 512,
903
+ 512,
904
+ 512,
905
+ 512,
906
+ 512,
907
+ 512,
908
+ 512,
909
+ 512,
910
+ 512,
911
+ 512,
912
+ 512,
913
+ 512,
914
+ 512,
915
+ 512,
916
+ 512,
917
+ 512,
918
+ 512,
919
+ 512,
920
+ 512,
921
+ 512,
922
+ 512,
923
+ 512,
924
+ 512
925
+ ],
926
+ [
927
+ 512,
928
+ 512,
929
+ 512,
930
+ 512,
931
+ 512,
932
+ 512,
933
+ 512,
934
+ 512,
935
+ 512,
936
+ 512,
937
+ 512,
938
+ 512,
939
+ 512,
940
+ 512,
941
+ 512,
942
+ 512,
943
+ 512,
944
+ 512,
945
+ 512,
946
+ 512,
947
+ 512,
948
+ 512,
949
+ 512,
950
+ 512,
951
+ 512,
952
+ 512,
953
+ 512,
954
+ 512,
955
+ 512,
956
+ 512,
957
+ 512,
958
+ 512,
959
+ 512,
960
+ 512,
961
+ 512,
962
+ 512,
963
+ 512,
964
+ 512,
965
+ 512,
966
+ 512,
967
+ 512,
968
+ 512,
969
+ 512,
970
+ 512,
971
+ 512,
972
+ 512,
973
+ 512,
974
+ 512,
975
+ 512,
976
+ 512,
977
+ 512,
978
+ 512,
979
+ 512,
980
+ 512,
981
+ 512,
982
+ 512,
983
+ 512,
984
+ 512,
985
+ 512,
986
+ 512,
987
+ 512,
988
+ 512,
989
+ 512,
990
+ 512,
991
+ 512,
992
+ 512,
993
+ 512,
994
+ 512,
995
+ 512,
996
+ 512,
997
+ 512,
998
+ 512,
999
+ 512,
1000
+ 512,
1001
+ 512,
1002
+ 512,
1003
+ 512,
1004
+ 512,
1005
+ 512,
1006
+ 512,
1007
+ 512,
1008
+ 512,
1009
+ 512,
1010
+ 512,
1011
+ 512,
1012
+ 512,
1013
+ 512,
1014
+ 512,
1015
+ 512,
1016
+ 512,
1017
+ 512,
1018
+ 512,
1019
+ 512,
1020
+ 512,
1021
+ 512,
1022
+ 512,
1023
+ 512,
1024
+ 512,
1025
+ 512,
1026
+ 512,
1027
+ 512,
1028
+ 512,
1029
+ 512,
1030
+ 512,
1031
+ 512,
1032
+ 512,
1033
+ 512,
1034
+ 512,
1035
+ 512,
1036
+ 512,
1037
+ 512,
1038
+ 512,
1039
+ 512,
1040
+ 512,
1041
+ 512,
1042
+ 512,
1043
+ 512,
1044
+ 512,
1045
+ 512,
1046
+ 512,
1047
+ 512,
1048
+ 512,
1049
+ 512,
1050
+ 512,
1051
+ 512,
1052
+ 512,
1053
+ 512,
1054
+ 512
1055
+ ],
1056
+ [
1057
+ 512,
1058
+ 512,
1059
+ 512,
1060
+ 512,
1061
+ 512,
1062
+ 512,
1063
+ 512,
1064
+ 512,
1065
+ 512,
1066
+ 512,
1067
+ 512,
1068
+ 512,
1069
+ 512,
1070
+ 512,
1071
+ 512,
1072
+ 512,
1073
+ 512,
1074
+ 512,
1075
+ 512,
1076
+ 512,
1077
+ 512,
1078
+ 512,
1079
+ 512,
1080
+ 512,
1081
+ 512,
1082
+ 512,
1083
+ 512,
1084
+ 512,
1085
+ 512,
1086
+ 512,
1087
+ 512,
1088
+ 512,
1089
+ 512,
1090
+ 512,
1091
+ 512,
1092
+ 512,
1093
+ 512,
1094
+ 512,
1095
+ 512,
1096
+ 512,
1097
+ 512,
1098
+ 512,
1099
+ 512,
1100
+ 512,
1101
+ 512,
1102
+ 512,
1103
+ 512,
1104
+ 512,
1105
+ 512,
1106
+ 512,
1107
+ 512,
1108
+ 512,
1109
+ 512,
1110
+ 512,
1111
+ 512,
1112
+ 512,
1113
+ 512,
1114
+ 512,
1115
+ 512,
1116
+ 512,
1117
+ 512,
1118
+ 512,
1119
+ 512,
1120
+ 512,
1121
+ 512,
1122
+ 512,
1123
+ 512,
1124
+ 512,
1125
+ 512,
1126
+ 512,
1127
+ 512,
1128
+ 512,
1129
+ 512,
1130
+ 512,
1131
+ 512,
1132
+ 512,
1133
+ 512,
1134
+ 512,
1135
+ 512,
1136
+ 512,
1137
+ 512,
1138
+ 512,
1139
+ 512,
1140
+ 512,
1141
+ 512,
1142
+ 512,
1143
+ 512,
1144
+ 512,
1145
+ 512,
1146
+ 512,
1147
+ 512,
1148
+ 512,
1149
+ 512,
1150
+ 512,
1151
+ 512,
1152
+ 512,
1153
+ 512,
1154
+ 512,
1155
+ 512,
1156
+ 512,
1157
+ 512,
1158
+ 512,
1159
+ 512,
1160
+ 512,
1161
+ 512,
1162
+ 512,
1163
+ 512,
1164
+ 512,
1165
+ 512,
1166
+ 512,
1167
+ 512,
1168
+ 512,
1169
+ 512,
1170
+ 512,
1171
+ 512,
1172
+ 512,
1173
+ 512,
1174
+ 512,
1175
+ 512,
1176
+ 512,
1177
+ 512,
1178
+ 512,
1179
+ 512,
1180
+ 512,
1181
+ 512,
1182
+ 512,
1183
+ 512,
1184
+ 512
1185
+ ],
1186
+ [
1187
+ 512,
1188
+ 512,
1189
+ 512,
1190
+ 512,
1191
+ 512,
1192
+ 512,
1193
+ 512,
1194
+ 512,
1195
+ 512,
1196
+ 512,
1197
+ 512,
1198
+ 512,
1199
+ 512,
1200
+ 512,
1201
+ 512,
1202
+ 512,
1203
+ 512,
1204
+ 512,
1205
+ 512,
1206
+ 512,
1207
+ 512,
1208
+ 512,
1209
+ 512,
1210
+ 512,
1211
+ 512,
1212
+ 512,
1213
+ 512,
1214
+ 512,
1215
+ 512,
1216
+ 512,
1217
+ 512,
1218
+ 512,
1219
+ 512,
1220
+ 512,
1221
+ 512,
1222
+ 512,
1223
+ 512,
1224
+ 512,
1225
+ 512,
1226
+ 512,
1227
+ 512,
1228
+ 512,
1229
+ 512,
1230
+ 512,
1231
+ 512,
1232
+ 512,
1233
+ 512,
1234
+ 512,
1235
+ 512,
1236
+ 512,
1237
+ 512,
1238
+ 512,
1239
+ 512,
1240
+ 512,
1241
+ 512,
1242
+ 512,
1243
+ 512,
1244
+ 512,
1245
+ 512,
1246
+ 512,
1247
+ 512,
1248
+ 512,
1249
+ 512,
1250
+ 512,
1251
+ 512,
1252
+ 512,
1253
+ 512,
1254
+ 512,
1255
+ 512,
1256
+ 512,
1257
+ 512,
1258
+ 512,
1259
+ 512,
1260
+ 512,
1261
+ 512,
1262
+ 512,
1263
+ 512,
1264
+ 512,
1265
+ 512,
1266
+ 512,
1267
+ 512,
1268
+ 512,
1269
+ 512,
1270
+ 512,
1271
+ 512,
1272
+ 512,
1273
+ 512,
1274
+ 512,
1275
+ 512,
1276
+ 512,
1277
+ 512,
1278
+ 512,
1279
+ 512,
1280
+ 512,
1281
+ 512,
1282
+ 512,
1283
+ 512,
1284
+ 512,
1285
+ 512,
1286
+ 512,
1287
+ 512,
1288
+ 512,
1289
+ 512,
1290
+ 512,
1291
+ 512,
1292
+ 512,
1293
+ 512,
1294
+ 512,
1295
+ 512,
1296
+ 512,
1297
+ 512,
1298
+ 512,
1299
+ 512,
1300
+ 512,
1301
+ 512,
1302
+ 512,
1303
+ 512,
1304
+ 512,
1305
+ 512,
1306
+ 512,
1307
+ 512,
1308
+ 512,
1309
+ 512,
1310
+ 512,
1311
+ 512,
1312
+ 512,
1313
+ 512,
1314
+ 512
1315
+ ],
1316
+ [
1317
+ 512,
1318
+ 512,
1319
+ 512,
1320
+ 512,
1321
+ 512,
1322
+ 512,
1323
+ 512,
1324
+ 512,
1325
+ 512,
1326
+ 512,
1327
+ 512,
1328
+ 512,
1329
+ 512,
1330
+ 512,
1331
+ 512,
1332
+ 512,
1333
+ 512,
1334
+ 512,
1335
+ 512,
1336
+ 512,
1337
+ 512,
1338
+ 512,
1339
+ 512,
1340
+ 512,
1341
+ 512,
1342
+ 512,
1343
+ 512,
1344
+ 512,
1345
+ 512,
1346
+ 512,
1347
+ 512,
1348
+ 512,
1349
+ 512,
1350
+ 512,
1351
+ 512,
1352
+ 512,
1353
+ 512,
1354
+ 512,
1355
+ 512,
1356
+ 512,
1357
+ 512,
1358
+ 512,
1359
+ 512,
1360
+ 512,
1361
+ 512,
1362
+ 512,
1363
+ 512,
1364
+ 512,
1365
+ 512,
1366
+ 512,
1367
+ 512,
1368
+ 512,
1369
+ 512,
1370
+ 512,
1371
+ 512,
1372
+ 512,
1373
+ 512,
1374
+ 512,
1375
+ 512,
1376
+ 512,
1377
+ 512,
1378
+ 512,
1379
+ 512,
1380
+ 512,
1381
+ 512,
1382
+ 512,
1383
+ 512,
1384
+ 512,
1385
+ 512,
1386
+ 512,
1387
+ 512,
1388
+ 512,
1389
+ 512,
1390
+ 512,
1391
+ 512,
1392
+ 512,
1393
+ 512,
1394
+ 512,
1395
+ 512,
1396
+ 512,
1397
+ 512,
1398
+ 512,
1399
+ 512,
1400
+ 512,
1401
+ 512,
1402
+ 512,
1403
+ 512,
1404
+ 512,
1405
+ 512,
1406
+ 512,
1407
+ 512,
1408
+ 512,
1409
+ 512,
1410
+ 512,
1411
+ 512,
1412
+ 512,
1413
+ 512,
1414
+ 512,
1415
+ 512,
1416
+ 512,
1417
+ 512,
1418
+ 512,
1419
+ 512,
1420
+ 512,
1421
+ 512,
1422
+ 512,
1423
+ 512,
1424
+ 512,
1425
+ 512,
1426
+ 512,
1427
+ 512,
1428
+ 512,
1429
+ 512,
1430
+ 512,
1431
+ 512,
1432
+ 512,
1433
+ 512,
1434
+ 512,
1435
+ 512,
1436
+ 512,
1437
+ 512,
1438
+ 512,
1439
+ 512,
1440
+ 512,
1441
+ 512,
1442
+ 512,
1443
+ 512,
1444
+ 512
1445
+ ],
1446
+ [
1447
+ 512,
1448
+ 512,
1449
+ 512,
1450
+ 512,
1451
+ 512,
1452
+ 512,
1453
+ 512,
1454
+ 512,
1455
+ 512,
1456
+ 512,
1457
+ 512,
1458
+ 512,
1459
+ 512,
1460
+ 512,
1461
+ 512,
1462
+ 512,
1463
+ 512,
1464
+ 512,
1465
+ 512,
1466
+ 512,
1467
+ 512,
1468
+ 512,
1469
+ 512,
1470
+ 512,
1471
+ 512,
1472
+ 512,
1473
+ 512,
1474
+ 512,
1475
+ 512,
1476
+ 512,
1477
+ 512,
1478
+ 512,
1479
+ 512,
1480
+ 512,
1481
+ 512,
1482
+ 512,
1483
+ 512,
1484
+ 512,
1485
+ 512,
1486
+ 512,
1487
+ 512,
1488
+ 512,
1489
+ 512,
1490
+ 512,
1491
+ 512,
1492
+ 512,
1493
+ 512,
1494
+ 512,
1495
+ 512,
1496
+ 512,
1497
+ 512,
1498
+ 512,
1499
+ 512,
1500
+ 512,
1501
+ 512,
1502
+ 512,
1503
+ 512,
1504
+ 512,
1505
+ 512,
1506
+ 512,
1507
+ 512,
1508
+ 512,
1509
+ 512,
1510
+ 512,
1511
+ 512,
1512
+ 512,
1513
+ 512,
1514
+ 512,
1515
+ 512,
1516
+ 512,
1517
+ 512,
1518
+ 512,
1519
+ 512,
1520
+ 512,
1521
+ 512,
1522
+ 512,
1523
+ 512,
1524
+ 512,
1525
+ 512,
1526
+ 512,
1527
+ 512,
1528
+ 512,
1529
+ 512,
1530
+ 512,
1531
+ 512,
1532
+ 512,
1533
+ 512,
1534
+ 512,
1535
+ 512,
1536
+ 512,
1537
+ 512,
1538
+ 512,
1539
+ 512,
1540
+ 512,
1541
+ 512,
1542
+ 512,
1543
+ 512,
1544
+ 512,
1545
+ 512,
1546
+ 512,
1547
+ 512,
1548
+ 512,
1549
+ 512,
1550
+ 512,
1551
+ 512,
1552
+ 512,
1553
+ 512,
1554
+ 512,
1555
+ 512,
1556
+ 512,
1557
+ 512,
1558
+ 512,
1559
+ 512,
1560
+ 512,
1561
+ 512,
1562
+ 512,
1563
+ 512,
1564
+ 512,
1565
+ 512,
1566
+ 512,
1567
+ 512,
1568
+ 512,
1569
+ 512,
1570
+ 512,
1571
+ 512,
1572
+ 512,
1573
+ 512,
1574
+ 512
1575
+ ],
1576
+ [
1577
+ 512,
1578
+ 512,
1579
+ 512,
1580
+ 512,
1581
+ 512,
1582
+ 512,
1583
+ 512,
1584
+ 512,
1585
+ 512,
1586
+ 512,
1587
+ 512,
1588
+ 512,
1589
+ 512,
1590
+ 512,
1591
+ 512,
1592
+ 512,
1593
+ 512,
1594
+ 512,
1595
+ 512,
1596
+ 512,
1597
+ 512,
1598
+ 512,
1599
+ 512,
1600
+ 512,
1601
+ 512,
1602
+ 512,
1603
+ 512,
1604
+ 512,
1605
+ 512,
1606
+ 512,
1607
+ 512,
1608
+ 512,
1609
+ 512,
1610
+ 512,
1611
+ 512,
1612
+ 512,
1613
+ 512,
1614
+ 512,
1615
+ 512,
1616
+ 512,
1617
+ 512,
1618
+ 512,
1619
+ 512,
1620
+ 512,
1621
+ 512,
1622
+ 512,
1623
+ 512,
1624
+ 512,
1625
+ 512,
1626
+ 512,
1627
+ 512,
1628
+ 512,
1629
+ 512,
1630
+ 512,
1631
+ 512,
1632
+ 512,
1633
+ 512,
1634
+ 512,
1635
+ 512,
1636
+ 512,
1637
+ 512,
1638
+ 512,
1639
+ 512,
1640
+ 512,
1641
+ 512,
1642
+ 512,
1643
+ 512,
1644
+ 512,
1645
+ 512,
1646
+ 512,
1647
+ 512,
1648
+ 512,
1649
+ 512,
1650
+ 512,
1651
+ 512,
1652
+ 512,
1653
+ 512,
1654
+ 512,
1655
+ 512,
1656
+ 512,
1657
+ 512,
1658
+ 512,
1659
+ 512,
1660
+ 512,
1661
+ 512,
1662
+ 512,
1663
+ 512,
1664
+ 512,
1665
+ 512,
1666
+ 512,
1667
+ 512,
1668
+ 512,
1669
+ 512,
1670
+ 512,
1671
+ 512,
1672
+ 512,
1673
+ 512,
1674
+ 512,
1675
+ 512,
1676
+ 512,
1677
+ 512,
1678
+ 512,
1679
+ 512,
1680
+ 512,
1681
+ 512,
1682
+ 512,
1683
+ 512,
1684
+ 512,
1685
+ 512,
1686
+ 512,
1687
+ 512,
1688
+ 512,
1689
+ 512,
1690
+ 512,
1691
+ 512,
1692
+ 512,
1693
+ 512,
1694
+ 512,
1695
+ 512,
1696
+ 512,
1697
+ 512,
1698
+ 512,
1699
+ 512,
1700
+ 512,
1701
+ 512,
1702
+ 512,
1703
+ 512,
1704
+ 512
1705
+ ],
1706
+ [
1707
+ 512,
1708
+ 512,
1709
+ 512,
1710
+ 512,
1711
+ 512,
1712
+ 512,
1713
+ 512,
1714
+ 512,
1715
+ 512,
1716
+ 512,
1717
+ 512,
1718
+ 512,
1719
+ 512,
1720
+ 512,
1721
+ 512,
1722
+ 512,
1723
+ 512,
1724
+ 512,
1725
+ 512,
1726
+ 512,
1727
+ 512,
1728
+ 512,
1729
+ 512,
1730
+ 512,
1731
+ 512,
1732
+ 512,
1733
+ 512,
1734
+ 512,
1735
+ 512,
1736
+ 512,
1737
+ 512,
1738
+ 512,
1739
+ 512,
1740
+ 512,
1741
+ 512,
1742
+ 512,
1743
+ 512,
1744
+ 512,
1745
+ 512,
1746
+ 512,
1747
+ 512,
1748
+ 512,
1749
+ 512,
1750
+ 512,
1751
+ 512,
1752
+ 512,
1753
+ 512,
1754
+ 512,
1755
+ 512,
1756
+ 512,
1757
+ 512,
1758
+ 512,
1759
+ 512,
1760
+ 512,
1761
+ 512,
1762
+ 512,
1763
+ 512,
1764
+ 512,
1765
+ 512,
1766
+ 512,
1767
+ 512,
1768
+ 512,
1769
+ 512,
1770
+ 512,
1771
+ 512,
1772
+ 512,
1773
+ 512,
1774
+ 512,
1775
+ 512,
1776
+ 512,
1777
+ 512,
1778
+ 512,
1779
+ 512,
1780
+ 512,
1781
+ 512,
1782
+ 512,
1783
+ 512,
1784
+ 512,
1785
+ 512,
1786
+ 512,
1787
+ 512,
1788
+ 512,
1789
+ 512,
1790
+ 512,
1791
+ 512,
1792
+ 512,
1793
+ 512,
1794
+ 512,
1795
+ 512,
1796
+ 512,
1797
+ 512,
1798
+ 512,
1799
+ 512,
1800
+ 512,
1801
+ 512,
1802
+ 512,
1803
+ 512,
1804
+ 512,
1805
+ 512,
1806
+ 512,
1807
+ 512,
1808
+ 512,
1809
+ 512,
1810
+ 512,
1811
+ 512,
1812
+ 512,
1813
+ 512,
1814
+ 512,
1815
+ 512,
1816
+ 512,
1817
+ 512,
1818
+ 512,
1819
+ 512,
1820
+ 512,
1821
+ 512,
1822
+ 512,
1823
+ 512,
1824
+ 512,
1825
+ 512,
1826
+ 512,
1827
+ 512,
1828
+ 512,
1829
+ 512,
1830
+ 512,
1831
+ 512,
1832
+ 512,
1833
+ 512,
1834
+ 512
1835
+ ],
1836
+ [
1837
+ 512,
1838
+ 512,
1839
+ 512,
1840
+ 512,
1841
+ 512,
1842
+ 512,
1843
+ 512,
1844
+ 512,
1845
+ 512,
1846
+ 512,
1847
+ 512,
1848
+ 512,
1849
+ 512,
1850
+ 512,
1851
+ 512,
1852
+ 512,
1853
+ 512,
1854
+ 512,
1855
+ 512,
1856
+ 512,
1857
+ 512,
1858
+ 512,
1859
+ 512,
1860
+ 512,
1861
+ 512,
1862
+ 512,
1863
+ 512,
1864
+ 512,
1865
+ 512,
1866
+ 512,
1867
+ 512,
1868
+ 512,
1869
+ 512,
1870
+ 512,
1871
+ 512,
1872
+ 512,
1873
+ 512,
1874
+ 512,
1875
+ 512,
1876
+ 512,
1877
+ 512,
1878
+ 512,
1879
+ 512,
1880
+ 512,
1881
+ 512,
1882
+ 512,
1883
+ 512,
1884
+ 512,
1885
+ 512,
1886
+ 512,
1887
+ 512,
1888
+ 512,
1889
+ 512,
1890
+ 512,
1891
+ 512,
1892
+ 512,
1893
+ 512,
1894
+ 512,
1895
+ 512,
1896
+ 512,
1897
+ 512,
1898
+ 512,
1899
+ 512,
1900
+ 512,
1901
+ 512,
1902
+ 512,
1903
+ 512,
1904
+ 512,
1905
+ 512,
1906
+ 512,
1907
+ 512,
1908
+ 512,
1909
+ 512,
1910
+ 512,
1911
+ 512,
1912
+ 512,
1913
+ 512,
1914
+ 512,
1915
+ 512,
1916
+ 512,
1917
+ 512,
1918
+ 512,
1919
+ 512,
1920
+ 512,
1921
+ 512,
1922
+ 512,
1923
+ 512,
1924
+ 512,
1925
+ 512,
1926
+ 512,
1927
+ 512,
1928
+ 512,
1929
+ 512,
1930
+ 512,
1931
+ 512,
1932
+ 512,
1933
+ 512,
1934
+ 512,
1935
+ 512,
1936
+ 512,
1937
+ 512,
1938
+ 512,
1939
+ 512,
1940
+ 512,
1941
+ 512,
1942
+ 512,
1943
+ 512,
1944
+ 512,
1945
+ 512,
1946
+ 512,
1947
+ 512,
1948
+ 512,
1949
+ 512,
1950
+ 512,
1951
+ 512,
1952
+ 512,
1953
+ 512,
1954
+ 512,
1955
+ 512,
1956
+ 512,
1957
+ 512,
1958
+ 512,
1959
+ 512,
1960
+ 512,
1961
+ 512,
1962
+ 512,
1963
+ 512,
1964
+ 512
1965
+ ],
1966
+ [
1967
+ 512,
1968
+ 512,
1969
+ 512,
1970
+ 512,
1971
+ 512,
1972
+ 512,
1973
+ 512,
1974
+ 512,
1975
+ 512,
1976
+ 512,
1977
+ 512,
1978
+ 512,
1979
+ 512,
1980
+ 512,
1981
+ 512,
1982
+ 512,
1983
+ 512,
1984
+ 512,
1985
+ 512,
1986
+ 512,
1987
+ 512,
1988
+ 512,
1989
+ 512,
1990
+ 512,
1991
+ 512,
1992
+ 512,
1993
+ 512,
1994
+ 512,
1995
+ 512,
1996
+ 512,
1997
+ 512,
1998
+ 512,
1999
+ 512,
2000
+ 512,
2001
+ 512,
2002
+ 512,
2003
+ 512,
2004
+ 512,
2005
+ 512,
2006
+ 512,
2007
+ 512,
2008
+ 512,
2009
+ 512,
2010
+ 512,
2011
+ 512,
2012
+ 512,
2013
+ 512,
2014
+ 512,
2015
+ 512,
2016
+ 512,
2017
+ 512,
2018
+ 512,
2019
+ 512,
2020
+ 512,
2021
+ 512,
2022
+ 512,
2023
+ 512,
2024
+ 512,
2025
+ 512,
2026
+ 512,
2027
+ 512,
2028
+ 512,
2029
+ 512,
2030
+ 512,
2031
+ 512,
2032
+ 512,
2033
+ 512,
2034
+ 512,
2035
+ 512,
2036
+ 512,
2037
+ 512,
2038
+ 512,
2039
+ 512,
2040
+ 512,
2041
+ 512,
2042
+ 512,
2043
+ 512,
2044
+ 512,
2045
+ 512,
2046
+ 512,
2047
+ 512,
2048
+ 512,
2049
+ 512,
2050
+ 512,
2051
+ 512,
2052
+ 512,
2053
+ 512,
2054
+ 512,
2055
+ 512,
2056
+ 512,
2057
+ 512,
2058
+ 512,
2059
+ 512,
2060
+ 512,
2061
+ 512,
2062
+ 512,
2063
+ 512,
2064
+ 512,
2065
+ 512,
2066
+ 512,
2067
+ 512,
2068
+ 512,
2069
+ 512,
2070
+ 512,
2071
+ 512,
2072
+ 512,
2073
+ 512,
2074
+ 512,
2075
+ 512,
2076
+ 512,
2077
+ 512,
2078
+ 512,
2079
+ 512,
2080
+ 512,
2081
+ 512,
2082
+ 512,
2083
+ 512,
2084
+ 512,
2085
+ 512,
2086
+ 512,
2087
+ 512,
2088
+ 512,
2089
+ 512,
2090
+ 512,
2091
+ 512,
2092
+ 512,
2093
+ 512,
2094
+ 512
2095
+ ],
2096
+ [
2097
+ 512,
2098
+ 512,
2099
+ 512,
2100
+ 512,
2101
+ 512,
2102
+ 512,
2103
+ 512,
2104
+ 512,
2105
+ 512,
2106
+ 512,
2107
+ 512,
2108
+ 512,
2109
+ 512,
2110
+ 512,
2111
+ 512,
2112
+ 512,
2113
+ 512,
2114
+ 512,
2115
+ 512,
2116
+ 512,
2117
+ 512,
2118
+ 512,
2119
+ 512,
2120
+ 512,
2121
+ 512,
2122
+ 512,
2123
+ 512,
2124
+ 512,
2125
+ 512,
2126
+ 512,
2127
+ 512,
2128
+ 512,
2129
+ 512,
2130
+ 512,
2131
+ 512,
2132
+ 512,
2133
+ 512,
2134
+ 512,
2135
+ 512,
2136
+ 512,
2137
+ 512,
2138
+ 512,
2139
+ 512,
2140
+ 512,
2141
+ 512,
2142
+ 512,
2143
+ 512,
2144
+ 512,
2145
+ 512,
2146
+ 512,
2147
+ 512,
2148
+ 512,
2149
+ 512,
2150
+ 512,
2151
+ 512,
2152
+ 512,
2153
+ 512,
2154
+ 512,
2155
+ 512,
2156
+ 512,
2157
+ 512,
2158
+ 512,
2159
+ 512,
2160
+ 512,
2161
+ 512,
2162
+ 512,
2163
+ 512,
2164
+ 512,
2165
+ 512,
2166
+ 512,
2167
+ 512,
2168
+ 512,
2169
+ 512,
2170
+ 512,
2171
+ 512,
2172
+ 512,
2173
+ 512,
2174
+ 512,
2175
+ 512,
2176
+ 512,
2177
+ 512,
2178
+ 512,
2179
+ 512,
2180
+ 512,
2181
+ 512,
2182
+ 512,
2183
+ 512,
2184
+ 512,
2185
+ 512,
2186
+ 512,
2187
+ 512,
2188
+ 512,
2189
+ 512,
2190
+ 512,
2191
+ 512,
2192
+ 512,
2193
+ 512,
2194
+ 512,
2195
+ 512,
2196
+ 512,
2197
+ 512,
2198
+ 512,
2199
+ 512,
2200
+ 512,
2201
+ 512,
2202
+ 512,
2203
+ 512,
2204
+ 512,
2205
+ 512,
2206
+ 512,
2207
+ 512,
2208
+ 512,
2209
+ 512,
2210
+ 512,
2211
+ 512,
2212
+ 512,
2213
+ 512,
2214
+ 512,
2215
+ 512,
2216
+ 512,
2217
+ 512,
2218
+ 512,
2219
+ 512,
2220
+ 512,
2221
+ 512,
2222
+ 512,
2223
+ 512,
2224
+ 512
2225
+ ],
2226
+ [
2227
+ 512,
2228
+ 512,
2229
+ 512,
2230
+ 512,
2231
+ 512,
2232
+ 512,
2233
+ 512,
2234
+ 512,
2235
+ 512,
2236
+ 512,
2237
+ 512,
2238
+ 512,
2239
+ 512,
2240
+ 512,
2241
+ 512,
2242
+ 512,
2243
+ 512,
2244
+ 512,
2245
+ 512,
2246
+ 512,
2247
+ 512,
2248
+ 512,
2249
+ 512,
2250
+ 512,
2251
+ 512,
2252
+ 512,
2253
+ 512,
2254
+ 512,
2255
+ 512,
2256
+ 512,
2257
+ 512,
2258
+ 512,
2259
+ 512,
2260
+ 512,
2261
+ 512,
2262
+ 512,
2263
+ 512,
2264
+ 512,
2265
+ 512,
2266
+ 512,
2267
+ 512,
2268
+ 512,
2269
+ 512,
2270
+ 512,
2271
+ 512,
2272
+ 512,
2273
+ 512,
2274
+ 512,
2275
+ 512,
2276
+ 512,
2277
+ 512,
2278
+ 512,
2279
+ 512,
2280
+ 512,
2281
+ 512,
2282
+ 512,
2283
+ 512,
2284
+ 512,
2285
+ 512,
2286
+ 512,
2287
+ 512,
2288
+ 512,
2289
+ 512,
2290
+ 512,
2291
+ 512,
2292
+ 512,
2293
+ 512,
2294
+ 512,
2295
+ 512,
2296
+ 512,
2297
+ 512,
2298
+ 512,
2299
+ 512,
2300
+ 512,
2301
+ 512,
2302
+ 512,
2303
+ 512,
2304
+ 512,
2305
+ 512,
2306
+ 512,
2307
+ 512,
2308
+ 512,
2309
+ 512,
2310
+ 512,
2311
+ 512,
2312
+ 512,
2313
+ 512,
2314
+ 512,
2315
+ 512,
2316
+ 512,
2317
+ 512,
2318
+ 512,
2319
+ 512,
2320
+ 512,
2321
+ 512,
2322
+ 512,
2323
+ 512,
2324
+ 512,
2325
+ 512,
2326
+ 512,
2327
+ 512,
2328
+ 512,
2329
+ 512,
2330
+ 512,
2331
+ 512,
2332
+ 512,
2333
+ 512,
2334
+ 512,
2335
+ 512,
2336
+ 512,
2337
+ 512,
2338
+ 512,
2339
+ 512,
2340
+ 512,
2341
+ 512,
2342
+ 512,
2343
+ 512,
2344
+ 512,
2345
+ 512,
2346
+ 512,
2347
+ 512,
2348
+ 512,
2349
+ 512,
2350
+ 512,
2351
+ 512,
2352
+ 512,
2353
+ 512,
2354
+ 512
2355
+ ],
2356
+ [
2357
+ 512,
2358
+ 512,
2359
+ 512,
2360
+ 512,
2361
+ 512,
2362
+ 512,
2363
+ 512,
2364
+ 512,
2365
+ 512,
2366
+ 512,
2367
+ 512,
2368
+ 512,
2369
+ 512,
2370
+ 512,
2371
+ 512,
2372
+ 512,
2373
+ 512,
2374
+ 512,
2375
+ 512,
2376
+ 512,
2377
+ 512,
2378
+ 512,
2379
+ 512,
2380
+ 512,
2381
+ 512,
2382
+ 512,
2383
+ 512,
2384
+ 512,
2385
+ 512,
2386
+ 512,
2387
+ 512,
2388
+ 512,
2389
+ 512,
2390
+ 512,
2391
+ 512,
2392
+ 512,
2393
+ 512,
2394
+ 512,
2395
+ 512,
2396
+ 512,
2397
+ 512,
2398
+ 512,
2399
+ 512,
2400
+ 512,
2401
+ 512,
2402
+ 512,
2403
+ 512,
2404
+ 512,
2405
+ 512,
2406
+ 512,
2407
+ 512,
2408
+ 512,
2409
+ 512,
2410
+ 512,
2411
+ 512,
2412
+ 512,
2413
+ 512,
2414
+ 512,
2415
+ 512,
2416
+ 512,
2417
+ 512,
2418
+ 512,
2419
+ 512,
2420
+ 512,
2421
+ 512,
2422
+ 512,
2423
+ 512,
2424
+ 512,
2425
+ 512,
2426
+ 512,
2427
+ 512,
2428
+ 512,
2429
+ 512,
2430
+ 512,
2431
+ 512,
2432
+ 512,
2433
+ 512,
2434
+ 512,
2435
+ 512,
2436
+ 512,
2437
+ 512,
2438
+ 512,
2439
+ 512,
2440
+ 512,
2441
+ 512,
2442
+ 512,
2443
+ 512,
2444
+ 512,
2445
+ 512,
2446
+ 512,
2447
+ 512,
2448
+ 512,
2449
+ 512,
2450
+ 512,
2451
+ 512,
2452
+ 512,
2453
+ 512,
2454
+ 512,
2455
+ 512,
2456
+ 512,
2457
+ 512,
2458
+ 512,
2459
+ 512,
2460
+ 512,
2461
+ 512,
2462
+ 512,
2463
+ 512,
2464
+ 512,
2465
+ 512,
2466
+ 512,
2467
+ 512,
2468
+ 512,
2469
+ 512,
2470
+ 512,
2471
+ 512,
2472
+ 512,
2473
+ 512,
2474
+ 512,
2475
+ 512,
2476
+ 512,
2477
+ 512,
2478
+ 512,
2479
+ 512,
2480
+ 512,
2481
+ 512,
2482
+ 512,
2483
+ 512,
2484
+ 512
2485
+ ],
2486
+ [
2487
+ 512,
2488
+ 512,
2489
+ 512,
2490
+ 512,
2491
+ 512,
2492
+ 512,
2493
+ 512,
2494
+ 512,
2495
+ 512,
2496
+ 512,
2497
+ 512,
2498
+ 512,
2499
+ 512,
2500
+ 512,
2501
+ 512,
2502
+ 512,
2503
+ 512,
2504
+ 512,
2505
+ 512,
2506
+ 512,
2507
+ 512,
2508
+ 512,
2509
+ 512,
2510
+ 512,
2511
+ 512,
2512
+ 512,
2513
+ 512,
2514
+ 512,
2515
+ 512,
2516
+ 512,
2517
+ 512,
2518
+ 512,
2519
+ 512,
2520
+ 512,
2521
+ 512,
2522
+ 512,
2523
+ 512,
2524
+ 512,
2525
+ 512,
2526
+ 512,
2527
+ 512,
2528
+ 512,
2529
+ 512,
2530
+ 512,
2531
+ 512,
2532
+ 512,
2533
+ 512,
2534
+ 512,
2535
+ 512,
2536
+ 512,
2537
+ 512,
2538
+ 512,
2539
+ 512,
2540
+ 512,
2541
+ 512,
2542
+ 512,
2543
+ 512,
2544
+ 512,
2545
+ 512,
2546
+ 512,
2547
+ 512,
2548
+ 512,
2549
+ 512,
2550
+ 512,
2551
+ 512,
2552
+ 512,
2553
+ 512,
2554
+ 512,
2555
+ 512,
2556
+ 512,
2557
+ 512,
2558
+ 512,
2559
+ 512,
2560
+ 512,
2561
+ 512,
2562
+ 512,
2563
+ 512,
2564
+ 512,
2565
+ 512,
2566
+ 512,
2567
+ 512,
2568
+ 512,
2569
+ 512,
2570
+ 512,
2571
+ 512,
2572
+ 512,
2573
+ 512,
2574
+ 512,
2575
+ 512,
2576
+ 512,
2577
+ 512,
2578
+ 512,
2579
+ 512,
2580
+ 512,
2581
+ 512,
2582
+ 512,
2583
+ 512,
2584
+ 512,
2585
+ 512,
2586
+ 512,
2587
+ 512,
2588
+ 512,
2589
+ 512,
2590
+ 512,
2591
+ 512,
2592
+ 512,
2593
+ 512,
2594
+ 512,
2595
+ 512,
2596
+ 512,
2597
+ 512,
2598
+ 512,
2599
+ 512,
2600
+ 512,
2601
+ 512,
2602
+ 512,
2603
+ 512,
2604
+ 512,
2605
+ 512,
2606
+ 512,
2607
+ 512,
2608
+ 512,
2609
+ 512,
2610
+ 512,
2611
+ 512,
2612
+ 512,
2613
+ 512,
2614
+ 512
2615
+ ],
2616
+ [
2617
+ 512,
2618
+ 512,
2619
+ 512,
2620
+ 512,
2621
+ 512,
2622
+ 512,
2623
+ 512,
2624
+ 512,
2625
+ 512,
2626
+ 512,
2627
+ 512,
2628
+ 512,
2629
+ 512,
2630
+ 512,
2631
+ 512,
2632
+ 512,
2633
+ 512,
2634
+ 512,
2635
+ 512,
2636
+ 512,
2637
+ 512,
2638
+ 512,
2639
+ 512,
2640
+ 512,
2641
+ 512,
2642
+ 512,
2643
+ 512,
2644
+ 512,
2645
+ 512,
2646
+ 512,
2647
+ 512,
2648
+ 512,
2649
+ 512,
2650
+ 512,
2651
+ 512,
2652
+ 512,
2653
+ 512,
2654
+ 512,
2655
+ 512,
2656
+ 512,
2657
+ 512,
2658
+ 512,
2659
+ 512,
2660
+ 512,
2661
+ 512,
2662
+ 512,
2663
+ 512,
2664
+ 512,
2665
+ 512,
2666
+ 512,
2667
+ 512,
2668
+ 512,
2669
+ 512,
2670
+ 512,
2671
+ 512,
2672
+ 512,
2673
+ 512,
2674
+ 512,
2675
+ 512,
2676
+ 512,
2677
+ 512,
2678
+ 512,
2679
+ 512,
2680
+ 512,
2681
+ 512,
2682
+ 512,
2683
+ 512,
2684
+ 512,
2685
+ 512,
2686
+ 512,
2687
+ 512,
2688
+ 512,
2689
+ 512,
2690
+ 512,
2691
+ 512,
2692
+ 512,
2693
+ 512,
2694
+ 512,
2695
+ 512,
2696
+ 512,
2697
+ 512,
2698
+ 512,
2699
+ 512,
2700
+ 512,
2701
+ 512,
2702
+ 512,
2703
+ 512,
2704
+ 512,
2705
+ 512,
2706
+ 512,
2707
+ 512,
2708
+ 512,
2709
+ 512,
2710
+ 512,
2711
+ 512,
2712
+ 512,
2713
+ 512,
2714
+ 512,
2715
+ 512,
2716
+ 512,
2717
+ 512,
2718
+ 512,
2719
+ 512,
2720
+ 512,
2721
+ 512,
2722
+ 512,
2723
+ 512,
2724
+ 512,
2725
+ 512,
2726
+ 512,
2727
+ 512,
2728
+ 512,
2729
+ 512,
2730
+ 512,
2731
+ 512,
2732
+ 512,
2733
+ 512,
2734
+ 512,
2735
+ 512,
2736
+ 512,
2737
+ 512,
2738
+ 512,
2739
+ 512,
2740
+ 512,
2741
+ 512,
2742
+ 512,
2743
+ 512,
2744
+ 512
2745
+ ],
2746
+ [
2747
+ 512,
2748
+ 512,
2749
+ 512,
2750
+ 512,
2751
+ 512,
2752
+ 512,
2753
+ 512,
2754
+ 512,
2755
+ 512,
2756
+ 512,
2757
+ 512,
2758
+ 512,
2759
+ 512,
2760
+ 512,
2761
+ 512,
2762
+ 512,
2763
+ 512,
2764
+ 512,
2765
+ 512,
2766
+ 512,
2767
+ 512,
2768
+ 512,
2769
+ 512,
2770
+ 512,
2771
+ 512,
2772
+ 512,
2773
+ 512,
2774
+ 512,
2775
+ 512,
2776
+ 512,
2777
+ 512,
2778
+ 512,
2779
+ 512,
2780
+ 512,
2781
+ 512,
2782
+ 512,
2783
+ 512,
2784
+ 512,
2785
+ 512,
2786
+ 512,
2787
+ 512,
2788
+ 512,
2789
+ 512,
2790
+ 512,
2791
+ 512,
2792
+ 512,
2793
+ 512,
2794
+ 512,
2795
+ 512,
2796
+ 512,
2797
+ 512,
2798
+ 512,
2799
+ 512,
2800
+ 512,
2801
+ 512,
2802
+ 512,
2803
+ 512,
2804
+ 512,
2805
+ 512,
2806
+ 512,
2807
+ 512,
2808
+ 512,
2809
+ 512,
2810
+ 512,
2811
+ 512,
2812
+ 512,
2813
+ 512,
2814
+ 512,
2815
+ 512,
2816
+ 512,
2817
+ 512,
2818
+ 512,
2819
+ 512,
2820
+ 512,
2821
+ 512,
2822
+ 512,
2823
+ 512,
2824
+ 512,
2825
+ 512,
2826
+ 512,
2827
+ 512,
2828
+ 512,
2829
+ 512,
2830
+ 512,
2831
+ 512,
2832
+ 512,
2833
+ 512,
2834
+ 512,
2835
+ 512,
2836
+ 512,
2837
+ 512,
2838
+ 512,
2839
+ 512,
2840
+ 512,
2841
+ 512,
2842
+ 512,
2843
+ 512,
2844
+ 512,
2845
+ 512,
2846
+ 512,
2847
+ 512,
2848
+ 512,
2849
+ 512,
2850
+ 512,
2851
+ 512,
2852
+ 512,
2853
+ 512,
2854
+ 512,
2855
+ 512,
2856
+ 512,
2857
+ 512,
2858
+ 512,
2859
+ 512,
2860
+ 512,
2861
+ 512,
2862
+ 512,
2863
+ 512,
2864
+ 512,
2865
+ 512,
2866
+ 512,
2867
+ 512,
2868
+ 512,
2869
+ 512,
2870
+ 512,
2871
+ 512,
2872
+ 512,
2873
+ 512,
2874
+ 512
2875
+ ],
2876
+ [
2877
+ 512,
2878
+ 512,
2879
+ 512,
2880
+ 512,
2881
+ 512,
2882
+ 512,
2883
+ 512,
2884
+ 512,
2885
+ 512,
2886
+ 512,
2887
+ 512,
2888
+ 512,
2889
+ 512,
2890
+ 512,
2891
+ 512,
2892
+ 512,
2893
+ 512,
2894
+ 512,
2895
+ 512,
2896
+ 512,
2897
+ 512,
2898
+ 512,
2899
+ 512,
2900
+ 512,
2901
+ 512,
2902
+ 512,
2903
+ 512,
2904
+ 512,
2905
+ 512,
2906
+ 512,
2907
+ 512,
2908
+ 512,
2909
+ 512,
2910
+ 512,
2911
+ 512,
2912
+ 512,
2913
+ 512,
2914
+ 512,
2915
+ 512,
2916
+ 512,
2917
+ 512,
2918
+ 512,
2919
+ 512,
2920
+ 512,
2921
+ 512,
2922
+ 512,
2923
+ 512,
2924
+ 512,
2925
+ 512,
2926
+ 512,
2927
+ 512,
2928
+ 512,
2929
+ 512,
2930
+ 512,
2931
+ 512,
2932
+ 512,
2933
+ 512,
2934
+ 512,
2935
+ 512,
2936
+ 512,
2937
+ 512,
2938
+ 512,
2939
+ 512,
2940
+ 512,
2941
+ 512,
2942
+ 512,
2943
+ 512,
2944
+ 512,
2945
+ 512,
2946
+ 512,
2947
+ 512,
2948
+ 512,
2949
+ 512,
2950
+ 512,
2951
+ 512,
2952
+ 512,
2953
+ 512,
2954
+ 512,
2955
+ 512,
2956
+ 512,
2957
+ 512,
2958
+ 512,
2959
+ 512,
2960
+ 512,
2961
+ 512,
2962
+ 512,
2963
+ 512,
2964
+ 512,
2965
+ 512,
2966
+ 512,
2967
+ 512,
2968
+ 512,
2969
+ 512,
2970
+ 512,
2971
+ 512,
2972
+ 512,
2973
+ 512,
2974
+ 512,
2975
+ 512,
2976
+ 512,
2977
+ 512,
2978
+ 512,
2979
+ 512,
2980
+ 512,
2981
+ 512,
2982
+ 512,
2983
+ 512,
2984
+ 512,
2985
+ 512,
2986
+ 512,
2987
+ 512,
2988
+ 512,
2989
+ 512,
2990
+ 512,
2991
+ 512,
2992
+ 512,
2993
+ 512,
2994
+ 512,
2995
+ 512,
2996
+ 512,
2997
+ 512,
2998
+ 512,
2999
+ 512,
3000
+ 512,
3001
+ 512,
3002
+ 512,
3003
+ 512,
3004
+ 512
3005
+ ],
3006
+ [
3007
+ 512,
3008
+ 512,
3009
+ 512,
3010
+ 512,
3011
+ 512,
3012
+ 512,
3013
+ 512,
3014
+ 512,
3015
+ 512,
3016
+ 512,
3017
+ 512,
3018
+ 512,
3019
+ 512,
3020
+ 512,
3021
+ 512,
3022
+ 512,
3023
+ 512,
3024
+ 512,
3025
+ 512,
3026
+ 512,
3027
+ 512,
3028
+ 512,
3029
+ 512,
3030
+ 512,
3031
+ 512,
3032
+ 512,
3033
+ 512,
3034
+ 512,
3035
+ 512,
3036
+ 512,
3037
+ 512,
3038
+ 512,
3039
+ 512,
3040
+ 512,
3041
+ 512,
3042
+ 512,
3043
+ 512,
3044
+ 512,
3045
+ 512,
3046
+ 512,
3047
+ 512,
3048
+ 512,
3049
+ 512,
3050
+ 512,
3051
+ 512,
3052
+ 512,
3053
+ 512,
3054
+ 512,
3055
+ 512,
3056
+ 512,
3057
+ 512,
3058
+ 512,
3059
+ 512,
3060
+ 512,
3061
+ 512,
3062
+ 512,
3063
+ 512,
3064
+ 512,
3065
+ 512,
3066
+ 512,
3067
+ 512,
3068
+ 512,
3069
+ 512,
3070
+ 512,
3071
+ 512,
3072
+ 512,
3073
+ 512,
3074
+ 512,
3075
+ 512,
3076
+ 512,
3077
+ 512,
3078
+ 512,
3079
+ 512,
3080
+ 512,
3081
+ 512,
3082
+ 512,
3083
+ 512,
3084
+ 512,
3085
+ 512,
3086
+ 512,
3087
+ 512,
3088
+ 512,
3089
+ 512,
3090
+ 512,
3091
+ 512,
3092
+ 512,
3093
+ 512,
3094
+ 512,
3095
+ 512,
3096
+ 512,
3097
+ 512,
3098
+ 512,
3099
+ 512,
3100
+ 512,
3101
+ 512,
3102
+ 512,
3103
+ 512,
3104
+ 512,
3105
+ 512,
3106
+ 512,
3107
+ 512,
3108
+ 512,
3109
+ 512,
3110
+ 512,
3111
+ 512,
3112
+ 512,
3113
+ 512,
3114
+ 512,
3115
+ 512,
3116
+ 512,
3117
+ 512,
3118
+ 512,
3119
+ 512,
3120
+ 512,
3121
+ 512,
3122
+ 512,
3123
+ 512,
3124
+ 512,
3125
+ 512,
3126
+ 512,
3127
+ 512,
3128
+ 512,
3129
+ 512,
3130
+ 512,
3131
+ 512,
3132
+ 512,
3133
+ 512,
3134
+ 512
3135
+ ],
3136
+ [
3137
+ 512,
3138
+ 512,
3139
+ 512,
3140
+ 512,
3141
+ 512,
3142
+ 512,
3143
+ 512,
3144
+ 512,
3145
+ 512,
3146
+ 512,
3147
+ 512,
3148
+ 512,
3149
+ 512,
3150
+ 512,
3151
+ 512,
3152
+ 512,
3153
+ 512,
3154
+ 512,
3155
+ 512,
3156
+ 512,
3157
+ 512,
3158
+ 512,
3159
+ 512,
3160
+ 512,
3161
+ 512,
3162
+ 512,
3163
+ 512,
3164
+ 512,
3165
+ 512,
3166
+ 512,
3167
+ 512,
3168
+ 512,
3169
+ 512,
3170
+ 512,
3171
+ 512,
3172
+ 512,
3173
+ 512,
3174
+ 512,
3175
+ 512,
3176
+ 512,
3177
+ 512,
3178
+ 512,
3179
+ 512,
3180
+ 512,
3181
+ 512,
3182
+ 512,
3183
+ 512,
3184
+ 512,
3185
+ 512,
3186
+ 512,
3187
+ 512,
3188
+ 512,
3189
+ 512,
3190
+ 512,
3191
+ 512,
3192
+ 512,
3193
+ 512,
3194
+ 512,
3195
+ 512,
3196
+ 512,
3197
+ 512,
3198
+ 512,
3199
+ 512,
3200
+ 512,
3201
+ 512,
3202
+ 512,
3203
+ 512,
3204
+ 512,
3205
+ 512,
3206
+ 512,
3207
+ 512,
3208
+ 512,
3209
+ 512,
3210
+ 512,
3211
+ 512,
3212
+ 512,
3213
+ 512,
3214
+ 512,
3215
+ 512,
3216
+ 512,
3217
+ 512,
3218
+ 512,
3219
+ 512,
3220
+ 512,
3221
+ 512,
3222
+ 512,
3223
+ 512,
3224
+ 512,
3225
+ 512,
3226
+ 512,
3227
+ 512,
3228
+ 512,
3229
+ 512,
3230
+ 512,
3231
+ 512,
3232
+ 512,
3233
+ 512,
3234
+ 512,
3235
+ 512,
3236
+ 512,
3237
+ 512,
3238
+ 512,
3239
+ 512,
3240
+ 512,
3241
+ 512,
3242
+ 512,
3243
+ 512,
3244
+ 512,
3245
+ 512,
3246
+ 512,
3247
+ 512,
3248
+ 512,
3249
+ 512,
3250
+ 512,
3251
+ 512,
3252
+ 512,
3253
+ 512,
3254
+ 512,
3255
+ 512,
3256
+ 512,
3257
+ 512,
3258
+ 512,
3259
+ 512,
3260
+ 512,
3261
+ 512,
3262
+ 512,
3263
+ 512,
3264
+ 512
3265
+ ],
3266
+ [
3267
+ 512,
3268
+ 512,
3269
+ 512,
3270
+ 512,
3271
+ 512,
3272
+ 512,
3273
+ 512,
3274
+ 512,
3275
+ 512,
3276
+ 512,
3277
+ 512,
3278
+ 512,
3279
+ 512,
3280
+ 512,
3281
+ 512,
3282
+ 512,
3283
+ 512,
3284
+ 512,
3285
+ 512,
3286
+ 512,
3287
+ 512,
3288
+ 512,
3289
+ 512,
3290
+ 512,
3291
+ 512,
3292
+ 512,
3293
+ 512,
3294
+ 512,
3295
+ 512,
3296
+ 512,
3297
+ 512,
3298
+ 512,
3299
+ 512,
3300
+ 512,
3301
+ 512,
3302
+ 512,
3303
+ 512,
3304
+ 512,
3305
+ 512,
3306
+ 512,
3307
+ 512,
3308
+ 512,
3309
+ 512,
3310
+ 512,
3311
+ 512,
3312
+ 512,
3313
+ 512,
3314
+ 512,
3315
+ 512,
3316
+ 512,
3317
+ 512,
3318
+ 512,
3319
+ 512,
3320
+ 512,
3321
+ 512,
3322
+ 512,
3323
+ 512,
3324
+ 512,
3325
+ 512,
3326
+ 512,
3327
+ 512,
3328
+ 512,
3329
+ 512,
3330
+ 512,
3331
+ 512,
3332
+ 512,
3333
+ 512,
3334
+ 512,
3335
+ 512,
3336
+ 512,
3337
+ 512,
3338
+ 512,
3339
+ 512,
3340
+ 512,
3341
+ 512,
3342
+ 512,
3343
+ 512,
3344
+ 512,
3345
+ 512,
3346
+ 512,
3347
+ 512,
3348
+ 512,
3349
+ 512,
3350
+ 512,
3351
+ 512,
3352
+ 512,
3353
+ 512,
3354
+ 512,
3355
+ 512,
3356
+ 512,
3357
+ 512,
3358
+ 512,
3359
+ 512,
3360
+ 512,
3361
+ 512,
3362
+ 512,
3363
+ 512,
3364
+ 512,
3365
+ 512,
3366
+ 512,
3367
+ 512,
3368
+ 512,
3369
+ 512,
3370
+ 512,
3371
+ 512,
3372
+ 512,
3373
+ 512,
3374
+ 512,
3375
+ 512,
3376
+ 512,
3377
+ 512,
3378
+ 512,
3379
+ 512,
3380
+ 512,
3381
+ 512,
3382
+ 512,
3383
+ 512,
3384
+ 512,
3385
+ 512,
3386
+ 512,
3387
+ 512,
3388
+ 512,
3389
+ 512,
3390
+ 512,
3391
+ 512,
3392
+ 512,
3393
+ 512,
3394
+ 512
3395
+ ],
3396
+ [
3397
+ 512,
3398
+ 512,
3399
+ 512,
3400
+ 512,
3401
+ 512,
3402
+ 512,
3403
+ 512,
3404
+ 512,
3405
+ 512,
3406
+ 512,
3407
+ 512,
3408
+ 512,
3409
+ 512,
3410
+ 512,
3411
+ 512,
3412
+ 512,
3413
+ 512,
3414
+ 512,
3415
+ 512,
3416
+ 512,
3417
+ 512,
3418
+ 512,
3419
+ 512,
3420
+ 512,
3421
+ 512,
3422
+ 512,
3423
+ 512,
3424
+ 512,
3425
+ 512,
3426
+ 512,
3427
+ 512,
3428
+ 512,
3429
+ 512,
3430
+ 512,
3431
+ 512,
3432
+ 512,
3433
+ 512,
3434
+ 512,
3435
+ 512,
3436
+ 512,
3437
+ 512,
3438
+ 512,
3439
+ 512,
3440
+ 512,
3441
+ 512,
3442
+ 512,
3443
+ 512,
3444
+ 512,
3445
+ 512,
3446
+ 512,
3447
+ 512,
3448
+ 512,
3449
+ 512,
3450
+ 512,
3451
+ 512,
3452
+ 512,
3453
+ 512,
3454
+ 512,
3455
+ 512,
3456
+ 512,
3457
+ 512,
3458
+ 512,
3459
+ 512,
3460
+ 512,
3461
+ 512,
3462
+ 512,
3463
+ 512,
3464
+ 512,
3465
+ 512,
3466
+ 512,
3467
+ 512,
3468
+ 512,
3469
+ 512,
3470
+ 512,
3471
+ 512,
3472
+ 512,
3473
+ 512,
3474
+ 512,
3475
+ 512,
3476
+ 512,
3477
+ 512,
3478
+ 512,
3479
+ 512,
3480
+ 512,
3481
+ 512,
3482
+ 512,
3483
+ 512,
3484
+ 512,
3485
+ 512,
3486
+ 512,
3487
+ 512,
3488
+ 512,
3489
+ 512,
3490
+ 512,
3491
+ 512,
3492
+ 512,
3493
+ 512,
3494
+ 512,
3495
+ 512,
3496
+ 512,
3497
+ 512,
3498
+ 512,
3499
+ 512,
3500
+ 512,
3501
+ 512,
3502
+ 512,
3503
+ 512,
3504
+ 512,
3505
+ 512,
3506
+ 512,
3507
+ 512,
3508
+ 512,
3509
+ 512,
3510
+ 512,
3511
+ 512,
3512
+ 512,
3513
+ 512,
3514
+ 512,
3515
+ 512,
3516
+ 512,
3517
+ 512,
3518
+ 512,
3519
+ 512,
3520
+ 512,
3521
+ 512,
3522
+ 512,
3523
+ 512,
3524
+ 512
3525
+ ],
3526
+ [
3527
+ 512,
3528
+ 512,
3529
+ 512,
3530
+ 512,
3531
+ 512,
3532
+ 512,
3533
+ 512,
3534
+ 512,
3535
+ 512,
3536
+ 512,
3537
+ 512,
3538
+ 512,
3539
+ 512,
3540
+ 512,
3541
+ 512,
3542
+ 512,
3543
+ 512,
3544
+ 512,
3545
+ 512,
3546
+ 512,
3547
+ 512,
3548
+ 512,
3549
+ 512,
3550
+ 512,
3551
+ 512,
3552
+ 512,
3553
+ 512,
3554
+ 512,
3555
+ 512,
3556
+ 512,
3557
+ 512,
3558
+ 512,
3559
+ 512,
3560
+ 512,
3561
+ 512,
3562
+ 512,
3563
+ 512,
3564
+ 512,
3565
+ 512,
3566
+ 512,
3567
+ 512,
3568
+ 512,
3569
+ 512,
3570
+ 512,
3571
+ 512,
3572
+ 512,
3573
+ 512,
3574
+ 512,
3575
+ 512,
3576
+ 512,
3577
+ 512,
3578
+ 512,
3579
+ 512,
3580
+ 512,
3581
+ 512,
3582
+ 512,
3583
+ 512,
3584
+ 512,
3585
+ 512,
3586
+ 512,
3587
+ 512,
3588
+ 512,
3589
+ 512,
3590
+ 512,
3591
+ 512,
3592
+ 512,
3593
+ 512,
3594
+ 512,
3595
+ 512,
3596
+ 512,
3597
+ 512,
3598
+ 512,
3599
+ 512,
3600
+ 512,
3601
+ 512,
3602
+ 512,
3603
+ 512,
3604
+ 512,
3605
+ 512,
3606
+ 512,
3607
+ 512,
3608
+ 512,
3609
+ 512,
3610
+ 512,
3611
+ 512,
3612
+ 512,
3613
+ 512,
3614
+ 512,
3615
+ 512,
3616
+ 512,
3617
+ 512,
3618
+ 512,
3619
+ 512,
3620
+ 512,
3621
+ 512,
3622
+ 512,
3623
+ 512,
3624
+ 512,
3625
+ 512,
3626
+ 512,
3627
+ 512,
3628
+ 512,
3629
+ 512,
3630
+ 512,
3631
+ 512,
3632
+ 512,
3633
+ 512,
3634
+ 512,
3635
+ 512,
3636
+ 512,
3637
+ 512,
3638
+ 512,
3639
+ 512,
3640
+ 512,
3641
+ 512,
3642
+ 512,
3643
+ 512,
3644
+ 512,
3645
+ 512,
3646
+ 512,
3647
+ 512,
3648
+ 512,
3649
+ 512,
3650
+ 512,
3651
+ 512,
3652
+ 512,
3653
+ 512,
3654
+ 512
3655
+ ],
3656
+ [
3657
+ 512,
3658
+ 512,
3659
+ 512,
3660
+ 512,
3661
+ 512,
3662
+ 512,
3663
+ 512,
3664
+ 512,
3665
+ 512,
3666
+ 512,
3667
+ 512,
3668
+ 512,
3669
+ 512,
3670
+ 512,
3671
+ 512,
3672
+ 512,
3673
+ 512,
3674
+ 512,
3675
+ 512,
3676
+ 512,
3677
+ 512,
3678
+ 512,
3679
+ 512,
3680
+ 512,
3681
+ 512,
3682
+ 512,
3683
+ 512,
3684
+ 512,
3685
+ 512,
3686
+ 512,
3687
+ 512,
3688
+ 512,
3689
+ 512,
3690
+ 512,
3691
+ 512,
3692
+ 512,
3693
+ 512,
3694
+ 512,
3695
+ 512,
3696
+ 512,
3697
+ 512,
3698
+ 512,
3699
+ 512,
3700
+ 512,
3701
+ 512,
3702
+ 512,
3703
+ 512,
3704
+ 512,
3705
+ 512,
3706
+ 512,
3707
+ 512,
3708
+ 512,
3709
+ 512,
3710
+ 512,
3711
+ 512,
3712
+ 512,
3713
+ 512,
3714
+ 512,
3715
+ 512,
3716
+ 512,
3717
+ 512,
3718
+ 512,
3719
+ 512,
3720
+ 512,
3721
+ 512,
3722
+ 512,
3723
+ 512,
3724
+ 512,
3725
+ 512,
3726
+ 512,
3727
+ 512,
3728
+ 512,
3729
+ 512,
3730
+ 512,
3731
+ 512,
3732
+ 512,
3733
+ 512,
3734
+ 512,
3735
+ 512,
3736
+ 512,
3737
+ 512,
3738
+ 512,
3739
+ 512,
3740
+ 512,
3741
+ 512,
3742
+ 512,
3743
+ 512,
3744
+ 512,
3745
+ 512,
3746
+ 512,
3747
+ 512,
3748
+ 512,
3749
+ 512,
3750
+ 512,
3751
+ 512,
3752
+ 512,
3753
+ 512,
3754
+ 512,
3755
+ 512,
3756
+ 512,
3757
+ 512,
3758
+ 512,
3759
+ 512,
3760
+ 512,
3761
+ 512,
3762
+ 512,
3763
+ 512,
3764
+ 512,
3765
+ 512,
3766
+ 512,
3767
+ 512,
3768
+ 512,
3769
+ 512,
3770
+ 512,
3771
+ 512,
3772
+ 512,
3773
+ 512,
3774
+ 512,
3775
+ 512,
3776
+ 512,
3777
+ 512,
3778
+ 512,
3779
+ 512,
3780
+ 512,
3781
+ 512,
3782
+ 512,
3783
+ 512,
3784
+ 512
3785
+ ],
3786
+ [
3787
+ 512,
3788
+ 512,
3789
+ 512,
3790
+ 512,
3791
+ 512,
3792
+ 512,
3793
+ 512,
3794
+ 512,
3795
+ 512,
3796
+ 512,
3797
+ 512,
3798
+ 512,
3799
+ 512,
3800
+ 512,
3801
+ 512,
3802
+ 512,
3803
+ 512,
3804
+ 512,
3805
+ 512,
3806
+ 512,
3807
+ 512,
3808
+ 512,
3809
+ 512,
3810
+ 512,
3811
+ 512,
3812
+ 512,
3813
+ 512,
3814
+ 512,
3815
+ 512,
3816
+ 512,
3817
+ 512,
3818
+ 512,
3819
+ 512,
3820
+ 512,
3821
+ 512,
3822
+ 512,
3823
+ 512,
3824
+ 512,
3825
+ 512,
3826
+ 512,
3827
+ 512,
3828
+ 512,
3829
+ 512,
3830
+ 512,
3831
+ 512,
3832
+ 512,
3833
+ 512,
3834
+ 512,
3835
+ 512,
3836
+ 512,
3837
+ 512,
3838
+ 512,
3839
+ 512,
3840
+ 512,
3841
+ 512,
3842
+ 512,
3843
+ 512,
3844
+ 512,
3845
+ 512,
3846
+ 512,
3847
+ 512,
3848
+ 512,
3849
+ 512,
3850
+ 512,
3851
+ 512,
3852
+ 512,
3853
+ 512,
3854
+ 512,
3855
+ 512,
3856
+ 512,
3857
+ 512,
3858
+ 512,
3859
+ 512,
3860
+ 512,
3861
+ 512,
3862
+ 512,
3863
+ 512,
3864
+ 512,
3865
+ 512,
3866
+ 512,
3867
+ 512,
3868
+ 512,
3869
+ 512,
3870
+ 512,
3871
+ 512,
3872
+ 512,
3873
+ 512,
3874
+ 512,
3875
+ 512,
3876
+ 512,
3877
+ 512,
3878
+ 512,
3879
+ 512,
3880
+ 512,
3881
+ 512,
3882
+ 512,
3883
+ 512,
3884
+ 512,
3885
+ 512,
3886
+ 512,
3887
+ 512,
3888
+ 512,
3889
+ 512,
3890
+ 512,
3891
+ 512,
3892
+ 512,
3893
+ 512,
3894
+ 512,
3895
+ 512,
3896
+ 512,
3897
+ 512,
3898
+ 512,
3899
+ 512,
3900
+ 512,
3901
+ 512,
3902
+ 512,
3903
+ 512,
3904
+ 512,
3905
+ 512,
3906
+ 512,
3907
+ 512,
3908
+ 512,
3909
+ 512,
3910
+ 512,
3911
+ 512,
3912
+ 512,
3913
+ 512,
3914
+ 512
3915
+ ],
3916
+ [
3917
+ 512,
3918
+ 512,
3919
+ 512,
3920
+ 512,
3921
+ 512,
3922
+ 512,
3923
+ 512,
3924
+ 512,
3925
+ 512,
3926
+ 512,
3927
+ 512,
3928
+ 512,
3929
+ 512,
3930
+ 512,
3931
+ 512,
3932
+ 512,
3933
+ 512,
3934
+ 512,
3935
+ 512,
3936
+ 512,
3937
+ 512,
3938
+ 512,
3939
+ 512,
3940
+ 512,
3941
+ 512,
3942
+ 512,
3943
+ 512,
3944
+ 512,
3945
+ 512,
3946
+ 512,
3947
+ 512,
3948
+ 512,
3949
+ 512,
3950
+ 512,
3951
+ 512,
3952
+ 512,
3953
+ 512,
3954
+ 512,
3955
+ 512,
3956
+ 512,
3957
+ 512,
3958
+ 512,
3959
+ 512,
3960
+ 512,
3961
+ 512,
3962
+ 512,
3963
+ 512,
3964
+ 512,
3965
+ 512,
3966
+ 512,
3967
+ 512,
3968
+ 512,
3969
+ 512,
3970
+ 512,
3971
+ 512,
3972
+ 512,
3973
+ 512,
3974
+ 512,
3975
+ 512,
3976
+ 512,
3977
+ 512,
3978
+ 512,
3979
+ 512,
3980
+ 512,
3981
+ 512,
3982
+ 512,
3983
+ 512,
3984
+ 512,
3985
+ 512,
3986
+ 512,
3987
+ 512,
3988
+ 512,
3989
+ 512,
3990
+ 512,
3991
+ 512,
3992
+ 512,
3993
+ 512,
3994
+ 512,
3995
+ 512,
3996
+ 512,
3997
+ 512,
3998
+ 512,
3999
+ 512,
4000
+ 512,
4001
+ 512,
4002
+ 512,
4003
+ 512,
4004
+ 512,
4005
+ 512,
4006
+ 512,
4007
+ 512,
4008
+ 512,
4009
+ 512,
4010
+ 512,
4011
+ 512,
4012
+ 512,
4013
+ 512,
4014
+ 512,
4015
+ 512,
4016
+ 512,
4017
+ 512,
4018
+ 512,
4019
+ 512,
4020
+ 512,
4021
+ 512,
4022
+ 512,
4023
+ 512,
4024
+ 512,
4025
+ 512,
4026
+ 512,
4027
+ 512,
4028
+ 512,
4029
+ 512,
4030
+ 512,
4031
+ 512,
4032
+ 512,
4033
+ 512,
4034
+ 512,
4035
+ 512,
4036
+ 512,
4037
+ 512,
4038
+ 512,
4039
+ 512,
4040
+ 512,
4041
+ 512,
4042
+ 512,
4043
+ 512,
4044
+ 512
4045
+ ],
4046
+ [
4047
+ 512,
4048
+ 512,
4049
+ 512,
4050
+ 512,
4051
+ 512,
4052
+ 512,
4053
+ 512,
4054
+ 512,
4055
+ 512,
4056
+ 512,
4057
+ 512,
4058
+ 512,
4059
+ 512,
4060
+ 512,
4061
+ 512,
4062
+ 512,
4063
+ 512,
4064
+ 512,
4065
+ 512,
4066
+ 512,
4067
+ 512,
4068
+ 512,
4069
+ 512,
4070
+ 512,
4071
+ 512,
4072
+ 512,
4073
+ 512,
4074
+ 512,
4075
+ 512,
4076
+ 512,
4077
+ 512,
4078
+ 512,
4079
+ 512,
4080
+ 512,
4081
+ 512,
4082
+ 512,
4083
+ 512,
4084
+ 512,
4085
+ 512,
4086
+ 512,
4087
+ 512,
4088
+ 512,
4089
+ 512,
4090
+ 512,
4091
+ 512,
4092
+ 512,
4093
+ 512,
4094
+ 512,
4095
+ 512,
4096
+ 512,
4097
+ 512,
4098
+ 512,
4099
+ 512,
4100
+ 512,
4101
+ 512,
4102
+ 512,
4103
+ 512,
4104
+ 512,
4105
+ 512,
4106
+ 512,
4107
+ 512,
4108
+ 512,
4109
+ 512,
4110
+ 512,
4111
+ 512,
4112
+ 512,
4113
+ 512,
4114
+ 512,
4115
+ 512,
4116
+ 512,
4117
+ 512,
4118
+ 512,
4119
+ 512,
4120
+ 512,
4121
+ 512,
4122
+ 512,
4123
+ 512,
4124
+ 512,
4125
+ 512,
4126
+ 512,
4127
+ 512,
4128
+ 512,
4129
+ 512,
4130
+ 512,
4131
+ 512,
4132
+ 512,
4133
+ 512,
4134
+ 512,
4135
+ 512,
4136
+ 512,
4137
+ 512,
4138
+ 512,
4139
+ 512,
4140
+ 512,
4141
+ 512,
4142
+ 512,
4143
+ 512,
4144
+ 512,
4145
+ 512,
4146
+ 512,
4147
+ 512,
4148
+ 512,
4149
+ 512,
4150
+ 512,
4151
+ 512,
4152
+ 512,
4153
+ 512,
4154
+ 512,
4155
+ 512,
4156
+ 512,
4157
+ 512,
4158
+ 512,
4159
+ 512,
4160
+ 512,
4161
+ 512,
4162
+ 512,
4163
+ 512,
4164
+ 512,
4165
+ 512,
4166
+ 512,
4167
+ 512,
4168
+ 512,
4169
+ 512,
4170
+ 512,
4171
+ 512,
4172
+ 512,
4173
+ 512,
4174
+ 512
4175
+ ],
4176
+ [
4177
+ 512,
4178
+ 512,
4179
+ 512,
4180
+ 512,
4181
+ 512,
4182
+ 512,
4183
+ 512,
4184
+ 512,
4185
+ 512,
4186
+ 512,
4187
+ 512,
4188
+ 512,
4189
+ 512,
4190
+ 512,
4191
+ 512,
4192
+ 512,
4193
+ 512,
4194
+ 512,
4195
+ 512,
4196
+ 512,
4197
+ 512,
4198
+ 512,
4199
+ 512,
4200
+ 512,
4201
+ 512,
4202
+ 512,
4203
+ 512,
4204
+ 512,
4205
+ 512,
4206
+ 512,
4207
+ 512,
4208
+ 512,
4209
+ 512,
4210
+ 512,
4211
+ 512,
4212
+ 512,
4213
+ 512,
4214
+ 512,
4215
+ 512,
4216
+ 512,
4217
+ 512,
4218
+ 512,
4219
+ 512,
4220
+ 512,
4221
+ 512,
4222
+ 512,
4223
+ 512,
4224
+ 512,
4225
+ 512,
4226
+ 512,
4227
+ 512,
4228
+ 512,
4229
+ 512,
4230
+ 512,
4231
+ 512,
4232
+ 512,
4233
+ 512,
4234
+ 512,
4235
+ 512,
4236
+ 512,
4237
+ 512,
4238
+ 512,
4239
+ 512,
4240
+ 512,
4241
+ 512,
4242
+ 512,
4243
+ 512,
4244
+ 512,
4245
+ 512,
4246
+ 512,
4247
+ 512,
4248
+ 512,
4249
+ 512,
4250
+ 512,
4251
+ 512,
4252
+ 512,
4253
+ 512,
4254
+ 512,
4255
+ 512,
4256
+ 512,
4257
+ 512,
4258
+ 512,
4259
+ 512,
4260
+ 512,
4261
+ 512,
4262
+ 512,
4263
+ 512,
4264
+ 512,
4265
+ 512,
4266
+ 512,
4267
+ 512,
4268
+ 512,
4269
+ 512,
4270
+ 512,
4271
+ 512,
4272
+ 512,
4273
+ 512,
4274
+ 512,
4275
+ 512,
4276
+ 512,
4277
+ 512,
4278
+ 512,
4279
+ 512,
4280
+ 512,
4281
+ 512,
4282
+ 512,
4283
+ 512,
4284
+ 512,
4285
+ 512,
4286
+ 512,
4287
+ 512,
4288
+ 512,
4289
+ 512,
4290
+ 512,
4291
+ 512,
4292
+ 512,
4293
+ 512,
4294
+ 512,
4295
+ 512,
4296
+ 512,
4297
+ 512,
4298
+ 512,
4299
+ 512,
4300
+ 512,
4301
+ 512,
4302
+ 512,
4303
+ 512,
4304
+ 512
4305
+ ],
4306
+ [
4307
+ 512,
4308
+ 512,
4309
+ 512,
4310
+ 512,
4311
+ 512,
4312
+ 512,
4313
+ 512,
4314
+ 512,
4315
+ 512,
4316
+ 512,
4317
+ 512,
4318
+ 512,
4319
+ 512,
4320
+ 512,
4321
+ 512,
4322
+ 512,
4323
+ 512,
4324
+ 512,
4325
+ 512,
4326
+ 512,
4327
+ 512,
4328
+ 512,
4329
+ 512,
4330
+ 512,
4331
+ 512,
4332
+ 512,
4333
+ 512,
4334
+ 512,
4335
+ 512,
4336
+ 512,
4337
+ 512,
4338
+ 512,
4339
+ 512,
4340
+ 512,
4341
+ 512,
4342
+ 512,
4343
+ 512,
4344
+ 512,
4345
+ 512,
4346
+ 512,
4347
+ 512,
4348
+ 512,
4349
+ 512,
4350
+ 512,
4351
+ 512,
4352
+ 512,
4353
+ 512,
4354
+ 512,
4355
+ 512,
4356
+ 512,
4357
+ 512,
4358
+ 512,
4359
+ 512,
4360
+ 512,
4361
+ 512,
4362
+ 512,
4363
+ 512,
4364
+ 512,
4365
+ 512,
4366
+ 512,
4367
+ 512,
4368
+ 512,
4369
+ 512,
4370
+ 512,
4371
+ 512,
4372
+ 512,
4373
+ 512,
4374
+ 512,
4375
+ 512,
4376
+ 512,
4377
+ 512,
4378
+ 512,
4379
+ 512,
4380
+ 512,
4381
+ 512,
4382
+ 512,
4383
+ 512,
4384
+ 512,
4385
+ 512,
4386
+ 512,
4387
+ 512,
4388
+ 512,
4389
+ 512,
4390
+ 512,
4391
+ 512,
4392
+ 512,
4393
+ 512,
4394
+ 512,
4395
+ 512,
4396
+ 512,
4397
+ 512,
4398
+ 512,
4399
+ 512,
4400
+ 512,
4401
+ 512,
4402
+ 512,
4403
+ 512,
4404
+ 512,
4405
+ 512,
4406
+ 512,
4407
+ 512,
4408
+ 512,
4409
+ 512,
4410
+ 512,
4411
+ 512,
4412
+ 512,
4413
+ 512,
4414
+ 512,
4415
+ 512,
4416
+ 512,
4417
+ 512,
4418
+ 512,
4419
+ 512,
4420
+ 512,
4421
+ 512,
4422
+ 512,
4423
+ 512,
4424
+ 512,
4425
+ 512,
4426
+ 512,
4427
+ 512,
4428
+ 512,
4429
+ 512,
4430
+ 512,
4431
+ 512,
4432
+ 512,
4433
+ 512,
4434
+ 512
4435
+ ],
4436
+ [
4437
+ 512,
4438
+ 512,
4439
+ 512,
4440
+ 512,
4441
+ 512,
4442
+ 512,
4443
+ 512,
4444
+ 512,
4445
+ 512,
4446
+ 512,
4447
+ 512,
4448
+ 512,
4449
+ 512,
4450
+ 512,
4451
+ 512,
4452
+ 512,
4453
+ 512,
4454
+ 512,
4455
+ 512,
4456
+ 512,
4457
+ 512,
4458
+ 512,
4459
+ 512,
4460
+ 512,
4461
+ 512,
4462
+ 512,
4463
+ 512,
4464
+ 512,
4465
+ 512,
4466
+ 512,
4467
+ 512,
4468
+ 512,
4469
+ 512,
4470
+ 512,
4471
+ 512,
4472
+ 512,
4473
+ 512,
4474
+ 512,
4475
+ 512,
4476
+ 512,
4477
+ 512,
4478
+ 512,
4479
+ 512,
4480
+ 512,
4481
+ 512,
4482
+ 512,
4483
+ 512,
4484
+ 512,
4485
+ 512,
4486
+ 512,
4487
+ 512,
4488
+ 512,
4489
+ 512,
4490
+ 512,
4491
+ 512,
4492
+ 512,
4493
+ 512,
4494
+ 512,
4495
+ 512,
4496
+ 512,
4497
+ 512,
4498
+ 512,
4499
+ 512,
4500
+ 512,
4501
+ 512,
4502
+ 512,
4503
+ 512,
4504
+ 512,
4505
+ 512,
4506
+ 512,
4507
+ 512,
4508
+ 512,
4509
+ 512,
4510
+ 512,
4511
+ 512,
4512
+ 512,
4513
+ 512,
4514
+ 512,
4515
+ 512,
4516
+ 512,
4517
+ 512,
4518
+ 512,
4519
+ 512,
4520
+ 512,
4521
+ 512,
4522
+ 512,
4523
+ 512,
4524
+ 512,
4525
+ 512,
4526
+ 512,
4527
+ 512,
4528
+ 512,
4529
+ 512,
4530
+ 512,
4531
+ 512,
4532
+ 512,
4533
+ 512,
4534
+ 512,
4535
+ 512,
4536
+ 512,
4537
+ 512,
4538
+ 512,
4539
+ 512,
4540
+ 512,
4541
+ 512,
4542
+ 512,
4543
+ 512,
4544
+ 512,
4545
+ 512,
4546
+ 512,
4547
+ 512,
4548
+ 512,
4549
+ 512,
4550
+ 512,
4551
+ 512,
4552
+ 512,
4553
+ 512,
4554
+ 512,
4555
+ 512,
4556
+ 512,
4557
+ 512,
4558
+ 512,
4559
+ 512,
4560
+ 512,
4561
+ 512,
4562
+ 512,
4563
+ 512,
4564
+ 512
4565
+ ],
4566
+ [
4567
+ 512,
4568
+ 512,
4569
+ 512,
4570
+ 512,
4571
+ 512,
4572
+ 512,
4573
+ 512,
4574
+ 512,
4575
+ 512,
4576
+ 512,
4577
+ 512,
4578
+ 512,
4579
+ 512,
4580
+ 512,
4581
+ 512,
4582
+ 512,
4583
+ 512,
4584
+ 512,
4585
+ 512,
4586
+ 512,
4587
+ 512,
4588
+ 512,
4589
+ 512,
4590
+ 512,
4591
+ 512,
4592
+ 512,
4593
+ 512,
4594
+ 512,
4595
+ 512,
4596
+ 512,
4597
+ 512,
4598
+ 512,
4599
+ 512,
4600
+ 512,
4601
+ 512,
4602
+ 512,
4603
+ 512,
4604
+ 512,
4605
+ 512,
4606
+ 512,
4607
+ 512,
4608
+ 512,
4609
+ 512,
4610
+ 512,
4611
+ 512,
4612
+ 512,
4613
+ 512,
4614
+ 512,
4615
+ 512,
4616
+ 512,
4617
+ 512,
4618
+ 512,
4619
+ 512,
4620
+ 512,
4621
+ 512,
4622
+ 512,
4623
+ 512,
4624
+ 512,
4625
+ 512,
4626
+ 512,
4627
+ 512,
4628
+ 512,
4629
+ 512,
4630
+ 512,
4631
+ 512,
4632
+ 512,
4633
+ 512,
4634
+ 512,
4635
+ 512,
4636
+ 512,
4637
+ 512,
4638
+ 512,
4639
+ 512,
4640
+ 512,
4641
+ 512,
4642
+ 512,
4643
+ 512,
4644
+ 512,
4645
+ 512,
4646
+ 512,
4647
+ 512,
4648
+ 512,
4649
+ 512,
4650
+ 512,
4651
+ 512,
4652
+ 512,
4653
+ 512,
4654
+ 512,
4655
+ 512,
4656
+ 512,
4657
+ 512,
4658
+ 512,
4659
+ 512,
4660
+ 512,
4661
+ 512,
4662
+ 512,
4663
+ 512,
4664
+ 512,
4665
+ 512,
4666
+ 512,
4667
+ 512,
4668
+ 512,
4669
+ 512,
4670
+ 512,
4671
+ 512,
4672
+ 512,
4673
+ 512,
4674
+ 512,
4675
+ 512,
4676
+ 512,
4677
+ 512,
4678
+ 512,
4679
+ 512,
4680
+ 512,
4681
+ 512,
4682
+ 512,
4683
+ 512,
4684
+ 512,
4685
+ 512,
4686
+ 512,
4687
+ 512,
4688
+ 512,
4689
+ 512,
4690
+ 512,
4691
+ 512,
4692
+ 512,
4693
+ 512,
4694
+ 512
4695
+ ],
4696
+ [
4697
+ 512,
4698
+ 512,
4699
+ 512,
4700
+ 512,
4701
+ 512,
4702
+ 512,
4703
+ 512,
4704
+ 512,
4705
+ 512,
4706
+ 512,
4707
+ 512,
4708
+ 512,
4709
+ 512,
4710
+ 512,
4711
+ 512,
4712
+ 512,
4713
+ 512,
4714
+ 512,
4715
+ 512,
4716
+ 512,
4717
+ 512,
4718
+ 512,
4719
+ 512,
4720
+ 512,
4721
+ 512,
4722
+ 512,
4723
+ 512,
4724
+ 512,
4725
+ 512,
4726
+ 512,
4727
+ 512,
4728
+ 512,
4729
+ 512,
4730
+ 512,
4731
+ 512,
4732
+ 512,
4733
+ 512,
4734
+ 512,
4735
+ 512,
4736
+ 512,
4737
+ 512,
4738
+ 512,
4739
+ 512,
4740
+ 512,
4741
+ 512,
4742
+ 512,
4743
+ 512,
4744
+ 512,
4745
+ 512,
4746
+ 512,
4747
+ 512,
4748
+ 512,
4749
+ 512,
4750
+ 512,
4751
+ 512,
4752
+ 512,
4753
+ 512,
4754
+ 512,
4755
+ 512,
4756
+ 512,
4757
+ 512,
4758
+ 512,
4759
+ 512,
4760
+ 512,
4761
+ 512,
4762
+ 512,
4763
+ 512,
4764
+ 512,
4765
+ 512,
4766
+ 512,
4767
+ 512,
4768
+ 512,
4769
+ 512,
4770
+ 512,
4771
+ 512,
4772
+ 512,
4773
+ 512,
4774
+ 512,
4775
+ 512,
4776
+ 512,
4777
+ 512,
4778
+ 512,
4779
+ 512,
4780
+ 512,
4781
+ 512,
4782
+ 512,
4783
+ 512,
4784
+ 512,
4785
+ 512,
4786
+ 512,
4787
+ 512,
4788
+ 512,
4789
+ 512,
4790
+ 512,
4791
+ 512,
4792
+ 512,
4793
+ 512,
4794
+ 512,
4795
+ 512,
4796
+ 512,
4797
+ 512,
4798
+ 512,
4799
+ 512,
4800
+ 512,
4801
+ 512,
4802
+ 512,
4803
+ 512,
4804
+ 512,
4805
+ 512,
4806
+ 512,
4807
+ 512,
4808
+ 512,
4809
+ 512,
4810
+ 512,
4811
+ 512,
4812
+ 512,
4813
+ 512,
4814
+ 512,
4815
+ 512,
4816
+ 512,
4817
+ 512,
4818
+ 512,
4819
+ 512,
4820
+ 512,
4821
+ 512,
4822
+ 512,
4823
+ 512,
4824
+ 512
4825
+ ],
4826
+ [
4827
+ 512,
4828
+ 512,
4829
+ 512,
4830
+ 512,
4831
+ 512,
4832
+ 512,
4833
+ 512,
4834
+ 512,
4835
+ 512,
4836
+ 512,
4837
+ 512,
4838
+ 512,
4839
+ 512,
4840
+ 512,
4841
+ 512,
4842
+ 512,
4843
+ 512,
4844
+ 512,
4845
+ 512,
4846
+ 512,
4847
+ 512,
4848
+ 512,
4849
+ 512,
4850
+ 512,
4851
+ 512,
4852
+ 512,
4853
+ 512,
4854
+ 512,
4855
+ 512,
4856
+ 512,
4857
+ 512,
4858
+ 512,
4859
+ 512,
4860
+ 512,
4861
+ 512,
4862
+ 512,
4863
+ 512,
4864
+ 512,
4865
+ 512,
4866
+ 512,
4867
+ 512,
4868
+ 512,
4869
+ 512,
4870
+ 512,
4871
+ 512,
4872
+ 512,
4873
+ 512,
4874
+ 512,
4875
+ 512,
4876
+ 512,
4877
+ 512,
4878
+ 512,
4879
+ 512,
4880
+ 512,
4881
+ 512,
4882
+ 512,
4883
+ 512,
4884
+ 512,
4885
+ 512,
4886
+ 512,
4887
+ 512,
4888
+ 512,
4889
+ 512,
4890
+ 512,
4891
+ 512,
4892
+ 512,
4893
+ 512,
4894
+ 512,
4895
+ 512,
4896
+ 512,
4897
+ 512,
4898
+ 512,
4899
+ 512,
4900
+ 512,
4901
+ 512,
4902
+ 512,
4903
+ 512,
4904
+ 512,
4905
+ 512,
4906
+ 512,
4907
+ 512,
4908
+ 512,
4909
+ 512,
4910
+ 512,
4911
+ 512,
4912
+ 512,
4913
+ 512,
4914
+ 512,
4915
+ 512,
4916
+ 512,
4917
+ 512,
4918
+ 512,
4919
+ 512,
4920
+ 512,
4921
+ 512,
4922
+ 512,
4923
+ 512,
4924
+ 512,
4925
+ 512,
4926
+ 512,
4927
+ 512,
4928
+ 512,
4929
+ 512,
4930
+ 512,
4931
+ 512,
4932
+ 512,
4933
+ 512,
4934
+ 512,
4935
+ 512,
4936
+ 512,
4937
+ 512,
4938
+ 512,
4939
+ 512,
4940
+ 512,
4941
+ 512,
4942
+ 512,
4943
+ 512,
4944
+ 512,
4945
+ 512,
4946
+ 512,
4947
+ 512,
4948
+ 512,
4949
+ 512,
4950
+ 512,
4951
+ 512,
4952
+ 512,
4953
+ 512,
4954
+ 512
4955
+ ],
4956
+ [
4957
+ 512,
4958
+ 512,
4959
+ 512,
4960
+ 512,
4961
+ 512,
4962
+ 512,
4963
+ 512,
4964
+ 512,
4965
+ 512,
4966
+ 512,
4967
+ 512,
4968
+ 512,
4969
+ 512,
4970
+ 512,
4971
+ 512,
4972
+ 512,
4973
+ 512,
4974
+ 512,
4975
+ 512,
4976
+ 512,
4977
+ 512,
4978
+ 512,
4979
+ 512,
4980
+ 512,
4981
+ 512,
4982
+ 512,
4983
+ 512,
4984
+ 512,
4985
+ 512,
4986
+ 512,
4987
+ 512,
4988
+ 512,
4989
+ 512,
4990
+ 512,
4991
+ 512,
4992
+ 512,
4993
+ 512,
4994
+ 512,
4995
+ 512,
4996
+ 512,
4997
+ 512,
4998
+ 512,
4999
+ 512,
5000
+ 512,
5001
+ 512,
5002
+ 512,
5003
+ 512,
5004
+ 512,
5005
+ 512,
5006
+ 512,
5007
+ 512,
5008
+ 512,
5009
+ 512,
5010
+ 512,
5011
+ 512,
5012
+ 512,
5013
+ 512,
5014
+ 512,
5015
+ 512,
5016
+ 512,
5017
+ 512,
5018
+ 512,
5019
+ 512,
5020
+ 512,
5021
+ 512,
5022
+ 512,
5023
+ 512,
5024
+ 512,
5025
+ 512,
5026
+ 512,
5027
+ 512,
5028
+ 512,
5029
+ 512,
5030
+ 512,
5031
+ 512,
5032
+ 512,
5033
+ 512,
5034
+ 512,
5035
+ 512,
5036
+ 512,
5037
+ 512,
5038
+ 512,
5039
+ 512,
5040
+ 512,
5041
+ 512,
5042
+ 512,
5043
+ 512,
5044
+ 512,
5045
+ 512,
5046
+ 512,
5047
+ 512,
5048
+ 512,
5049
+ 512,
5050
+ 512,
5051
+ 512,
5052
+ 512,
5053
+ 512,
5054
+ 512,
5055
+ 512,
5056
+ 512,
5057
+ 512,
5058
+ 512,
5059
+ 512,
5060
+ 512,
5061
+ 512,
5062
+ 512,
5063
+ 512,
5064
+ 512,
5065
+ 512,
5066
+ 512,
5067
+ 512,
5068
+ 512,
5069
+ 512,
5070
+ 512,
5071
+ 512,
5072
+ 512,
5073
+ 512,
5074
+ 512,
5075
+ 512,
5076
+ 512,
5077
+ 512,
5078
+ 512,
5079
+ 512,
5080
+ 512,
5081
+ 512,
5082
+ 512,
5083
+ 512,
5084
+ 512
5085
+ ],
5086
+ [
5087
+ 512,
5088
+ 512,
5089
+ 512,
5090
+ 512,
5091
+ 512,
5092
+ 512,
5093
+ 512,
5094
+ 512,
5095
+ 512,
5096
+ 512,
5097
+ 512,
5098
+ 512,
5099
+ 512,
5100
+ 512,
5101
+ 512,
5102
+ 512,
5103
+ 512,
5104
+ 512,
5105
+ 512,
5106
+ 512,
5107
+ 512,
5108
+ 512,
5109
+ 512,
5110
+ 512,
5111
+ 512,
5112
+ 512,
5113
+ 512,
5114
+ 512,
5115
+ 512,
5116
+ 512,
5117
+ 512,
5118
+ 512,
5119
+ 512,
5120
+ 512,
5121
+ 512,
5122
+ 512,
5123
+ 512,
5124
+ 512,
5125
+ 512,
5126
+ 512,
5127
+ 512,
5128
+ 512,
5129
+ 512,
5130
+ 512,
5131
+ 512,
5132
+ 512,
5133
+ 512,
5134
+ 512,
5135
+ 512,
5136
+ 512,
5137
+ 512,
5138
+ 512,
5139
+ 512,
5140
+ 512,
5141
+ 512,
5142
+ 512,
5143
+ 512,
5144
+ 512,
5145
+ 512,
5146
+ 512,
5147
+ 512,
5148
+ 512,
5149
+ 512,
5150
+ 512,
5151
+ 512,
5152
+ 512,
5153
+ 512,
5154
+ 512,
5155
+ 512,
5156
+ 512,
5157
+ 512,
5158
+ 512,
5159
+ 512,
5160
+ 512,
5161
+ 512,
5162
+ 512,
5163
+ 512,
5164
+ 512,
5165
+ 512,
5166
+ 512,
5167
+ 512,
5168
+ 512,
5169
+ 512,
5170
+ 512,
5171
+ 512,
5172
+ 512,
5173
+ 512,
5174
+ 512,
5175
+ 512,
5176
+ 512,
5177
+ 512,
5178
+ 512,
5179
+ 512,
5180
+ 512,
5181
+ 512,
5182
+ 512,
5183
+ 512,
5184
+ 512,
5185
+ 512,
5186
+ 512,
5187
+ 512,
5188
+ 512,
5189
+ 512,
5190
+ 512,
5191
+ 512,
5192
+ 512,
5193
+ 512,
5194
+ 512,
5195
+ 512,
5196
+ 512,
5197
+ 512,
5198
+ 512,
5199
+ 512,
5200
+ 512,
5201
+ 512,
5202
+ 512,
5203
+ 512,
5204
+ 512,
5205
+ 512,
5206
+ 512,
5207
+ 512,
5208
+ 512,
5209
+ 512,
5210
+ 512,
5211
+ 512,
5212
+ 512,
5213
+ 512,
5214
+ 512
5215
+ ]
5216
+ ],
5217
+ "full_attention_interval": 4,
5218
+ "glean_metadata": {
5219
+ "keep_fraction": 0.5,
5220
+ "method": "REAP whole-expert prune (reap), uniform per-layer keep",
5221
+ "score": {
5222
+ "alpha": 1,
5223
+ "b": 1.0,
5224
+ "beta": 1
5225
+ },
5226
+ "source_model": "Qwen/Qwen3.6-35B-A3B",
5227
+ "stats_path": "outputs/qwen35_stats1024.pt"
5228
+ },
5229
+ "head_dim": 256,
5230
+ "hidden_act": "silu",
5231
+ "hidden_size": 2048,
5232
+ "initializer_range": 0.02,
5233
+ "layer_types": [
5234
+ "linear_attention",
5235
+ "linear_attention",
5236
+ "linear_attention",
5237
+ "full_attention",
5238
+ "linear_attention",
5239
+ "linear_attention",
5240
+ "linear_attention",
5241
+ "full_attention",
5242
+ "linear_attention",
5243
+ "linear_attention",
5244
+ "linear_attention",
5245
+ "full_attention",
5246
+ "linear_attention",
5247
+ "linear_attention",
5248
+ "linear_attention",
5249
+ "full_attention",
5250
+ "linear_attention",
5251
+ "linear_attention",
5252
+ "linear_attention",
5253
+ "full_attention",
5254
+ "linear_attention",
5255
+ "linear_attention",
5256
+ "linear_attention",
5257
+ "full_attention",
5258
+ "linear_attention",
5259
+ "linear_attention",
5260
+ "linear_attention",
5261
+ "full_attention",
5262
+ "linear_attention",
5263
+ "linear_attention",
5264
+ "linear_attention",
5265
+ "full_attention",
5266
+ "linear_attention",
5267
+ "linear_attention",
5268
+ "linear_attention",
5269
+ "full_attention",
5270
+ "linear_attention",
5271
+ "linear_attention",
5272
+ "linear_attention",
5273
+ "full_attention"
5274
+ ],
5275
+ "linear_conv_kernel_dim": 4,
5276
+ "linear_key_head_dim": 128,
5277
+ "linear_num_key_heads": 16,
5278
+ "linear_num_value_heads": 32,
5279
+ "linear_value_head_dim": 128,
5280
+ "mamba_ssm_dtype": "float32",
5281
+ "max_position_embeddings": 262144,
5282
+ "model_type": "pruned_qwen3_5_moe",
5283
+ "moe_intermediate_size": 512,
5284
+ "mtp_num_hidden_layers": 1,
5285
+ "mtp_use_dedicated_embeddings": false,
5286
+ "num_attention_heads": 16,
5287
+ "num_experts": 256,
5288
+ "num_experts_per_tok": 8,
5289
+ "num_hidden_layers": 40,
5290
+ "num_key_value_heads": 2,
5291
+ "output_router_logits": false,
5292
+ "pad_token_id": null,
5293
+ "partial_rotary_factor": 0.25,
5294
+ "rms_norm_eps": 1e-06,
5295
+ "rope_parameters": {
5296
+ "mrope_interleaved": true,
5297
+ "mrope_section": [
5298
+ 11,
5299
+ 11,
5300
+ 10
5301
+ ],
5302
+ "partial_rotary_factor": 0.25,
5303
+ "rope_theta": 10000000,
5304
+ "rope_type": "default"
5305
+ },
5306
+ "router_aux_loss_coef": 0.001,
5307
+ "shared_expert_intermediate_size": 512,
5308
+ "tie_word_embeddings": false,
5309
+ "transformers_version": "5.13.1",
5310
+ "use_cache": true,
5311
+ "vocab_size": 248320
5312
+ }
qwen35_reap_keep50/configuration_pruned_qwen3_5_moe.py ADDED
@@ -0,0 +1,31 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ """Configuration for GLEAN-pruned Qwen3.5-MoE: variable-width, variable-count experts.
2
+
3
+ Requires the ``qwen3_5_moe`` architecture (transformers >= 5.x); checkpoints
4
+ carrying this module import the upstream classes rather than vendoring the
5
+ ~1.5k-line hybrid (GatedDeltaNet + attention) stack.
6
+ """
7
+
8
+ from transformers.models.qwen3_5_moe.configuration_qwen3_5_moe import (
9
+ Qwen3_5MoeTextConfig,
10
+ )
11
+
12
+
13
+ class PrunedQwen3_5MoeTextConfig(Qwen3_5MoeTextConfig):
14
+ """Qwen3_5MoeTextConfig plus a per-(layer, expert) width table.
15
+
16
+ ``expert_widths[l]`` lists the SwiGLU intermediate width of each surviving
17
+ routed expert in decoder layer ``l``, in original expert order. Lists are
18
+ ragged: layers may keep different numbers of experts (deleted experts
19
+ simply don't appear — the router in layer ``l`` has
20
+ ``len(expert_widths[l])`` rows), and each width may differ. ``None`` means
21
+ an unpruned model (uniform ``num_experts`` x ``moe_intermediate_size``).
22
+
23
+ The inherited ``num_experts`` / ``moe_intermediate_size`` keep their
24
+ ORIGINAL (pre-pruning) values for provenance; the width table is
25
+ authoritative for the built architecture. The shared expert and its
26
+ sigmoid gate are untouched by pruning and keep their stock config fields.
27
+ """
28
+
29
+ model_type = "pruned_qwen3_5_moe"
30
+
31
+ expert_widths: list[list[int]] | None = None
qwen35_reap_keep50/generation_config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_from_model_config": true,
3
+ "bos_token_id": 248044,
4
+ "eos_token_id": 248044,
5
+ "output_attentions": false,
6
+ "output_hidden_states": false,
7
+ "transformers_version": "5.13.1",
8
+ "use_cache": true
9
+ }
qwen35_reap_keep50/modeling_pruned_qwen3_5_moe.py ADDED
@@ -0,0 +1,122 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ """GLEAN-pruned Qwen3.5-MoE: HF-loadable model with ragged (variable-width) experts.
2
+
3
+ Same pattern as ``modeling_pruned_olmoe``: ``super().__init__`` builds the
4
+ uniform architecture from the config (free on the meta device during
5
+ ``from_pretrained``), then every MoE block's router and fused experts are
6
+ rebuilt to their pruned shape — surviving experts only, each at its own width,
7
+ router rows sliced to match. The token mixers (GatedDeltaNet / attention), the
8
+ shared expert, and its sigmoid gate are stock and untouched.
9
+
10
+ Unlike OLMoE's per-expert ``nn.Linear`` modules, Qwen3.5-MoE fuses experts
11
+ into 3-D parameters, which cannot hold ragged widths — so the pruned experts
12
+ module stores per-expert 2-D parameters in ``nn.ParameterList``s
13
+ (``gate_up_projs.{j}`` ``[2*w_j, H]``, ``down_projs.{j}`` ``[H, w_j]``) and
14
+ runs the same routed per-expert loop as upstream ``Qwen3_5MoeExperts.forward``
15
+ (verified against transformers 5.13.1; re-verify after upgrades).
16
+
17
+ Caveats: ``output_router_logits=True`` (the load-balancing aux loss) assumes a
18
+ uniform ``config.num_experts`` and is unsupported on ragged models; the eager
19
+ per-expert loop bypasses the fused ``_experts_implementation`` kernels.
20
+ """
21
+
22
+ import torch
23
+ import torch.nn.functional as F
24
+ from torch import nn
25
+ from transformers.activations import ACT2FN
26
+ from transformers.models.qwen3_5_moe.modeling_qwen3_5_moe import (
27
+ Qwen3_5MoeForCausalLM,
28
+ )
29
+
30
+ from .configuration_pruned_qwen3_5_moe import PrunedQwen3_5MoeTextConfig
31
+
32
+
33
+ class PrunedQwen3_5MoeExperts(nn.Module):
34
+ """Ragged replacement for the fused ``Qwen3_5MoeExperts``.
35
+
36
+ Forward replicates the upstream routed per-expert loop exactly, indexing
37
+ per-expert parameters instead of slices of a stacked 3-D tensor.
38
+ """
39
+
40
+ def __init__(self, config: PrunedQwen3_5MoeTextConfig, widths: list[int]):
41
+ super().__init__()
42
+ self.num_experts = len(widths)
43
+ self.hidden_dim = config.hidden_size
44
+ self.gate_up_projs = nn.ParameterList(
45
+ nn.Parameter(torch.empty(2 * w, config.hidden_size)) for w in widths
46
+ )
47
+ self.down_projs = nn.ParameterList(
48
+ nn.Parameter(torch.empty(config.hidden_size, w)) for w in widths
49
+ )
50
+ self.act_fn = ACT2FN[config.hidden_act]
51
+
52
+ def forward(
53
+ self,
54
+ hidden_states: torch.Tensor,
55
+ top_k_index: torch.Tensor,
56
+ top_k_weights: torch.Tensor,
57
+ ) -> torch.Tensor:
58
+ final_hidden_states = torch.zeros_like(hidden_states)
59
+ with torch.no_grad():
60
+ expert_mask = F.one_hot(top_k_index, num_classes=self.num_experts)
61
+ expert_mask = expert_mask.permute(2, 1, 0)
62
+ expert_hit = torch.greater(expert_mask.sum(dim=(-1, -2)), 0).nonzero()
63
+
64
+ for expert_idx in expert_hit:
65
+ expert_idx = int(expert_idx[0])
66
+ top_k_pos, token_idx = torch.where(expert_mask[expert_idx])
67
+ current_state = hidden_states[token_idx]
68
+ gate, up = F.linear(current_state, self.gate_up_projs[expert_idx]).chunk(2, dim=-1)
69
+ current_hidden_states = self.act_fn(gate) * up
70
+ current_hidden_states = F.linear(current_hidden_states, self.down_projs[expert_idx])
71
+ current_hidden_states = current_hidden_states * top_k_weights[token_idx, top_k_pos, None]
72
+ final_hidden_states.index_add_(
73
+ 0, token_idx, current_hidden_states.to(final_hidden_states.dtype)
74
+ )
75
+
76
+ return final_hidden_states
77
+
78
+
79
+ class PrunedQwen3_5MoeForCausalLM(Qwen3_5MoeForCausalLM):
80
+ """Qwen3.5-MoE with per-layer surviving-expert lists at per-expert widths."""
81
+
82
+ config_class = PrunedQwen3_5MoeTextConfig
83
+
84
+ def __init__(self, config: PrunedQwen3_5MoeTextConfig):
85
+ super().__init__(config)
86
+ widths_table = getattr(config, "expert_widths", None)
87
+ if widths_table is None:
88
+ return # unpruned: plain Qwen3.5-MoE
89
+ if len(widths_table) != len(self.model.layers):
90
+ raise ValueError(
91
+ f"expert_widths has {len(widths_table)} rows but the model has "
92
+ f"{len(self.model.layers)} decoder layers"
93
+ )
94
+ for layer, widths in zip(self.model.layers, widths_table):
95
+ if any(w <= 0 for w in widths):
96
+ raise ValueError("expert_widths must list surviving experts only (>0)")
97
+ block = layer.mlp
98
+ if len(widths) < block.gate.top_k:
99
+ raise ValueError(
100
+ f"a layer keeps {len(widths)} experts < top_k={block.gate.top_k}"
101
+ )
102
+ # keep the stock router class (OutputRecorder isinstance, forward
103
+ # unchanged) but shrink it to the surviving experts' rows
104
+ block.gate.num_experts = len(widths)
105
+ block.gate.weight = nn.Parameter(
106
+ torch.empty(
107
+ len(widths),
108
+ config.hidden_size,
109
+ dtype=block.gate.weight.dtype,
110
+ device=block.gate.weight.device,
111
+ )
112
+ )
113
+ block.experts = PrunedQwen3_5MoeExperts(config, list(widths))
114
+
115
+ @torch.no_grad()
116
+ def _init_weights(self, module):
117
+ super()._init_weights(module)
118
+ # the base isinstance(Qwen3_5MoeExperts) branch never sees our ragged
119
+ # module; raw nn.Parameters get no default init otherwise
120
+ if isinstance(module, PrunedQwen3_5MoeExperts):
121
+ for p in list(module.gate_up_projs) + list(module.down_projs):
122
+ nn.init.normal_(p, mean=0.0, std=self.config.initializer_range)
qwen35_reap_keep50/reap_verify.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "keep": 0.5,
3
+ "criterion": "reap",
4
+ "ppl": 14.886332511901855,
5
+ "n_params_b": 18.543997568,
6
+ "eval_seq": 32,
7
+ "seq_len": 2048,
8
+ "dataset": "c4"
9
+ }
qwen35_reap_keep50/tokenizer_config.json ADDED
@@ -0,0 +1,32 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_prefix_space": false,
3
+ "audio_bos_token": "<|audio_start|>",
4
+ "audio_eos_token": "<|audio_end|>",
5
+ "audio_token": "<|audio_pad|>",
6
+ "backend": "tokenizers",
7
+ "bos_token": null,
8
+ "clean_up_tokenization_spaces": false,
9
+ "eos_token": "<|im_end|>",
10
+ "errors": "replace",
11
+ "image_token": "<|image_pad|>",
12
+ "is_local": false,
13
+ "local_files_only": false,
14
+ "model_max_length": 262144,
15
+ "model_specific_special_tokens": {
16
+ "audio_bos_token": "<|audio_start|>",
17
+ "audio_eos_token": "<|audio_end|>",
18
+ "audio_token": "<|audio_pad|>",
19
+ "image_token": "<|image_pad|>",
20
+ "video_token": "<|video_pad|>",
21
+ "vision_bos_token": "<|vision_start|>",
22
+ "vision_eos_token": "<|vision_end|>"
23
+ },
24
+ "pad_token": "<|endoftext|>",
25
+ "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
26
+ "split_special_tokens": false,
27
+ "tokenizer_class": "Qwen2Tokenizer",
28
+ "unk_token": null,
29
+ "video_token": "<|video_pad|>",
30
+ "vision_bos_token": "<|vision_start|>",
31
+ "vision_eos_token": "<|vision_end|>"
32
+ }
redo_policy/score_divergence.json ADDED
@@ -0,0 +1,38 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ [
2
+ {
3
+ "keep": 25,
4
+ "validated": true,
5
+ "kept": 262144,
6
+ "total": 1048576,
7
+ "jaccard_kept": 0.6221832369530846,
8
+ "disagree_all": 0.11645317077636719,
9
+ "math_pruned_general_keeps": 0.0776354471842448,
10
+ "dead_math": 442,
11
+ "dead_general": 405,
12
+ "dead_overlap": 359
13
+ },
14
+ {
15
+ "keep": 50,
16
+ "validated": true,
17
+ "kept": 524288,
18
+ "total": 1048576,
19
+ "jaccard_kept": 0.7086550917816143,
20
+ "disagree_all": 0.17051124572753906,
21
+ "math_pruned_general_keeps": 0.17051124572753906,
22
+ "dead_math": 217,
23
+ "dead_general": 186,
24
+ "dead_overlap": 150
25
+ },
26
+ {
27
+ "keep": 75,
28
+ "validated": true,
29
+ "kept": 786432,
30
+ "total": 1048576,
31
+ "jaccard_kept": 0.828006127197458,
32
+ "disagree_all": 0.14113235473632812,
33
+ "math_pruned_general_keeps": 0.28226470947265625,
34
+ "dead_math": 88,
35
+ "dead_general": 82,
36
+ "dead_overlap": 56
37
+ }
38
+ ]
teacher_trajectories/dolci_math_curated.stats.json ADDED
@@ -0,0 +1,29 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "input": [
3
+ "outputs/teacher_trajectories/dolci_math_chat_n4_t0.8.jsonl",
4
+ "outputs/teacher_trajectories/dolci_math_chat_n4_t0.8.chunks27-32.jsonl"
5
+ ],
6
+ "output": "outputs/teacher_trajectories/dolci_math_curated.jsonl",
7
+ "min_agreement": 1,
8
+ "require_finished": true,
9
+ "require_gold_match": true,
10
+ "max_prompt_idx": null,
11
+ "prompts": 63977,
12
+ "selected_prompts": 6673,
13
+ "selected_trajectories": 10110,
14
+ "selected_completion_tokens": 4946904,
15
+ "mean_completion_tokens": 489.3080118694362,
16
+ "prompt_keep_rate": 0.10430310892977164,
17
+ "consensus_vote_counts_by_prompt": {
18
+ "1": 4527,
19
+ "3": 563,
20
+ "2": 1219,
21
+ "4": 364
22
+ },
23
+ "reasons": {
24
+ "unfinished": 16552,
25
+ "gold_mismatch": 40740,
26
+ "kept": 6673,
27
+ "no_answer": 12
28
+ }
29
+ }
teacher_trajectories/dolci_math_curated_opd.stats.json ADDED
@@ -0,0 +1,30 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "input": [
3
+ "outputs/teacher_trajectories/dolci_math_chat_n4_t0.8.jsonl",
4
+ "outputs/teacher_trajectories/dolci_math_chat_n4_t0.8.chunks27-32.jsonl"
5
+ ],
6
+ "output": "outputs/teacher_trajectories/dolci_math_curated_opd.jsonl",
7
+ "min_agreement": 1,
8
+ "require_finished": false,
9
+ "unfinished_min_agreement": 2,
10
+ "require_gold_match": true,
11
+ "max_prompt_idx": null,
12
+ "prompts": 63977,
13
+ "selected_prompts": 7322,
14
+ "selected_trajectories": 12115,
15
+ "selected_completion_tokens": 6476634,
16
+ "mean_completion_tokens": 534.5962855963681,
17
+ "prompt_keep_rate": 0.11444737952701753,
18
+ "consensus_vote_counts_by_prompt": {
19
+ "1": 4112,
20
+ "3": 767,
21
+ "2": 2035,
22
+ "4": 408
23
+ },
24
+ "reasons": {
25
+ "gold_mismatch": 53958,
26
+ "kept": 7322,
27
+ "low_agreement_unfinished": 2694,
28
+ "no_answer": 3
29
+ }
30
+ }
teacher_trajectories/dolci_math_nogold_matched.stats.json ADDED
@@ -0,0 +1,10 @@
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "input": "outputs/teacher_trajectories/dolci_math_nogold.jsonl",
3
+ "output": "outputs/teacher_trajectories/dolci_math_nogold_matched.jsonl",
4
+ "seed": 1224,
5
+ "target_tokens": 6476634,
6
+ "selected_tokens": 6476712,
7
+ "selected_trajectories": 9918,
8
+ "unique_prompts": 8748,
9
+ "pool_trajectories": 57420
10
+ }
teacher_trajectories/dolci_math_nogold_matched_top128.score.log ADDED
@@ -0,0 +1,647 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ reconstructing exact chat prompts ...
2
+ loading teacher on cuda:0 ...
3
+
4
+ shard 00000: 128/9918 records | 86,738 tokens | 8.71 records/s
5
+ shard 00001: 256/9918 records | 167,821 tokens | 8.96 records/s
6
+ shard 00002: 384/9918 records | 252,309 tokens | 9.04 records/s
7
+ shard 00003: 512/9918 records | 334,964 tokens | 9.09 records/s
8
+ shard 00004: 640/9918 records | 417,335 tokens | 9.13 records/s
9
+ shard 00005: 768/9918 records | 500,766 tokens | 9.14 records/s
10
+ shard 00006: 896/9918 records | 581,333 tokens | 9.16 records/s
11
+ shard 00007: 1024/9918 records | 661,718 tokens | 9.18 records/s
12
+ shard 00008: 1152/9918 records | 748,886 tokens | 9.19 records/s
13
+ shard 00009: 1280/9918 records | 836,216 tokens | 9.19 records/s
14
+ shard 00010: 1408/9918 records | 920,766 tokens | 9.20 records/s
15
+ shard 00011: 1536/9918 records | 1,006,776 tokens | 9.21 records/s
16
+ shard 00012: 1664/9918 records | 1,089,613 tokens | 9.21 records/s
17
+ shard 00013: 1792/9918 records | 1,176,076 tokens | 9.21 records/s
18
+ shard 00014: 1920/9918 records | 1,261,191 tokens | 9.21 records/s
19
+ shard 00015: 2048/9918 records | 1,344,996 tokens | 9.21 records/s
20
+ shard 00016: 2176/9918 records | 1,428,549 tokens | 9.22 records/s
21
+ shard 00017: 2304/9918 records | 1,509,478 tokens | 9.22 records/s
22
+ shard 00018: 2432/9918 records | 1,595,067 tokens | 9.23 records/s
23
+ shard 00019: 2560/9918 records | 1,678,874 tokens | 9.23 records/s
24
+ shard 00020: 2688/9918 records | 1,763,838 tokens | 9.23 records/s
25
+ shard 00021: 2816/9918 records | 1,838,106 tokens | 9.23 records/s
26
+ shard 00022: 2944/9918 records | 1,915,047 tokens | 9.23 records/s
27
+ shard 00023: 3072/9918 records | 1,991,387 tokens | 9.24 records/s
28
+ shard 00024: 3200/9918 records | 2,067,294 tokens | 9.25 records/s
29
+ shard 00025: 3328/9918 records | 2,140,318 tokens | 9.26 records/s
30
+ shard 00026: 3456/9918 records | 2,216,508 tokens | 9.27 records/s
31
+ shard 00027: 3584/9918 records | 2,288,848 tokens | 9.27 records/s
32
+ shard 00028: 3712/9918 records | 2,365,928 tokens | 9.27 records/s
33
+ shard 00029: 3840/9918 records | 2,439,058 tokens | 9.28 records/s
34
+ shard 00030: 3968/9918 records | 2,517,122 tokens | 9.28 records/s
35
+ shard 00031: 4096/9918 records | 2,593,654 tokens | 9.28 records/s
36
+ shard 00032: 4224/9918 records | 2,667,587 tokens | 9.29 records/s
37
+ shard 00033: 4352/9918 records | 2,742,631 tokens | 9.29 records/s
38
+ shard 00034: 4480/9918 records | 2,814,842 tokens | 9.29 records/s
39
+ shard 00035: 4608/9918 records | 2,890,157 tokens | 9.30 records/s
40
+ shard 00036: 4736/9918 records | 2,966,385 tokens | 9.30 records/s
41
+ shard 00037: 4864/9918 records | 3,039,781 tokens | 9.30 records/s
42
+ shard 00038: 4992/9918 records | 3,116,074 tokens | 9.31 records/s
43
+ shard 00039: 5120/9918 records | 3,203,251 tokens | 9.30 records/s
44
+ shard 00040: 5248/9918 records | 3,293,236 tokens | 9.30 records/s
45
+ shard 00041: 5376/9918 records | 3,382,951 tokens | 9.29 records/s
46
+ shard 00042: 5504/9918 records | 3,471,094 tokens | 9.28 records/s
47
+ shard 00043: 5632/9918 records | 3,562,058 tokens | 9.27 records/s
48
+ shard 00044: 5760/9918 records | 3,650,120 tokens | 9.26 records/s
49
+ shard 00045: 5888/9918 records | 3,738,179 tokens | 9.25 records/s
50
+ shard 00046: 6016/9918 records | 3,827,029 tokens | 9.24 records/s
51
+ shard 00047: 6144/9918 records | 3,917,892 tokens | 9.23 records/s
52
+ shard 00048: 6272/9918 records | 4,005,920 tokens | 9.22 records/s
53
+ shard 00049: 6400/9918 records | 4,095,241 tokens | 9.21 records/s
54
+ shard 00050: 6528/9918 records | 4,183,072 tokens | 9.20 records/s
55
+ shard 00051: 6656/9918 records | 4,269,396 tokens | 9.20 records/s
56
+ shard 00052: 6784/9918 records | 4,353,524 tokens | 9.20 records/s
57
+ shard 00053: 6912/9918 records | 4,445,816 tokens | 9.19 records/s
58
+ shard 00054: 7040/9918 records | 4,537,739 tokens | 9.18 records/s
59
+ shard 00055: 7168/9918 records | 4,627,901 tokens | 9.18 records/s
60
+ shard 00056: 7296/9918 records | 4,716,890 tokens | 9.17 records/s
61
+ shard 00057: 7424/9918 records | 4,802,719 tokens | 9.17 records/s
62
+ shard 00058: 7552/9918 records | 4,888,043 tokens | 9.17 records/s
63
+ shard 00059: 7680/9918 records | 4,972,545 tokens | 9.17 records/s
64
+ shard 00060: 7808/9918 records | 5,056,045 tokens | 9.17 records/s
65
+ shard 00061: 7936/9918 records | 5,144,939 tokens | 9.17 records/s
66
+ shard 00062: 8064/9918 records | 5,228,021 tokens | 9.17 records/s
67
+ shard 00063: 8192/9918 records | 5,311,363 tokens | 9.17 records/s
68
+ shard 00064: 8320/9918 records | 5,395,159 tokens | 9.18 records/s
69
+ shard 00065: 8448/9918 records | 5,478,974 tokens | 9.18 records/s
70
+ shard 00066: 8576/9918 records | 5,562,357 tokens | 9.18 records/s
71
+ shard 00067: 8704/9918 records | 5,648,649 tokens | 9.18 records/s
72
+ shard 00068: 8832/9918 records | 5,732,310 tokens | 9.18 records/s
73
+ shard 00069: 8960/9918 records | 5,814,685 tokens | 9.18 records/s
74
+ shard 00070: 9088/9918 records | 5,902,712 tokens | 9.18 records/s
75
+ shard 00071: 9216/9918 records | 5,991,814 tokens | 9.18 records/s
76
+ shard 00072: 9344/9918 records | 6,081,881 tokens | 9.18 records/s
77
+ shard 00073: 9472/9918 records | 6,174,078 tokens | 9.18 records/s
78
+ shard 00074: 9600/9918 records | 6,263,194 tokens | 9.18 records/s
79
+ shard 00075: 9728/9918 records | 6,345,278 tokens | 9.18 records/s
80
+ shard 00076: 9856/9918 records | 6,436,191 tokens | 9.18 records/s
81
+ shard 00077: 9918/9918 records | 6,476,712 tokens | 9.18 records/s
82
+ {
83
+ "version": 1,
84
+ "complete": true,
85
+ "teacher": "allenai/OLMoE-1B-7B-0125-Instruct",
86
+ "tokenizer": "allenai/OLMoE-1B-7B-0125-Instruct",
87
+ "source": "/home/henry/Documents/PythonProjects/variable-reap/outputs/teacher_trajectories/dolci_math_nogold_matched.jsonl",
88
+ "source_size": 56024209,
89
+ "frame": "chat",
90
+ "topk": 128,
91
+ "max_prompt_len": 1024,
92
+ "max_seq_len": 2048,
93
+ "pad_token_id": 50280,
94
+ "eos_token_id": 50279,
95
+ "total_records": 9918,
96
+ "total_tokens": 6476712,
97
+ "shards": [
98
+ {
99
+ "path": "shard_00000.pt",
100
+ "records": 128,
101
+ "tokens": 86738,
102
+ "first_record": 0,
103
+ "last_record": 127
104
+ },
105
+ {
106
+ "path": "shard_00001.pt",
107
+ "records": 128,
108
+ "tokens": 81083,
109
+ "first_record": 128,
110
+ "last_record": 255
111
+ },
112
+ {
113
+ "path": "shard_00002.pt",
114
+ "records": 128,
115
+ "tokens": 84488,
116
+ "first_record": 256,
117
+ "last_record": 383
118
+ },
119
+ {
120
+ "path": "shard_00003.pt",
121
+ "records": 128,
122
+ "tokens": 82655,
123
+ "first_record": 384,
124
+ "last_record": 511
125
+ },
126
+ {
127
+ "path": "shard_00004.pt",
128
+ "records": 128,
129
+ "tokens": 82371,
130
+ "first_record": 512,
131
+ "last_record": 639
132
+ },
133
+ {
134
+ "path": "shard_00005.pt",
135
+ "records": 128,
136
+ "tokens": 83431,
137
+ "first_record": 640,
138
+ "last_record": 767
139
+ },
140
+ {
141
+ "path": "shard_00006.pt",
142
+ "records": 128,
143
+ "tokens": 80567,
144
+ "first_record": 768,
145
+ "last_record": 895
146
+ },
147
+ {
148
+ "path": "shard_00007.pt",
149
+ "records": 128,
150
+ "tokens": 80385,
151
+ "first_record": 896,
152
+ "last_record": 1023
153
+ },
154
+ {
155
+ "path": "shard_00008.pt",
156
+ "records": 128,
157
+ "tokens": 87168,
158
+ "first_record": 1024,
159
+ "last_record": 1151
160
+ },
161
+ {
162
+ "path": "shard_00009.pt",
163
+ "records": 128,
164
+ "tokens": 87330,
165
+ "first_record": 1152,
166
+ "last_record": 1279
167
+ },
168
+ {
169
+ "path": "shard_00010.pt",
170
+ "records": 128,
171
+ "tokens": 84550,
172
+ "first_record": 1280,
173
+ "last_record": 1407
174
+ },
175
+ {
176
+ "path": "shard_00011.pt",
177
+ "records": 128,
178
+ "tokens": 86010,
179
+ "first_record": 1408,
180
+ "last_record": 1535
181
+ },
182
+ {
183
+ "path": "shard_00012.pt",
184
+ "records": 128,
185
+ "tokens": 82837,
186
+ "first_record": 1536,
187
+ "last_record": 1663
188
+ },
189
+ {
190
+ "path": "shard_00013.pt",
191
+ "records": 128,
192
+ "tokens": 86463,
193
+ "first_record": 1664,
194
+ "last_record": 1791
195
+ },
196
+ {
197
+ "path": "shard_00014.pt",
198
+ "records": 128,
199
+ "tokens": 85115,
200
+ "first_record": 1792,
201
+ "last_record": 1919
202
+ },
203
+ {
204
+ "path": "shard_00015.pt",
205
+ "records": 128,
206
+ "tokens": 83805,
207
+ "first_record": 1920,
208
+ "last_record": 2047
209
+ },
210
+ {
211
+ "path": "shard_00016.pt",
212
+ "records": 128,
213
+ "tokens": 83553,
214
+ "first_record": 2048,
215
+ "last_record": 2175
216
+ },
217
+ {
218
+ "path": "shard_00017.pt",
219
+ "records": 128,
220
+ "tokens": 80929,
221
+ "first_record": 2176,
222
+ "last_record": 2303
223
+ },
224
+ {
225
+ "path": "shard_00018.pt",
226
+ "records": 128,
227
+ "tokens": 85589,
228
+ "first_record": 2304,
229
+ "last_record": 2431
230
+ },
231
+ {
232
+ "path": "shard_00019.pt",
233
+ "records": 128,
234
+ "tokens": 83807,
235
+ "first_record": 2432,
236
+ "last_record": 2559
237
+ },
238
+ {
239
+ "path": "shard_00020.pt",
240
+ "records": 128,
241
+ "tokens": 84964,
242
+ "first_record": 2560,
243
+ "last_record": 2687
244
+ },
245
+ {
246
+ "path": "shard_00021.pt",
247
+ "records": 128,
248
+ "tokens": 74268,
249
+ "first_record": 2688,
250
+ "last_record": 2815
251
+ },
252
+ {
253
+ "path": "shard_00022.pt",
254
+ "records": 128,
255
+ "tokens": 76941,
256
+ "first_record": 2816,
257
+ "last_record": 2943
258
+ },
259
+ {
260
+ "path": "shard_00023.pt",
261
+ "records": 128,
262
+ "tokens": 76340,
263
+ "first_record": 2944,
264
+ "last_record": 3071
265
+ },
266
+ {
267
+ "path": "shard_00024.pt",
268
+ "records": 128,
269
+ "tokens": 75907,
270
+ "first_record": 3072,
271
+ "last_record": 3199
272
+ },
273
+ {
274
+ "path": "shard_00025.pt",
275
+ "records": 128,
276
+ "tokens": 73024,
277
+ "first_record": 3200,
278
+ "last_record": 3327
279
+ },
280
+ {
281
+ "path": "shard_00026.pt",
282
+ "records": 128,
283
+ "tokens": 76190,
284
+ "first_record": 3328,
285
+ "last_record": 3455
286
+ },
287
+ {
288
+ "path": "shard_00027.pt",
289
+ "records": 128,
290
+ "tokens": 72340,
291
+ "first_record": 3456,
292
+ "last_record": 3583
293
+ },
294
+ {
295
+ "path": "shard_00028.pt",
296
+ "records": 128,
297
+ "tokens": 77080,
298
+ "first_record": 3584,
299
+ "last_record": 3711
300
+ },
301
+ {
302
+ "path": "shard_00029.pt",
303
+ "records": 128,
304
+ "tokens": 73130,
305
+ "first_record": 3712,
306
+ "last_record": 3839
307
+ },
308
+ {
309
+ "path": "shard_00030.pt",
310
+ "records": 128,
311
+ "tokens": 78064,
312
+ "first_record": 3840,
313
+ "last_record": 3967
314
+ },
315
+ {
316
+ "path": "shard_00031.pt",
317
+ "records": 128,
318
+ "tokens": 76532,
319
+ "first_record": 3968,
320
+ "last_record": 4095
321
+ },
322
+ {
323
+ "path": "shard_00032.pt",
324
+ "records": 128,
325
+ "tokens": 73933,
326
+ "first_record": 4096,
327
+ "last_record": 4223
328
+ },
329
+ {
330
+ "path": "shard_00033.pt",
331
+ "records": 128,
332
+ "tokens": 75044,
333
+ "first_record": 4224,
334
+ "last_record": 4351
335
+ },
336
+ {
337
+ "path": "shard_00034.pt",
338
+ "records": 128,
339
+ "tokens": 72211,
340
+ "first_record": 4352,
341
+ "last_record": 4479
342
+ },
343
+ {
344
+ "path": "shard_00035.pt",
345
+ "records": 128,
346
+ "tokens": 75315,
347
+ "first_record": 4480,
348
+ "last_record": 4607
349
+ },
350
+ {
351
+ "path": "shard_00036.pt",
352
+ "records": 128,
353
+ "tokens": 76228,
354
+ "first_record": 4608,
355
+ "last_record": 4735
356
+ },
357
+ {
358
+ "path": "shard_00037.pt",
359
+ "records": 128,
360
+ "tokens": 73396,
361
+ "first_record": 4736,
362
+ "last_record": 4863
363
+ },
364
+ {
365
+ "path": "shard_00038.pt",
366
+ "records": 128,
367
+ "tokens": 76293,
368
+ "first_record": 4864,
369
+ "last_record": 4991
370
+ },
371
+ {
372
+ "path": "shard_00039.pt",
373
+ "records": 128,
374
+ "tokens": 87177,
375
+ "first_record": 4992,
376
+ "last_record": 5119
377
+ },
378
+ {
379
+ "path": "shard_00040.pt",
380
+ "records": 128,
381
+ "tokens": 89985,
382
+ "first_record": 5120,
383
+ "last_record": 5247
384
+ },
385
+ {
386
+ "path": "shard_00041.pt",
387
+ "records": 128,
388
+ "tokens": 89715,
389
+ "first_record": 5248,
390
+ "last_record": 5375
391
+ },
392
+ {
393
+ "path": "shard_00042.pt",
394
+ "records": 128,
395
+ "tokens": 88143,
396
+ "first_record": 5376,
397
+ "last_record": 5503
398
+ },
399
+ {
400
+ "path": "shard_00043.pt",
401
+ "records": 128,
402
+ "tokens": 90964,
403
+ "first_record": 5504,
404
+ "last_record": 5631
405
+ },
406
+ {
407
+ "path": "shard_00044.pt",
408
+ "records": 128,
409
+ "tokens": 88062,
410
+ "first_record": 5632,
411
+ "last_record": 5759
412
+ },
413
+ {
414
+ "path": "shard_00045.pt",
415
+ "records": 128,
416
+ "tokens": 88059,
417
+ "first_record": 5760,
418
+ "last_record": 5887
419
+ },
420
+ {
421
+ "path": "shard_00046.pt",
422
+ "records": 128,
423
+ "tokens": 88850,
424
+ "first_record": 5888,
425
+ "last_record": 6015
426
+ },
427
+ {
428
+ "path": "shard_00047.pt",
429
+ "records": 128,
430
+ "tokens": 90863,
431
+ "first_record": 6016,
432
+ "last_record": 6143
433
+ },
434
+ {
435
+ "path": "shard_00048.pt",
436
+ "records": 128,
437
+ "tokens": 88028,
438
+ "first_record": 6144,
439
+ "last_record": 6271
440
+ },
441
+ {
442
+ "path": "shard_00049.pt",
443
+ "records": 128,
444
+ "tokens": 89321,
445
+ "first_record": 6272,
446
+ "last_record": 6399
447
+ },
448
+ {
449
+ "path": "shard_00050.pt",
450
+ "records": 128,
451
+ "tokens": 87831,
452
+ "first_record": 6400,
453
+ "last_record": 6527
454
+ },
455
+ {
456
+ "path": "shard_00051.pt",
457
+ "records": 128,
458
+ "tokens": 86324,
459
+ "first_record": 6528,
460
+ "last_record": 6655
461
+ },
462
+ {
463
+ "path": "shard_00052.pt",
464
+ "records": 128,
465
+ "tokens": 84128,
466
+ "first_record": 6656,
467
+ "last_record": 6783
468
+ },
469
+ {
470
+ "path": "shard_00053.pt",
471
+ "records": 128,
472
+ "tokens": 92292,
473
+ "first_record": 6784,
474
+ "last_record": 6911
475
+ },
476
+ {
477
+ "path": "shard_00054.pt",
478
+ "records": 128,
479
+ "tokens": 91923,
480
+ "first_record": 6912,
481
+ "last_record": 7039
482
+ },
483
+ {
484
+ "path": "shard_00055.pt",
485
+ "records": 128,
486
+ "tokens": 90162,
487
+ "first_record": 7040,
488
+ "last_record": 7167
489
+ },
490
+ {
491
+ "path": "shard_00056.pt",
492
+ "records": 128,
493
+ "tokens": 88989,
494
+ "first_record": 7168,
495
+ "last_record": 7295
496
+ },
497
+ {
498
+ "path": "shard_00057.pt",
499
+ "records": 128,
500
+ "tokens": 85829,
501
+ "first_record": 7296,
502
+ "last_record": 7423
503
+ },
504
+ {
505
+ "path": "shard_00058.pt",
506
+ "records": 128,
507
+ "tokens": 85324,
508
+ "first_record": 7424,
509
+ "last_record": 7551
510
+ },
511
+ {
512
+ "path": "shard_00059.pt",
513
+ "records": 128,
514
+ "tokens": 84502,
515
+ "first_record": 7552,
516
+ "last_record": 7679
517
+ },
518
+ {
519
+ "path": "shard_00060.pt",
520
+ "records": 128,
521
+ "tokens": 83500,
522
+ "first_record": 7680,
523
+ "last_record": 7807
524
+ },
525
+ {
526
+ "path": "shard_00061.pt",
527
+ "records": 128,
528
+ "tokens": 88894,
529
+ "first_record": 7808,
530
+ "last_record": 7935
531
+ },
532
+ {
533
+ "path": "shard_00062.pt",
534
+ "records": 128,
535
+ "tokens": 83082,
536
+ "first_record": 7936,
537
+ "last_record": 8063
538
+ },
539
+ {
540
+ "path": "shard_00063.pt",
541
+ "records": 128,
542
+ "tokens": 83342,
543
+ "first_record": 8064,
544
+ "last_record": 8191
545
+ },
546
+ {
547
+ "path": "shard_00064.pt",
548
+ "records": 128,
549
+ "tokens": 83796,
550
+ "first_record": 8192,
551
+ "last_record": 8319
552
+ },
553
+ {
554
+ "path": "shard_00065.pt",
555
+ "records": 128,
556
+ "tokens": 83815,
557
+ "first_record": 8320,
558
+ "last_record": 8447
559
+ },
560
+ {
561
+ "path": "shard_00066.pt",
562
+ "records": 128,
563
+ "tokens": 83383,
564
+ "first_record": 8448,
565
+ "last_record": 8575
566
+ },
567
+ {
568
+ "path": "shard_00067.pt",
569
+ "records": 128,
570
+ "tokens": 86292,
571
+ "first_record": 8576,
572
+ "last_record": 8703
573
+ },
574
+ {
575
+ "path": "shard_00068.pt",
576
+ "records": 128,
577
+ "tokens": 83661,
578
+ "first_record": 8704,
579
+ "last_record": 8831
580
+ },
581
+ {
582
+ "path": "shard_00069.pt",
583
+ "records": 128,
584
+ "tokens": 82375,
585
+ "first_record": 8832,
586
+ "last_record": 8959
587
+ },
588
+ {
589
+ "path": "shard_00070.pt",
590
+ "records": 128,
591
+ "tokens": 88027,
592
+ "first_record": 8960,
593
+ "last_record": 9087
594
+ },
595
+ {
596
+ "path": "shard_00071.pt",
597
+ "records": 128,
598
+ "tokens": 89102,
599
+ "first_record": 9088,
600
+ "last_record": 9215
601
+ },
602
+ {
603
+ "path": "shard_00072.pt",
604
+ "records": 128,
605
+ "tokens": 90067,
606
+ "first_record": 9216,
607
+ "last_record": 9343
608
+ },
609
+ {
610
+ "path": "shard_00073.pt",
611
+ "records": 128,
612
+ "tokens": 92197,
613
+ "first_record": 9344,
614
+ "last_record": 9471
615
+ },
616
+ {
617
+ "path": "shard_00074.pt",
618
+ "records": 128,
619
+ "tokens": 89116,
620
+ "first_record": 9472,
621
+ "last_record": 9599
622
+ },
623
+ {
624
+ "path": "shard_00075.pt",
625
+ "records": 128,
626
+ "tokens": 82084,
627
+ "first_record": 9600,
628
+ "last_record": 9727
629
+ },
630
+ {
631
+ "path": "shard_00076.pt",
632
+ "records": 128,
633
+ "tokens": 90913,
634
+ "first_record": 9728,
635
+ "last_record": 9855
636
+ },
637
+ {
638
+ "path": "shard_00077.pt",
639
+ "records": 62,
640
+ "tokens": 40521,
641
+ "first_record": 9856,
642
+ "last_record": 9917
643
+ }
644
+ ],
645
+ "requested_records": 9918,
646
+ "source_records": 9918
647
+ }
teacher_trajectories/gen_state.chunks27-32.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"chunks_done": 32, "n_chunks": 32, "end_chunk": 32, "samples_per_prompt": 4}
teacher_trajectories/generalgen.log ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ 104214 prompts, 51 chunks, generating [0, 51)
2
+ 104214 prompts, 51 chunks, generating [0, 10)
3
+ chunk 1/10: 3.5M tokens in 1069s
teacher_trajectories/generalgen_A.log ADDED
@@ -0,0 +1,4 @@
 
 
 
 
 
1
+ 104214 prompts, 51 chunks, generating [1, 3)
2
+ chunk 2/3: 3.5M tokens in 1418s
3
+ chunk 3/3: 3.5M tokens in 1520s
4
+ TEACHER TRAJECTORIES DONE
teacher_trajectories/generalgen_C.log ADDED
@@ -0,0 +1,4 @@
 
 
 
 
 
1
+ 104214 prompts, 51 chunks, generating [5, 7)
2
+ chunk 6/7: 3.4M tokens in 1014s
3
+ chunk 7/7: 3.5M tokens in 1040s
4
+ TEACHER TRAJECTORIES DONE
teacher_trajectories/server_general.log ADDED
The diff for this file is too large to render. See raw diff
 
teacher_trajectories/server_general_A.log ADDED
@@ -0,0 +1,41 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ (APIServer pid=113486) INFO 07-17 07:11:17 [api_utils.py:339]
2
+ (APIServer pid=113486) INFO 07-17 07:11:17 [api_utils.py:339] █ █ █▄ ▄█
3
+ (APIServer pid=113486) INFO 07-17 07:11:17 [api_utils.py:339] ▄▄ ▄█ █ █ █ ▀▄▀ █ version 0.25.0
4
+ (APIServer pid=113486) INFO 07-17 07:11:17 [api_utils.py:339] █▄█▀ █ █ █ █ model allenai/OLMoE-1B-7B-0125-Instruct
5
+ (APIServer pid=113486) INFO 07-17 07:11:17 [api_utils.py:339] ▀▀ ▀▀▀▀▀ ▀▀▀▀▀ ▀ ▀
6
+ (APIServer pid=113486) INFO 07-17 07:11:17 [api_utils.py:339]
7
+ (APIServer pid=113486) INFO 07-17 07:11:17 [api_utils.py:273] non-default args: {'model_tag': 'allenai/OLMoE-1B-7B-0125-Instruct', 'host': '127.0.0.1', 'port': 8383, 'model': 'allenai/OLMoE-1B-7B-0125-Instruct', 'max_model_len': 2048, 'enforce_eager': True, 'served_model_name': ['student'], 'gpu_memory_utilization': 0.85}
8
+ (APIServer pid=113486) Traceback (most recent call last):
9
+ (APIServer pid=113486) File "<frozen runpy>", line 198, in _run_module_as_main
10
+ (APIServer pid=113486) File "<frozen runpy>", line 88, in _run_code
11
+ (APIServer pid=113486) File "/home/henry/Documents/PythonProjects/variable-reap/vllm-plugin/.venv25/lib/python3.12/site-packages/vllm/entrypoints/cli/main.py", line 101, in <module>
12
+ (APIServer pid=113486) main()
13
+ (APIServer pid=113486) File "/home/henry/Documents/PythonProjects/variable-reap/vllm-plugin/.venv25/lib/python3.12/site-packages/vllm/entrypoints/cli/main.py", line 95, in main
14
+ (APIServer pid=113486) args.dispatch_function(args)
15
+ (APIServer pid=113486) File "/home/henry/Documents/PythonProjects/variable-reap/vllm-plugin/.venv25/lib/python3.12/site-packages/vllm/entrypoints/cli/serve.py", line 148, in cmd
16
+ (APIServer pid=113486) uvloop.run(run_server(args))
17
+ (APIServer pid=113486) File "/home/henry/Documents/PythonProjects/variable-reap/vllm-plugin/.venv25/lib/python3.12/site-packages/uvloop/__init__.py", line 96, in run
18
+ (APIServer pid=113486) return __asyncio.run(
19
+ (APIServer pid=113486) ^^^^^^^^^^^^^^
20
+ (APIServer pid=113486) File "/home/henry/.local/share/uv/python/cpython-3.12.12-linux-x86_64-gnu/lib/python3.12/asyncio/runners.py", line 195, in run
21
+ (APIServer pid=113486) return runner.run(main)
22
+ (APIServer pid=113486) ^^^^^^^^^^^^^^^^
23
+ (APIServer pid=113486) File "/home/henry/.local/share/uv/python/cpython-3.12.12-linux-x86_64-gnu/lib/python3.12/asyncio/runners.py", line 118, in run
24
+ (APIServer pid=113486) return self._loop.run_until_complete(task)
25
+ (APIServer pid=113486) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
26
+ (APIServer pid=113486) File "uvloop/loop.pyx", line 1518, in uvloop.loop.Loop.run_until_complete
27
+ (APIServer pid=113486) File "/home/henry/Documents/PythonProjects/variable-reap/vllm-plugin/.venv25/lib/python3.12/site-packages/uvloop/__init__.py", line 48, in wrapper
28
+ (APIServer pid=113486) return await main
29
+ (APIServer pid=113486) ^^^^^^^^^^
30
+ (APIServer pid=113486) File "/home/henry/Documents/PythonProjects/variable-reap/vllm-plugin/.venv25/lib/python3.12/site-packages/vllm/entrypoints/openai/api_server.py", line 697, in run_server
31
+ (APIServer pid=113486) listen_address, sock = setup_server(args, reuse_port=False)
32
+ (APIServer pid=113486) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
33
+ (APIServer pid=113486) File "/home/henry/Documents/PythonProjects/variable-reap/vllm-plugin/.venv25/lib/python3.12/site-packages/vllm/tracing/otel.py", line 178, in sync_wrapper
34
+ (APIServer pid=113486) return func(*args, **kwargs)
35
+ (APIServer pid=113486) ^^^^^^^^^^^^^^^^^^^^^
36
+ (APIServer pid=113486) File "/home/henry/Documents/PythonProjects/variable-reap/vllm-plugin/.venv25/lib/python3.12/site-packages/vllm/entrypoints/openai/api_server.py", line 576, in setup_server
37
+ (APIServer pid=113486) sock = create_server_socket(sock_addr, reuse_port=reuse_port)
38
+ (APIServer pid=113486) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
39
+ (APIServer pid=113486) File "/home/henry/Documents/PythonProjects/variable-reap/vllm-plugin/.venv25/lib/python3.12/site-packages/vllm/entrypoints/openai/api_server.py", line 525, in create_server_socket
40
+ (APIServer pid=113486) sock.bind(addr)
41
+ (APIServer pid=113486) OSError: [Errno 98] Address already in use
teacher_trajectories/server_general_C.log ADDED
@@ -0,0 +1,331 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ (APIServer pid=114640) INFO 07-17 07:15:16 [api_utils.py:339]
2
+ (APIServer pid=114640) INFO 07-17 07:15:16 [api_utils.py:339] █ █ █▄ ▄█
3
+ (APIServer pid=114640) INFO 07-17 07:15:16 [api_utils.py:339] ▄▄ ▄█ █ █ █ ▀▄▀ █ version 0.25.0
4
+ (APIServer pid=114640) INFO 07-17 07:15:16 [api_utils.py:339] █▄█▀ █ █ █ █ model allenai/OLMoE-1B-7B-0125-Instruct
5
+ (APIServer pid=114640) INFO 07-17 07:15:16 [api_utils.py:339] ▀▀ ▀▀▀▀▀ ▀▀▀▀▀ ▀ ▀
6
+ (APIServer pid=114640) INFO 07-17 07:15:16 [api_utils.py:339]
7
+ (APIServer pid=114640) INFO 07-17 07:15:16 [api_utils.py:273] non-default args: {'model_tag': 'allenai/OLMoE-1B-7B-0125-Instruct', 'host': '127.0.0.1', 'port': 8385, 'model': 'allenai/OLMoE-1B-7B-0125-Instruct', 'max_model_len': 2048, 'enforce_eager': True, 'served_model_name': ['student'], 'gpu_memory_utilization': 0.85}
8
+ (APIServer pid=114640) INFO 07-17 07:15:16 [model.py:619] Resolved architecture: OlmoeForCausalLM
9
+ (APIServer pid=114640) INFO 07-17 07:15:16 [model.py:1776] Using max model len 2048
10
+ (APIServer pid=114640) INFO 07-17 07:15:16 [vllm.py:1042] Asynchronous scheduling is enabled.
11
+ (APIServer pid=114640) WARNING 07-17 07:15:16 [vllm.py:1096] Enforce eager set, disabling torch.compile and CUDAGraphs. This is equivalent to setting -cc.mode=none -cc.cudagraph_mode=none
12
+ (APIServer pid=114640) WARNING 07-17 07:15:16 [vllm.py:1144] Inductor compilation was disabled by user settings, optimizations settings that are only active during inductor compilation will be ignored.
13
+ (APIServer pid=114640) INFO 07-17 07:15:16 [kernel.py:292] Final IR op priority after setting platform defaults: IrOpPriorityConfig(rms_norm=['vllm_c', 'native'], fused_add_rms_norm=['vllm_c', 'native'])
14
+ (APIServer pid=114640) INFO 07-17 07:15:17 [vllm.py:1322] Cudagraph is disabled under eager mode
15
+ (APIServer pid=114640) INFO 07-17 07:15:17 [compilation.py:312] Enabled custom fusions: norm_quant, act_quant
16
+ (EngineCore pid=114758) INFO 07-17 07:15:24 [core.py:114] Initializing a V1 LLM engine (v0.25.0) with config: model='allenai/OLMoE-1B-7B-0125-Instruct', speculative_config=None, tokenizer='allenai/OLMoE-1B-7B-0125-Instruct', skip_tokenizer_init=False, tokenizer_mode=auto, revision=None, tokenizer_revision=None, trust_remote_code=False, dtype=torch.bfloat16, max_seq_len=2048, download_dir=None, load_format=auto, tensor_parallel_size=1, pipeline_parallel_size=1, data_parallel_size=1, decode_context_parallel_size=1, dcp_comm_backend=ag_rs, disable_custom_all_reduce=False, quantization=None, quantization_config=None, enforce_eager=True, enable_return_routed_experts=False, kv_cache_dtype=auto, device_config=cuda, structured_outputs_config=StructuredOutputsConfig(backend='auto', disable_any_whitespace=False, disable_additional_properties=False, reasoning_parser='', reasoning_parser_plugin='', enable_in_reasoning=False), observability_config=ObservabilityConfig(show_hidden_metrics_for_version=None, otlp_traces_endpoint=None, collect_detailed_traces=None, kv_cache_metrics=False, kv_cache_metrics_sample=0.01, cudagraph_metrics=False, enable_layerwise_nvtx_tracing=False, enable_mfu_metrics=False, enable_mm_processor_stats=False, enable_logging_iteration_details=False, jit_monitor_mode='warn', jit_monitor_verbose=False), seed=0, served_model_name=student, enable_prefix_caching=True, enable_chunked_prefill=True, pooler_config=None, compilation_config={'mode': <CompilationMode.NONE: 0>, 'debug_dump_path': None, 'cache_dir': '', 'compile_cache_save_format': 'binary', 'backend': 'inductor', 'custom_ops': ['all'], 'ir_enable_torch_wrap': False, 'splitting_ops': [], 'compile_mm_encoder': False, 'cudagraph_mm_encoder': False, 'encoder_cudagraph_token_budgets': [], 'encoder_cudagraph_max_vision_items_per_batch': 0, 'encoder_cudagraph_max_frames_per_batch': None, 'compile_sizes': [], 'compile_ranges_endpoints': [2048], 'inductor_compile_config': {'enable_auto_functionalized_v2': False, 'size_asserts': False, 'alignment_asserts': False, 'scalar_asserts': False, 'combo_kernels': True, 'benchmark_combo_kernel': True}, 'inductor_passes': {}, 'cudagraph_mode': <CUDAGraphMode.NONE: 0>, 'cudagraph_num_of_warmups': 0, 'cudagraph_capture_sizes': [], 'cudagraph_copy_inputs': False, 'cudagraph_specialize_lora': True, 'use_inductor_graph_partition': False, 'pass_config': {'fuse_norm_quant': True, 'fuse_act_quant': True, 'fuse_attn_quant': False, 'enable_sp': False, 'fuse_gemm_comms': False, 'fuse_allreduce_rms': False, 'fuse_rope_kvcache_cat_mla': False, 'fuse_act_padding': False}, 'max_cudagraph_capture_size': 0, 'dynamic_shapes_config': {'type': <DynamicShapesType.BACKED: 'backed'>, 'evaluate_guards': False, 'assume_32_bit_indexing': False}, 'local_cache_dir': None, 'fast_moe_cold_start': False, 'static_all_moe_layers': []}, kernel_config=KernelConfig(ir_op_priority=IrOpPriorityConfig(rms_norm=['vllm_c', 'native'], fused_add_rms_norm=['vllm_c', 'native']), enable_flashinfer_autotune=True, enable_cutedsl_warmup=True, moe_backend='auto', linear_backend='auto')
17
+ (EngineCore pid=114758) INFO 07-17 07:15:25 [parallel_state.py:1607] world_size=1 rank=0 local_rank=0 distributed_init_method=tcp://192.168.0.15:44865 backend=nccl
18
+ (EngineCore pid=114758) INFO 07-17 07:15:25 [parallel_state.py:1942] rank 0 in world size 1 is assigned as DP rank 0, PP rank 0, PCP rank 0, TP rank 0, EP rank 0, EPLB rank N/A
19
+ (EngineCore pid=114758) INFO 07-17 07:15:26 [topk_topp_sampler.py:55] Using FlashInfer for top-p & top-k sampling.
20
+ (EngineCore pid=114758) INFO 07-17 07:15:26 [gpu_model_runner.py:5209] Starting to load model allenai/OLMoE-1B-7B-0125-Instruct...
21
+ (EngineCore pid=114758) INFO 07-17 07:15:26 [cuda.py:476] Using FLASH_ATTN attention backend out of potential backends: ['FLASH_ATTN', 'FLASHINFER', 'TRITON_ATTN', 'FLEX_ATTENTION'].
22
+ (EngineCore pid=114758) INFO 07-17 07:15:26 [flash_attn.py:718] Using FlashAttention version 2
23
+ (EngineCore pid=114758) INFO 07-17 07:15:26 [unquantized.py:262] Using TRITON Unquantized MoE backend out of potential backends: ['FlashInfer TRTLLM', 'FlashInfer CUTLASS', 'TRITON', 'BATCHED_TRITON'].
24
+ (EngineCore pid=114758) INFO 07-17 07:15:27 [weight_utils.py:849] Filesystem type for checkpoints: EXT4. Checkpoint size: 12.89 GiB. Available RAM: 102.39 GiB.
25
+ (EngineCore pid=114758) INFO 07-17 07:15:27 [weight_utils.py:872] Auto-prefetch is disabled because the filesystem (EXT4) is not a recognized network FS (NFS/Lustre). If you want to force prefetching, start vLLM with --safetensors-load-strategy=prefetch.
26
+ (EngineCore pid=114758)
27
+ (EngineCore pid=114758)
28
+ (EngineCore pid=114758)
29
+ (EngineCore pid=114758)
30
+ (EngineCore pid=114758)
31
+ (EngineCore pid=114758)
32
+ (EngineCore pid=114758) INFO 07-17 07:15:29 [default_loader.py:430] Loading weights took 1.89 seconds
33
+ (EngineCore pid=114758) INFO 07-17 07:15:29 [unquantized.py:334] Using MoEPrepareAndFinalizeNoDPEPModular
34
+ (EngineCore pid=114758) INFO 07-17 07:15:29 [gpu_model_runner.py:5306] Model loading took 12.89 GiB memory and 2.708411 seconds
35
+ (EngineCore pid=114758) WARNING 07-17 07:15:30 [fused_moe.py:1106] Using default MoE config. Performance might be sub-optimal! Config file not found at /home/henry/Documents/PythonProjects/variable-reap/vllm-plugin/.venv25/lib/python3.12/site-packages/vllm/model_executor/layers/fused_moe/configs/E=64,N=1024,device_name=NVIDIA_GeForce_RTX_3090.json
36
+ (EngineCore pid=114758) INFO 07-17 07:15:31 [gpu_worker.py:538] Available KV cache memory: 6.73 GiB
37
+ (EngineCore pid=114758) INFO 07-17 07:15:31 [kv_cache_utils.py:2146] GPU KV cache size: 55,104 tokens
38
+ (EngineCore pid=114758) INFO 07-17 07:15:31 [kv_cache_utils.py:2147] Maximum concurrency for 2,048 tokens per request: 26.91x
39
+ (EngineCore pid=114758) INFO 07-17 07:15:31 [cutedsl_warmup.py:97] Skipping CuTeDSL warmup because no compile units were requested.
40
+ (EngineCore pid=114758) INFO 07-17 07:15:31 [jit_monitor.py:73] Kernel JIT monitor activated; monitored JIT compilations during inference will use mode=warn.
41
+ (EngineCore pid=114758) INFO 07-17 07:15:31 [core.py:344] init engine (profile, create kv cache, warmup model) took 1.97 s
42
+ (EngineCore pid=114758) INFO 07-17 07:15:32 [vllm.py:1042] Asynchronous scheduling is enabled.
43
+ (EngineCore pid=114758) WARNING 07-17 07:15:32 [vllm.py:1096] Enforce eager set, disabling torch.compile and CUDAGraphs. This is equivalent to setting -cc.mode=none -cc.cudagraph_mode=none
44
+ (EngineCore pid=114758) WARNING 07-17 07:15:32 [vllm.py:1144] Inductor compilation was disabled by user settings, optimizations settings that are only active during inductor compilation will be ignored.
45
+ (EngineCore pid=114758) INFO 07-17 07:15:32 [kernel.py:292] Final IR op priority after setting platform defaults: IrOpPriorityConfig(rms_norm=['vllm_c', 'native'], fused_add_rms_norm=['vllm_c', 'native'])
46
+ (EngineCore pid=114758) INFO 07-17 07:15:32 [vllm.py:1322] Cudagraph is disabled under eager mode
47
+ (EngineCore pid=114758) INFO 07-17 07:15:32 [compilation.py:312] Enabled custom fusions: norm_quant, act_quant
48
+ (APIServer pid=114640) INFO 07-17 07:15:32 [api_server.py:612] Supported tasks: ['generate']
49
+ (APIServer pid=114640) WARNING 07-17 07:15:32 [__init__.py:36] SECURITY WARNING: Development endpoints are enabled! This should NOT be used in production!
50
+ (APIServer pid=114640) INFO 07-17 07:15:34 [hf.py:548] Detected the chat template content format to be 'string'. You can set `--chat-template-content-format` to override this.
51
+ (APIServer pid=114640) INFO 07-17 07:15:34 [api_server.py:616] Starting vLLM server on http://127.0.0.1:8385
52
+ (APIServer pid=114640) INFO 07-17 07:15:34 [launcher.py:37] Available routes are:
53
+ (APIServer pid=114640) INFO 07-17 07:15:34 [launcher.py:46] Route: /openapi.json, Methods: HEAD, GET
54
+ (APIServer pid=114640) INFO 07-17 07:15:34 [launcher.py:46] Route: /docs, Methods: HEAD, GET
55
+ (APIServer pid=114640) INFO 07-17 07:15:34 [launcher.py:46] Route: /docs/oauth2-redirect, Methods: HEAD, GET
56
+ (APIServer pid=114640) INFO 07-17 07:15:34 [launcher.py:46] Route: /redoc, Methods: HEAD, GET
57
+ (APIServer pid=114640) INFO 07-17 07:15:34 [launcher.py:46] Route: /load, Methods: GET
58
+ (APIServer pid=114640) INFO 07-17 07:15:34 [launcher.py:46] Route: /version, Methods: GET
59
+ (APIServer pid=114640) INFO 07-17 07:15:34 [launcher.py:46] Route: /health, Methods: GET
60
+ (APIServer pid=114640) INFO 07-17 07:15:34 [launcher.py:46] Route: /metrics, Methods: GET
61
+ (APIServer pid=114640) INFO 07-17 07:15:34 [launcher.py:46] Route: /tokenize, Methods: POST
62
+ (APIServer pid=114640) INFO 07-17 07:15:34 [launcher.py:46] Route: /detokenize, Methods: POST
63
+ (APIServer pid=114640) INFO 07-17 07:15:34 [launcher.py:46] Route: /v1/models, Methods: GET
64
+ (APIServer pid=114640) INFO 07-17 07:15:34 [launcher.py:46] Route: /ping, Methods: GET
65
+ (APIServer pid=114640) INFO 07-17 07:15:34 [launcher.py:46] Route: /ping, Methods: POST
66
+ (APIServer pid=114640) INFO 07-17 07:15:34 [launcher.py:46] Route: /invocations, Methods: POST
67
+ (APIServer pid=114640) INFO 07-17 07:15:34 [launcher.py:46] Route: /reset_prefix_cache, Methods: POST
68
+ (APIServer pid=114640) INFO 07-17 07:15:34 [launcher.py:46] Route: /reset_mm_cache, Methods: POST
69
+ (APIServer pid=114640) INFO 07-17 07:15:34 [launcher.py:46] Route: /reset_encoder_cache, Methods: POST
70
+ (APIServer pid=114640) INFO 07-17 07:15:34 [launcher.py:46] Route: /pause, Methods: POST
71
+ (APIServer pid=114640) INFO 07-17 07:15:34 [launcher.py:46] Route: /resume, Methods: POST
72
+ (APIServer pid=114640) INFO 07-17 07:15:34 [launcher.py:46] Route: /is_paused, Methods: GET
73
+ (APIServer pid=114640) INFO 07-17 07:15:34 [launcher.py:46] Route: /init_weight_transfer_engine, Methods: POST
74
+ (APIServer pid=114640) INFO 07-17 07:15:34 [launcher.py:46] Route: /start_weight_update, Methods: POST
75
+ (APIServer pid=114640) INFO 07-17 07:15:34 [launcher.py:46] Route: /update_weights, Methods: POST
76
+ (APIServer pid=114640) INFO 07-17 07:15:34 [launcher.py:46] Route: /finish_weight_update, Methods: POST
77
+ (APIServer pid=114640) INFO 07-17 07:15:34 [launcher.py:46] Route: /get_world_size, Methods: GET
78
+ (APIServer pid=114640) INFO 07-17 07:15:34 [launcher.py:46] Route: /collective_rpc, Methods: POST
79
+ (APIServer pid=114640) INFO 07-17 07:15:34 [launcher.py:46] Route: /server_info, Methods: GET
80
+ (APIServer pid=114640) INFO 07-17 07:15:34 [launcher.py:46] Route: /sleep, Methods: POST
81
+ (APIServer pid=114640) INFO 07-17 07:15:34 [launcher.py:46] Route: /wake_up, Methods: POST
82
+ (APIServer pid=114640) INFO 07-17 07:15:34 [launcher.py:46] Route: /is_sleeping, Methods: GET
83
+ (APIServer pid=114640) INFO 07-17 07:15:34 [launcher.py:46] Route: /v1/chat/completions, Methods: POST
84
+ (APIServer pid=114640) INFO 07-17 07:15:34 [launcher.py:46] Route: /v1/chat/completions/batch, Methods: POST
85
+ (APIServer pid=114640) INFO 07-17 07:15:34 [launcher.py:46] Route: /v1/responses, Methods: POST
86
+ (APIServer pid=114640) INFO 07-17 07:15:34 [launcher.py:46] Route: /v1/responses/{response_id}, Methods: GET
87
+ (APIServer pid=114640) INFO 07-17 07:15:34 [launcher.py:46] Route: /v1/responses/{response_id}/cancel, Methods: POST
88
+ (APIServer pid=114640) INFO 07-17 07:15:34 [launcher.py:46] Route: /v1/completions, Methods: POST
89
+ (APIServer pid=114640) INFO 07-17 07:15:34 [launcher.py:46] Route: /v1/messages, Methods: POST
90
+ (APIServer pid=114640) INFO 07-17 07:15:34 [launcher.py:46] Route: /v1/messages/count_tokens, Methods: POST
91
+ (APIServer pid=114640) INFO 07-17 07:15:34 [launcher.py:46] Route: /generative_scoring, Methods: POST
92
+ (APIServer pid=114640) INFO 07-17 07:15:34 [launcher.py:46] Route: /scale_elastic_ep, Methods: POST
93
+ (APIServer pid=114640) INFO 07-17 07:15:34 [launcher.py:46] Route: /is_scaling_elastic_ep, Methods: POST
94
+ (APIServer pid=114640) INFO 07-17 07:15:34 [launcher.py:46] Route: /v1/chat/completions/render, Methods: POST
95
+ (APIServer pid=114640) INFO 07-17 07:15:34 [launcher.py:46] Route: /v1/completions/render, Methods: POST
96
+ (APIServer pid=114640) INFO 07-17 07:15:34 [launcher.py:46] Route: /v1/chat/completions/derender, Methods: POST
97
+ (APIServer pid=114640) INFO 07-17 07:15:34 [launcher.py:46] Route: /v1/completions/derender, Methods: POST
98
+ (APIServer pid=114640) INFO 07-17 07:15:34 [launcher.py:46] Route: /inference/v1/generate, Methods: POST
99
+ (APIServer pid=114640) INFO: Started server process [114640]
100
+ (APIServer pid=114640) INFO: Waiting for application startup.
101
+ (APIServer pid=114640) INFO: Application startup complete.
102
+ (APIServer pid=114640) INFO: 127.0.0.1:34120 - "GET /health HTTP/1.1" 200 OK
103
+ (EngineCore pid=114758) WARNING 07-17 07:15:35 [jit_monitor.py:129] Triton kernel JIT compilation during inference: fused_moe_kernel. This causes a latency spike; consider extending warmup to cover this shape/config.
104
+ (APIServer pid=114640) INFO 07-17 07:15:45 [loggers.py:273] Engine 000: Avg prompt throughput: 1275.7 tokens/s, Avg generation throughput: 3910.1 tokens/s, Running: 217 reqs, Waiting: 7921 reqs, GPU KV cache usage: 100.0%, Prefix cache hit rate: 71.2%
105
+ (APIServer pid=114640) INFO 07-17 07:15:55 [loggers.py:273] Engine 000: Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 3688.1 tokens/s, Running: 108 reqs, Waiting: 7956 reqs, GPU KV cache usage: 98.9%, Prefix cache hit rate: 71.2%
106
+ (APIServer pid=114640) INFO 07-17 07:16:05 [loggers.py:273] Engine 000: Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 2335.5 tokens/s, Running: 119 reqs, Waiting: 7881 reqs, GPU KV cache usage: 100.0%, Prefix cache hit rate: 71.2%
107
+ (APIServer pid=114640) INFO 07-17 07:16:15 [loggers.py:273] Engine 000: Avg prompt throughput: 1507.4 tokens/s, Avg generation throughput: 4544.8 tokens/s, Running: 245 reqs, Waiting: 7610 reqs, GPU KV cache usage: 100.0%, Prefix cache hit rate: 64.7%
108
+ (APIServer pid=114640) INFO 07-17 07:16:25 [loggers.py:273] Engine 000: Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 3832.9 tokens/s, Running: 115 reqs, Waiting: 7654 reqs, GPU KV cache usage: 99.2%, Prefix cache hit rate: 64.7%
109
+ (APIServer pid=114640) INFO 07-17 07:16:35 [loggers.py:273] Engine 000: Avg prompt throughput: 591.7 tokens/s, Avg generation throughput: 3009.1 tokens/s, Running: 167 reqs, Waiting: 7512 reqs, GPU KV cache usage: 98.8%, Prefix cache hit rate: 59.4%
110
+ (APIServer pid=114640) INFO 07-17 07:16:45 [loggers.py:273] Engine 000: Avg prompt throughput: 534.0 tokens/s, Avg generation throughput: 3807.8 tokens/s, Running: 211 reqs, Waiting: 7394 reqs, GPU KV cache usage: 99.9%, Prefix cache hit rate: 60.1%
111
+ (APIServer pid=114640) INFO 07-17 07:16:55 [loggers.py:273] Engine 000: Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 3239.3 tokens/s, Running: 98 reqs, Waiting: 7448 reqs, GPU KV cache usage: 99.5%, Prefix cache hit rate: 60.1%
112
+ (APIServer pid=114640) INFO 07-17 07:17:05 [loggers.py:273] Engine 000: Avg prompt throughput: 953.8 tokens/s, Avg generation throughput: 3089.5 tokens/s, Running: 256 reqs, Waiting: 7187 reqs, GPU KV cache usage: 96.5%, Prefix cache hit rate: 56.9%
113
+ (APIServer pid=114640) INFO 07-17 07:17:15 [loggers.py:273] Engine 000: Avg prompt throughput: 282.1 tokens/s, Avg generation throughput: 4527.6 tokens/s, Running: 135 reqs, Waiting: 7214 reqs, GPU KV cache usage: 100.0%, Prefix cache hit rate: 54.5%
114
+ (APIServer pid=114640) INFO 07-17 07:17:25 [loggers.py:273] Engine 000: Avg prompt throughput: 37.9 tokens/s, Avg generation throughput: 2899.9 tokens/s, Running: 154 reqs, Waiting: 7112 reqs, GPU KV cache usage: 99.9%, Prefix cache hit rate: 53.9%
115
+ (APIServer pid=114640) INFO 07-17 07:17:35 [loggers.py:273] Engine 000: Avg prompt throughput: 746.4 tokens/s, Avg generation throughput: 3528.4 tokens/s, Running: 142 reqs, Waiting: 7028 reqs, GPU KV cache usage: 99.6%, Prefix cache hit rate: 54.4%
116
+ (APIServer pid=114640) INFO 07-17 07:17:45 [loggers.py:273] Engine 000: Avg prompt throughput: 594.7 tokens/s, Avg generation throughput: 3840.1 tokens/s, Running: 137 reqs, Waiting: 6933 reqs, GPU KV cache usage: 100.0%, Prefix cache hit rate: 53.4%
117
+ (APIServer pid=114640) INFO 07-17 07:17:55 [loggers.py:273] Engine 000: Avg prompt throughput: 233.3 tokens/s, Avg generation throughput: 3001.7 tokens/s, Running: 143 reqs, Waiting: 6844 reqs, GPU KV cache usage: 99.8%, Prefix cache hit rate: 52.7%
118
+ (APIServer pid=114640) INFO 07-17 07:18:05 [loggers.py:273] Engine 000: Avg prompt throughput: 488.8 tokens/s, Avg generation throughput: 3381.1 tokens/s, Running: 136 reqs, Waiting: 6760 reqs, GPU KV cache usage: 99.5%, Prefix cache hit rate: 51.0%
119
+ (APIServer pid=114640) INFO 07-17 07:18:15 [loggers.py:273] Engine 000: Avg prompt throughput: 619.5 tokens/s, Avg generation throughput: 3265.0 tokens/s, Running: 121 reqs, Waiting: 6698 reqs, GPU KV cache usage: 99.8%, Prefix cache hit rate: 52.6%
120
+ (APIServer pid=114640) INFO 07-17 07:18:25 [loggers.py:273] Engine 000: Avg prompt throughput: 487.6 tokens/s, Avg generation throughput: 3126.9 tokens/s, Running: 163 reqs, Waiting: 6576 reqs, GPU KV cache usage: 99.8%, Prefix cache hit rate: 50.5%
121
+ (APIServer pid=114640) INFO 07-17 07:18:35 [loggers.py:273] Engine 000: Avg prompt throughput: 583.6 tokens/s, Avg generation throughput: 3508.0 tokens/s, Running: 159 reqs, Waiting: 6471 reqs, GPU KV cache usage: 99.9%, Prefix cache hit rate: 52.7%
122
+ (APIServer pid=114640) INFO 07-17 07:18:45 [loggers.py:273] Engine 000: Avg prompt throughput: 512.7 tokens/s, Avg generation throughput: 3249.9 tokens/s, Running: 221 reqs, Waiting: 6335 reqs, GPU KV cache usage: 99.9%, Prefix cache hit rate: 55.3%
123
+ (APIServer pid=114640) INFO 07-17 07:18:55 [loggers.py:273] Engine 000: Avg prompt throughput: 79.6 tokens/s, Avg generation throughput: 3715.4 tokens/s, Running: 120 reqs, Waiting: 6354 reqs, GPU KV cache usage: 99.9%, Prefix cache hit rate: 55.3%
124
+ (APIServer pid=114640) INFO 07-17 07:19:05 [loggers.py:273] Engine 000: Avg prompt throughput: 595.6 tokens/s, Avg generation throughput: 2819.9 tokens/s, Running: 187 reqs, Waiting: 6216 reqs, GPU KV cache usage: 100.0%, Prefix cache hit rate: 56.8%
125
+ (APIServer pid=114640) INFO 07-17 07:19:15 [loggers.py:273] Engine 000: Avg prompt throughput: 242.2 tokens/s, Avg generation throughput: 3617.6 tokens/s, Running: 182 reqs, Waiting: 6140 reqs, GPU KV cache usage: 97.9%, Prefix cache hit rate: 53.6%
126
+ (APIServer pid=114640) INFO 07-17 07:19:25 [loggers.py:273] Engine 000: Avg prompt throughput: 482.4 tokens/s, Avg generation throughput: 3606.4 tokens/s, Running: 139 reqs, Waiting: 6106 reqs, GPU KV cache usage: 99.9%, Prefix cache hit rate: 53.4%
127
+ (APIServer pid=114640) INFO 07-17 07:19:35 [loggers.py:273] Engine 000: Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 2646.6 tokens/s, Running: 112 reqs, Waiting: 6080 reqs, GPU KV cache usage: 99.8%, Prefix cache hit rate: 53.4%
128
+ (APIServer pid=114640) INFO 07-17 07:19:45 [loggers.py:273] Engine 000: Avg prompt throughput: 941.6 tokens/s, Avg generation throughput: 4357.0 tokens/s, Running: 176 reqs, Waiting: 5897 reqs, GPU KV cache usage: 99.9%, Prefix cache hit rate: 55.1%
129
+ (APIServer pid=114640) INFO 07-17 07:19:55 [loggers.py:273] Engine 000: Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 3357.2 tokens/s, Running: 114 reqs, Waiting: 5888 reqs, GPU KV cache usage: 99.0%, Prefix cache hit rate: 55.1%
130
+ (APIServer pid=114640) INFO 07-17 07:20:05 [loggers.py:273] Engine 000: Avg prompt throughput: 417.1 tokens/s, Avg generation throughput: 3118.7 tokens/s, Running: 154 reqs, Waiting: 5768 reqs, GPU KV cache usage: 99.9%, Prefix cache hit rate: 56.9%
131
+ (APIServer pid=114640) INFO 07-17 07:20:15 [loggers.py:273] Engine 000: Avg prompt throughput: 918.0 tokens/s, Avg generation throughput: 4118.9 tokens/s, Running: 249 reqs, Waiting: 5560 reqs, GPU KV cache usage: 99.2%, Prefix cache hit rate: 60.5%
132
+ (APIServer pid=114640) INFO 07-17 07:20:25 [loggers.py:273] Engine 000: Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 3615.5 tokens/s, Running: 110 reqs, Waiting: 5639 reqs, GPU KV cache usage: 100.0%, Prefix cache hit rate: 60.5%
133
+ (APIServer pid=114640) INFO 07-17 07:20:35 [loggers.py:273] Engine 000: Avg prompt throughput: 476.6 tokens/s, Avg generation throughput: 2807.2 tokens/s, Running: 167 reqs, Waiting: 5496 reqs, GPU KV cache usage: 100.0%, Prefix cache hit rate: 60.5%
134
+ (APIServer pid=114640) INFO 07-17 07:20:45 [loggers.py:273] Engine 000: Avg prompt throughput: 398.8 tokens/s, Avg generation throughput: 3556.8 tokens/s, Running: 136 reqs, Waiting: 5449 reqs, GPU KV cache usage: 99.9%, Prefix cache hit rate: 60.1%
135
+ (APIServer pid=114640) INFO 07-17 07:20:55 [loggers.py:273] Engine 000: Avg prompt throughput: 268.0 tokens/s, Avg generation throughput: 3118.2 tokens/s, Running: 101 reqs, Waiting: 5414 reqs, GPU KV cache usage: 99.8%, Prefix cache hit rate: 60.1%
136
+ (APIServer pid=114640) INFO 07-17 07:21:05 [loggers.py:273] Engine 000: Avg prompt throughput: 415.6 tokens/s, Avg generation throughput: 3015.9 tokens/s, Running: 137 reqs, Waiting: 5304 reqs, GPU KV cache usage: 99.9%, Prefix cache hit rate: 60.5%
137
+ (APIServer pid=114640) INFO 07-17 07:21:15 [loggers.py:273] Engine 000: Avg prompt throughput: 597.2 tokens/s, Avg generation throughput: 3774.0 tokens/s, Running: 151 reqs, Waiting: 5187 reqs, GPU KV cache usage: 99.9%, Prefix cache hit rate: 63.5%
138
+ (APIServer pid=114640) INFO 07-17 07:21:25 [loggers.py:273] Engine 000: Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 2962.1 tokens/s, Running: 111 reqs, Waiting: 5165 reqs, GPU KV cache usage: 99.9%, Prefix cache hit rate: 63.5%
139
+ (APIServer pid=114640) INFO 07-17 07:21:35 [loggers.py:273] Engine 000: Avg prompt throughput: 733.1 tokens/s, Avg generation throughput: 2981.7 tokens/s, Running: 220 reqs, Waiting: 4980 reqs, GPU KV cache usage: 99.8%, Prefix cache hit rate: 62.7%
140
+ (APIServer pid=114640) INFO 07-17 07:21:45 [loggers.py:273] Engine 000: Avg prompt throughput: 436.0 tokens/s, Avg generation throughput: 4018.3 tokens/s, Running: 149 reqs, Waiting: 4981 reqs, GPU KV cache usage: 99.6%, Prefix cache hit rate: 62.7%
141
+ (APIServer pid=114640) INFO 07-17 07:21:55 [loggers.py:273] Engine 000: Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 2802.2 tokens/s, Running: 95 reqs, Waiting: 4983 reqs, GPU KV cache usage: 99.7%, Prefix cache hit rate: 62.7%
142
+ (APIServer pid=114640) INFO 07-17 07:22:05 [loggers.py:273] Engine 000: Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 2363.5 tokens/s, Running: 130 reqs, Waiting: 4900 reqs, GPU KV cache usage: 99.6%, Prefix cache hit rate: 62.7%
143
+ (APIServer pid=114640) INFO 07-17 07:22:15 [loggers.py:273] Engine 000: Avg prompt throughput: 1068.0 tokens/s, Avg generation throughput: 4933.2 tokens/s, Running: 168 reqs, Waiting: 4732 reqs, GPU KV cache usage: 100.0%, Prefix cache hit rate: 63.4%
144
+ (APIServer pid=114640) INFO 07-17 07:22:25 [loggers.py:273] Engine 000: Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 2989.7 tokens/s, Running: 106 reqs, Waiting: 4734 reqs, GPU KV cache usage: 99.9%, Prefix cache hit rate: 63.4%
145
+ (APIServer pid=114640) INFO 07-17 07:22:35 [loggers.py:273] Engine 000: Avg prompt throughput: 166.4 tokens/s, Avg generation throughput: 2422.3 tokens/s, Running: 123 reqs, Waiting: 4660 reqs, GPU KV cache usage: 99.6%, Prefix cache hit rate: 62.2%
146
+ (APIServer pid=114640) INFO 07-17 07:22:45 [loggers.py:273] Engine 000: Avg prompt throughput: 1179.2 tokens/s, Avg generation throughput: 4664.0 tokens/s, Running: 215 reqs, Waiting: 4454 reqs, GPU KV cache usage: 99.4%, Prefix cache hit rate: 62.0%
147
+ (APIServer pid=114640) INFO 07-17 07:22:55 [loggers.py:273] Engine 000: Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 3607.0 tokens/s, Running: 106 reqs, Waiting: 4504 reqs, GPU KV cache usage: 99.0%, Prefix cache hit rate: 62.0%
148
+ (APIServer pid=114640) INFO 07-17 07:23:05 [loggers.py:273] Engine 000: Avg prompt throughput: 288.8 tokens/s, Avg generation throughput: 2537.3 tokens/s, Running: 142 reqs, Waiting: 4388 reqs, GPU KV cache usage: 99.9%, Prefix cache hit rate: 64.2%
149
+ (APIServer pid=114640) INFO 07-17 07:23:15 [loggers.py:273] Engine 000: Avg prompt throughput: 867.6 tokens/s, Avg generation throughput: 3923.2 tokens/s, Running: 233 reqs, Waiting: 4195 reqs, GPU KV cache usage: 99.5%, Prefix cache hit rate: 63.7%
150
+ (APIServer pid=114640) INFO 07-17 07:23:25 [loggers.py:273] Engine 000: Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 3591.1 tokens/s, Running: 129 reqs, Waiting: 4232 reqs, GPU KV cache usage: 99.9%, Prefix cache hit rate: 63.7%
151
+ (APIServer pid=114640) INFO 07-17 07:23:35 [loggers.py:273] Engine 000: Avg prompt throughput: 199.7 tokens/s, Avg generation throughput: 3147.1 tokens/s, Running: 116 reqs, Waiting: 4165 reqs, GPU KV cache usage: 99.1%, Prefix cache hit rate: 59.8%
152
+ (APIServer pid=114640) INFO 07-17 07:23:45 [loggers.py:273] Engine 000: Avg prompt throughput: 835.1 tokens/s, Avg generation throughput: 3413.0 tokens/s, Running: 180 reqs, Waiting: 4000 reqs, GPU KV cache usage: 99.5%, Prefix cache hit rate: 61.2%
153
+ (APIServer pid=114640) INFO 07-17 07:23:55 [loggers.py:273] Engine 000: Avg prompt throughput: 405.6 tokens/s, Avg generation throughput: 3850.2 tokens/s, Running: 136 reqs, Waiting: 3964 reqs, GPU KV cache usage: 98.3%, Prefix cache hit rate: 60.7%
154
+ (APIServer pid=114640) INFO 07-17 07:24:05 [loggers.py:273] Engine 000: Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 2786.0 tokens/s, Running: 97 reqs, Waiting: 3953 reqs, GPU KV cache usage: 99.8%, Prefix cache hit rate: 60.7%
155
+ (APIServer pid=114640) INFO 07-17 07:24:15 [loggers.py:273] Engine 000: Avg prompt throughput: 903.9 tokens/s, Avg generation throughput: 2835.5 tokens/s, Running: 218 reqs, Waiting: 3760 reqs, GPU KV cache usage: 99.8%, Prefix cache hit rate: 59.0%
156
+ (APIServer pid=114640) INFO 07-17 07:24:25 [loggers.py:273] Engine 000: Avg prompt throughput: 316.4 tokens/s, Avg generation throughput: 4077.5 tokens/s, Running: 145 reqs, Waiting: 3758 reqs, GPU KV cache usage: 99.5%, Prefix cache hit rate: 59.1%
157
+ (APIServer pid=114640) INFO 07-17 07:24:35 [loggers.py:273] Engine 000: Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 2766.5 tokens/s, Running: 115 reqs, Waiting: 3721 reqs, GPU KV cache usage: 99.2%, Prefix cache hit rate: 59.1%
158
+ (APIServer pid=114640) INFO 07-17 07:24:45 [loggers.py:273] Engine 000: Avg prompt throughput: 281.3 tokens/s, Avg generation throughput: 2714.4 tokens/s, Running: 189 reqs, Waiting: 3576 reqs, GPU KV cache usage: 99.7%, Prefix cache hit rate: 59.2%
159
+ (APIServer pid=114640) INFO 07-17 07:24:55 [loggers.py:273] Engine 000: Avg prompt throughput: 928.8 tokens/s, Avg generation throughput: 4746.0 tokens/s, Running: 184 reqs, Waiting: 3456 reqs, GPU KV cache usage: 99.6%, Prefix cache hit rate: 59.6%
160
+ (APIServer pid=114640) INFO 07-17 07:25:05 [loggers.py:273] Engine 000: Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 3150.9 tokens/s, Running: 128 reqs, Waiting: 3433 reqs, GPU KV cache usage: 99.9%, Prefix cache hit rate: 59.6%
161
+ (APIServer pid=114640) INFO 07-17 07:25:15 [loggers.py:273] Engine 000: Avg prompt throughput: 686.9 tokens/s, Avg generation throughput: 2787.5 tokens/s, Running: 191 reqs, Waiting: 3288 reqs, GPU KV cache usage: 99.7%, Prefix cache hit rate: 57.9%
162
+ (APIServer pid=114640) INFO 07-17 07:25:25 [loggers.py:273] Engine 000: Avg prompt throughput: 579.2 tokens/s, Avg generation throughput: 4740.4 tokens/s, Running: 219 reqs, Waiting: 3147 reqs, GPU KV cache usage: 100.0%, Prefix cache hit rate: 61.5%
163
+ (APIServer pid=114640) INFO 07-17 07:25:35 [loggers.py:273] Engine 000: Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 3331.9 tokens/s, Running: 110 reqs, Waiting: 3202 reqs, GPU KV cache usage: 99.4%, Prefix cache hit rate: 61.5%
164
+ (APIServer pid=114640) INFO 07-17 07:25:45 [loggers.py:273] Engine 000: Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 2407.5 tokens/s, Running: 92 reqs, Waiting: 3170 reqs, GPU KV cache usage: 98.8%, Prefix cache hit rate: 61.5%
165
+ (APIServer pid=114640) INFO 07-17 07:25:55 [loggers.py:273] Engine 000: Avg prompt throughput: 1099.5 tokens/s, Avg generation throughput: 4171.3 tokens/s, Running: 213 reqs, Waiting: 2952 reqs, GPU KV cache usage: 97.8%, Prefix cache hit rate: 63.6%
166
+ (APIServer pid=114640) INFO 07-17 07:26:05 [loggers.py:273] Engine 000: Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 3499.0 tokens/s, Running: 116 reqs, Waiting: 2998 reqs, GPU KV cache usage: 99.4%, Prefix cache hit rate: 63.6%
167
+ (APIServer pid=114640) INFO 07-17 07:26:15 [loggers.py:273] Engine 000: Avg prompt throughput: 116.8 tokens/s, Avg generation throughput: 2488.9 tokens/s, Running: 143 reqs, Waiting: 2894 reqs, GPU KV cache usage: 100.0%, Prefix cache hit rate: 62.8%
168
+ (APIServer pid=114640) INFO 07-17 07:26:25 [loggers.py:273] Engine 000: Avg prompt throughput: 999.2 tokens/s, Avg generation throughput: 3831.9 tokens/s, Running: 254 reqs, Waiting: 2667 reqs, GPU KV cache usage: 90.7%, Prefix cache hit rate: 63.5%
169
+ (APIServer pid=114640) INFO 07-17 07:26:35 [loggers.py:273] Engine 000: Avg prompt throughput: 169.6 tokens/s, Avg generation throughput: 4115.8 tokens/s, Running: 120 reqs, Waiting: 2714 reqs, GPU KV cache usage: 99.8%, Prefix cache hit rate: 63.8%
170
+ (APIServer pid=114640) INFO 07-17 07:26:45 [loggers.py:273] Engine 000: Avg prompt throughput: 249.8 tokens/s, Avg generation throughput: 2590.2 tokens/s, Running: 160 reqs, Waiting: 2588 reqs, GPU KV cache usage: 100.0%, Prefix cache hit rate: 63.3%
171
+ (APIServer pid=114640) INFO 07-17 07:26:55 [loggers.py:273] Engine 000: Avg prompt throughput: 1049.3 tokens/s, Avg generation throughput: 3741.0 tokens/s, Running: 211 reqs, Waiting: 2440 reqs, GPU KV cache usage: 100.0%, Prefix cache hit rate: 63.3%
172
+ (APIServer pid=114640) INFO 07-17 07:27:05 [loggers.py:273] Engine 000: Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 3279.2 tokens/s, Running: 109 reqs, Waiting: 2480 reqs, GPU KV cache usage: 99.8%, Prefix cache hit rate: 63.3%
173
+ (APIServer pid=114640) INFO 07-17 07:27:15 [loggers.py:273] Engine 000: Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 2394.5 tokens/s, Running: 102 reqs, Waiting: 2436 reqs, GPU KV cache usage: 99.7%, Prefix cache hit rate: 63.3%
174
+ (APIServer pid=114640) INFO 07-17 07:27:25 [loggers.py:273] Engine 000: Avg prompt throughput: 1214.8 tokens/s, Avg generation throughput: 3507.5 tokens/s, Running: 222 reqs, Waiting: 2216 reqs, GPU KV cache usage: 99.8%, Prefix cache hit rate: 63.3%
175
+ (APIServer pid=114640) INFO 07-17 07:27:35 [loggers.py:273] Engine 000: Avg prompt throughput: 292.0 tokens/s, Avg generation throughput: 4258.7 tokens/s, Running: 143 reqs, Waiting: 2205 reqs, GPU KV cache usage: 98.9%, Prefix cache hit rate: 64.8%
176
+ (APIServer pid=114640) INFO 07-17 07:27:45 [loggers.py:273] Engine 000: Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 2717.3 tokens/s, Running: 124 reqs, Waiting: 2148 reqs, GPU KV cache usage: 99.8%, Prefix cache hit rate: 64.8%
177
+ (APIServer pid=114640) INFO 07-17 07:27:55 [loggers.py:273] Engine 000: Avg prompt throughput: 582.8 tokens/s, Avg generation throughput: 2965.9 tokens/s, Running: 158 reqs, Waiting: 2040 reqs, GPU KV cache usage: 99.7%, Prefix cache hit rate: 63.0%
178
+ (APIServer pid=114640) INFO 07-17 07:28:05 [loggers.py:273] Engine 000: Avg prompt throughput: 620.1 tokens/s, Avg generation throughput: 4109.2 tokens/s, Running: 178 reqs, Waiting: 1929 reqs, GPU KV cache usage: 99.9%, Prefix cache hit rate: 64.2%
179
+ (APIServer pid=114640) INFO 07-17 07:28:15 [loggers.py:273] Engine 000: Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 3079.7 tokens/s, Running: 114 reqs, Waiting: 1931 reqs, GPU KV cache usage: 99.9%, Prefix cache hit rate: 64.2%
180
+ (APIServer pid=114640) INFO 07-17 07:28:25 [loggers.py:273] Engine 000: Avg prompt throughput: 547.3 tokens/s, Avg generation throughput: 3258.1 tokens/s, Running: 152 reqs, Waiting: 1805 reqs, GPU KV cache usage: 99.9%, Prefix cache hit rate: 61.9%
181
+ (APIServer pid=114640) INFO 07-17 07:28:35 [loggers.py:273] Engine 000: Avg prompt throughput: 501.2 tokens/s, Avg generation throughput: 3917.4 tokens/s, Running: 174 reqs, Waiting: 1699 reqs, GPU KV cache usage: 99.2%, Prefix cache hit rate: 63.8%
182
+ (APIServer pid=114640) INFO 07-17 07:28:45 [loggers.py:273] Engine 000: Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 3086.8 tokens/s, Running: 110 reqs, Waiting: 1699 reqs, GPU KV cache usage: 99.9%, Prefix cache hit rate: 63.8%
183
+ (APIServer pid=114640) INFO 07-17 07:28:55 [loggers.py:273] Engine 000: Avg prompt throughput: 603.5 tokens/s, Avg generation throughput: 2933.5 tokens/s, Running: 180 reqs, Waiting: 1555 reqs, GPU KV cache usage: 100.0%, Prefix cache hit rate: 64.0%
184
+ (APIServer pid=114640) INFO 07-17 07:29:05 [loggers.py:273] Engine 000: Avg prompt throughput: 157.6 tokens/s, Avg generation throughput: 3746.9 tokens/s, Running: 175 reqs, Waiting: 1479 reqs, GPU KV cache usage: 99.9%, Prefix cache hit rate: 64.3%
185
+ (APIServer pid=114640) INFO 07-17 07:29:15 [loggers.py:273] Engine 000: Avg prompt throughput: 366.8 tokens/s, Avg generation throughput: 3153.3 tokens/s, Running: 156 reqs, Waiting: 1416 reqs, GPU KV cache usage: 99.7%, Prefix cache hit rate: 58.5%
186
+ (APIServer pid=114640) INFO 07-17 07:29:25 [loggers.py:273] Engine 000: Avg prompt throughput: 268.4 tokens/s, Avg generation throughput: 3149.1 tokens/s, Running: 130 reqs, Waiting: 1376 reqs, GPU KV cache usage: 99.7%, Prefix cache hit rate: 57.7%
187
+ (APIServer pid=114640) INFO 07-17 07:29:35 [loggers.py:273] Engine 000: Avg prompt throughput: 872.6 tokens/s, Avg generation throughput: 4096.3 tokens/s, Running: 179 reqs, Waiting: 1212 reqs, GPU KV cache usage: 100.0%, Prefix cache hit rate: 58.6%
188
+ (APIServer pid=114640) INFO 07-17 07:29:45 [loggers.py:273] Engine 000: Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 3091.5 tokens/s, Running: 123 reqs, Waiting: 1192 reqs, GPU KV cache usage: 99.9%, Prefix cache hit rate: 58.6%
189
+ (APIServer pid=114640) INFO 07-17 07:29:55 [loggers.py:273] Engine 000: Avg prompt throughput: 883.5 tokens/s, Avg generation throughput: 3621.3 tokens/s, Running: 162 reqs, Waiting: 1040 reqs, GPU KV cache usage: 97.2%, Prefix cache hit rate: 56.1%
190
+ (APIServer pid=114640) INFO 07-17 07:30:05 [loggers.py:273] Engine 000: Avg prompt throughput: 36.4 tokens/s, Avg generation throughput: 3044.9 tokens/s, Running: 131 reqs, Waiting: 1000 reqs, GPU KV cache usage: 96.2%, Prefix cache hit rate: 56.2%
191
+ (APIServer pid=114640) INFO 07-17 07:30:15 [loggers.py:273] Engine 000: Avg prompt throughput: 798.0 tokens/s, Avg generation throughput: 3699.6 tokens/s, Running: 123 reqs, Waiting: 911 reqs, GPU KV cache usage: 98.4%, Prefix cache hit rate: 57.7%
192
+ (APIServer pid=114640) INFO 07-17 07:30:25 [loggers.py:273] Engine 000: Avg prompt throughput: 598.9 tokens/s, Avg generation throughput: 2895.0 tokens/s, Running: 224 reqs, Waiting: 728 reqs, GPU KV cache usage: 99.9%, Prefix cache hit rate: 57.3%
193
+ (APIServer pid=114640) INFO 07-17 07:30:35 [loggers.py:273] Engine 000: Avg prompt throughput: 123.8 tokens/s, Avg generation throughput: 4034.0 tokens/s, Running: 139 reqs, Waiting: 718 reqs, GPU KV cache usage: 99.8%, Prefix cache hit rate: 57.5%
194
+ (APIServer pid=114640) INFO 07-17 07:30:45 [loggers.py:273] Engine 000: Avg prompt throughput: 166.8 tokens/s, Avg generation throughput: 2897.7 tokens/s, Running: 116 reqs, Waiting: 676 reqs, GPU KV cache usage: 99.7%, Prefix cache hit rate: 57.1%
195
+ (APIServer pid=114640) INFO 07-17 07:30:55 [loggers.py:273] Engine 000: Avg prompt throughput: 381.2 tokens/s, Avg generation throughput: 3042.8 tokens/s, Running: 186 reqs, Waiting: 538 reqs, GPU KV cache usage: 98.3%, Prefix cache hit rate: 61.2%
196
+ (APIServer pid=114640) INFO 07-17 07:31:05 [loggers.py:273] Engine 000: Avg prompt throughput: 851.9 tokens/s, Avg generation throughput: 4313.0 tokens/s, Running: 178 reqs, Waiting: 448 reqs, GPU KV cache usage: 99.4%, Prefix cache hit rate: 62.3%
197
+ (APIServer pid=114640) INFO 07-17 07:31:15 [loggers.py:273] Engine 000: Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 3098.5 tokens/s, Running: 106 reqs, Waiting: 462 reqs, GPU KV cache usage: 98.9%, Prefix cache hit rate: 62.3%
198
+ (APIServer pid=114640) INFO 07-17 07:31:25 [loggers.py:273] Engine 000: Avg prompt throughput: 218.0 tokens/s, Avg generation throughput: 2617.5 tokens/s, Running: 187 reqs, Waiting: 307 reqs, GPU KV cache usage: 97.6%, Prefix cache hit rate: 59.7%
199
+ (APIServer pid=114640) INFO 07-17 07:31:35 [loggers.py:273] Engine 000: Avg prompt throughput: 846.7 tokens/s, Avg generation throughput: 3900.3 tokens/s, Running: 215 reqs, Waiting: 188 reqs, GPU KV cache usage: 99.2%, Prefix cache hit rate: 60.3%
200
+ (APIServer pid=114640) INFO 07-17 07:31:45 [loggers.py:273] Engine 000: Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 3257.6 tokens/s, Running: 128 reqs, Waiting: 199 reqs, GPU KV cache usage: 99.9%, Prefix cache hit rate: 60.3%
201
+ (APIServer pid=114640) INFO 07-17 07:31:55 [loggers.py:273] Engine 000: Avg prompt throughput: 223.2 tokens/s, Avg generation throughput: 2874.4 tokens/s, Running: 99 reqs, Waiting: 174 reqs, GPU KV cache usage: 99.8%, Prefix cache hit rate: 60.2%
202
+ (APIServer pid=114640) INFO 07-17 07:32:05 [loggers.py:273] Engine 000: Avg prompt throughput: 617.1 tokens/s, Avg generation throughput: 3714.9 tokens/s, Running: 173 reqs, Waiting: 0 reqs, GPU KV cache usage: 89.9%, Prefix cache hit rate: 60.3%
203
+ (APIServer pid=114640) INFO 07-17 07:32:15 [loggers.py:273] Engine 000: Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 3560.9 tokens/s, Running: 111 reqs, Waiting: 0 reqs, GPU KV cache usage: 96.3%, Prefix cache hit rate: 60.3%
204
+ (APIServer pid=114640) INFO 07-17 07:32:25 [loggers.py:273] Engine 000: Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 2281.9 tokens/s, Running: 38 reqs, Waiting: 0 reqs, GPU KV cache usage: 54.9%, Prefix cache hit rate: 60.3%
205
+ (APIServer pid=114640) INFO: 127.0.0.1:34130 - "POST /v1/completions HTTP/1.1" 200 OK
206
+ (APIServer pid=114640) INFO 07-17 07:32:35 [loggers.py:273] Engine 000: Avg prompt throughput: 1423.7 tokens/s, Avg generation throughput: 2762.8 tokens/s, Running: 256 reqs, Waiting: 7893 reqs, GPU KV cache usage: 70.7%, Prefix cache hit rate: 61.7%
207
+ (APIServer pid=114640) INFO 07-17 07:32:45 [loggers.py:273] Engine 000: Avg prompt throughput: 82.6 tokens/s, Avg generation throughput: 4682.5 tokens/s, Running: 138 reqs, Waiting: 7961 reqs, GPU KV cache usage: 99.9%, Prefix cache hit rate: 62.4%
208
+ (APIServer pid=114640) INFO 07-17 07:32:55 [loggers.py:273] Engine 000: Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 2606.8 tokens/s, Running: 88 reqs, Waiting: 7955 reqs, GPU KV cache usage: 99.6%, Prefix cache hit rate: 62.4%
209
+ (APIServer pid=114640) INFO 07-17 07:33:05 [loggers.py:273] Engine 000: Avg prompt throughput: 900.2 tokens/s, Avg generation throughput: 2589.9 tokens/s, Running: 256 reqs, Waiting: 7686 reqs, GPU KV cache usage: 83.6%, Prefix cache hit rate: 63.4%
210
+ (APIServer pid=114640) INFO 07-17 07:33:15 [loggers.py:273] Engine 000: Avg prompt throughput: 289.2 tokens/s, Avg generation throughput: 4322.5 tokens/s, Running: 144 reqs, Waiting: 7710 reqs, GPU KV cache usage: 99.9%, Prefix cache hit rate: 63.5%
211
+ (APIServer pid=114640) INFO 07-17 07:33:25 [loggers.py:273] Engine 000: Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 2838.5 tokens/s, Running: 106 reqs, Waiting: 7679 reqs, GPU KV cache usage: 99.0%, Prefix cache hit rate: 63.5%
212
+ (APIServer pid=114640) INFO 07-17 07:33:35 [loggers.py:273] Engine 000: Avg prompt throughput: 625.0 tokens/s, Avg generation throughput: 2942.5 tokens/s, Running: 154 reqs, Waiting: 7556 reqs, GPU KV cache usage: 97.7%, Prefix cache hit rate: 61.5%
213
+ (APIServer pid=114640) INFO 07-17 07:33:45 [loggers.py:273] Engine 000: Avg prompt throughput: 932.3 tokens/s, Avg generation throughput: 4427.9 tokens/s, Running: 187 reqs, Waiting: 7402 reqs, GPU KV cache usage: 99.9%, Prefix cache hit rate: 62.0%
214
+ (APIServer pid=114640) INFO 07-17 07:33:55 [loggers.py:273] Engine 000: Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 3064.1 tokens/s, Running: 96 reqs, Waiting: 7435 reqs, GPU KV cache usage: 99.6%, Prefix cache hit rate: 62.0%
215
+ (APIServer pid=114640) INFO 07-17 07:34:05 [loggers.py:273] Engine 000: Avg prompt throughput: 274.4 tokens/s, Avg generation throughput: 2745.0 tokens/s, Running: 176 reqs, Waiting: 7288 reqs, GPU KV cache usage: 99.9%, Prefix cache hit rate: 61.8%
216
+ (APIServer pid=114640) INFO 07-17 07:34:15 [loggers.py:273] Engine 000: Avg prompt throughput: 940.8 tokens/s, Avg generation throughput: 3966.3 tokens/s, Running: 256 reqs, Waiting: 7103 reqs, GPU KV cache usage: 99.9%, Prefix cache hit rate: 60.3%
217
+ (APIServer pid=114640) INFO 07-17 07:34:25 [loggers.py:273] Engine 000: Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 3596.1 tokens/s, Running: 105 reqs, Waiting: 7202 reqs, GPU KV cache usage: 98.6%, Prefix cache hit rate: 60.3%
218
+ (APIServer pid=114640) INFO 07-17 07:34:35 [loggers.py:273] Engine 000: Avg prompt throughput: 2.8 tokens/s, Avg generation throughput: 2700.2 tokens/s, Running: 117 reqs, Waiting: 7123 reqs, GPU KV cache usage: 99.9%, Prefix cache hit rate: 60.5%
219
+ (APIServer pid=114640) INFO 07-17 07:34:45 [loggers.py:273] Engine 000: Avg prompt throughput: 860.2 tokens/s, Avg generation throughput: 3683.0 tokens/s, Running: 177 reqs, Waiting: 6962 reqs, GPU KV cache usage: 99.9%, Prefix cache hit rate: 58.8%
220
+ (APIServer pid=114640) INFO 07-17 07:34:55 [loggers.py:273] Engine 000: Avg prompt throughput: 243.6 tokens/s, Avg generation throughput: 3623.1 tokens/s, Running: 128 reqs, Waiting: 6924 reqs, GPU KV cache usage: 99.9%, Prefix cache hit rate: 58.3%
221
+ (APIServer pid=114640) INFO 07-17 07:35:05 [loggers.py:273] Engine 000: Avg prompt throughput: 296.4 tokens/s, Avg generation throughput: 3103.7 tokens/s, Running: 156 reqs, Waiting: 6832 reqs, GPU KV cache usage: 99.9%, Prefix cache hit rate: 56.9%
222
+ (APIServer pid=114640) INFO 07-17 07:35:15 [loggers.py:273] Engine 000: Avg prompt throughput: 248.2 tokens/s, Avg generation throughput: 3037.3 tokens/s, Running: 144 reqs, Waiting: 6772 reqs, GPU KV cache usage: 97.2%, Prefix cache hit rate: 56.7%
223
+ (APIServer pid=114640) INFO 07-17 07:35:25 [loggers.py:273] Engine 000: Avg prompt throughput: 536.8 tokens/s, Avg generation throughput: 4252.9 tokens/s, Running: 156 reqs, Waiting: 6668 reqs, GPU KV cache usage: 99.2%, Prefix cache hit rate: 64.2%
224
+ (APIServer pid=114640) INFO 07-17 07:35:35 [loggers.py:273] Engine 000: Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 2873.1 tokens/s, Running: 95 reqs, Waiting: 6675 reqs, GPU KV cache usage: 98.1%, Prefix cache hit rate: 64.2%
225
+ (APIServer pid=114640) INFO 07-17 07:35:45 [loggers.py:273] Engine 000: Avg prompt throughput: 543.7 tokens/s, Avg generation throughput: 2632.3 tokens/s, Running: 176 reqs, Waiting: 6524 reqs, GPU KV cache usage: 99.9%, Prefix cache hit rate: 61.1%
226
+ (APIServer pid=114640) INFO 07-17 07:35:55 [loggers.py:273] Engine 000: Avg prompt throughput: 717.7 tokens/s, Avg generation throughput: 4454.2 tokens/s, Running: 187 reqs, Waiting: 6423 reqs, GPU KV cache usage: 99.9%, Prefix cache hit rate: 61.3%
227
+ (APIServer pid=114640) INFO 07-17 07:36:05 [loggers.py:273] Engine 000: Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 3197.0 tokens/s, Running: 98 reqs, Waiting: 6465 reqs, GPU KV cache usage: 98.8%, Prefix cache hit rate: 61.3%
228
+ (APIServer pid=114640) INFO 07-17 07:36:15 [loggers.py:273] Engine 000: Avg prompt throughput: 165.1 tokens/s, Avg generation throughput: 2319.8 tokens/s, Running: 175 reqs, Waiting: 6325 reqs, GPU KV cache usage: 100.0%, Prefix cache hit rate: 61.3%
229
+ (APIServer pid=114640) INFO 07-17 07:36:25 [loggers.py:273] Engine 000: Avg prompt throughput: 771.1 tokens/s, Avg generation throughput: 4011.8 tokens/s, Running: 199 reqs, Waiting: 6216 reqs, GPU KV cache usage: 99.9%, Prefix cache hit rate: 58.5%
230
+ (APIServer pid=114640) INFO 07-17 07:36:35 [loggers.py:273] Engine 000: Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 3133.8 tokens/s, Running: 100 reqs, Waiting: 6256 reqs, GPU KV cache usage: 99.6%, Prefix cache hit rate: 58.5%
231
+ (APIServer pid=114640) INFO 07-17 07:36:45 [loggers.py:273] Engine 000: Avg prompt throughput: 543.0 tokens/s, Avg generation throughput: 2986.8 tokens/s, Running: 211 reqs, Waiting: 6068 reqs, GPU KV cache usage: 98.1%, Prefix cache hit rate: 58.1%
232
+ (APIServer pid=114640) INFO 07-17 07:36:55 [loggers.py:273] Engine 000: Avg prompt throughput: 208.9 tokens/s, Avg generation throughput: 3864.9 tokens/s, Running: 206 reqs, Waiting: 5996 reqs, GPU KV cache usage: 99.9%, Prefix cache hit rate: 58.1%
233
+ (APIServer pid=114640) INFO 07-17 07:37:05 [loggers.py:273] Engine 000: Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 3222.9 tokens/s, Running: 122 reqs, Waiting: 6009 reqs, GPU KV cache usage: 98.8%, Prefix cache hit rate: 58.1%
234
+ (APIServer pid=114640) INFO 07-17 07:37:15 [loggers.py:273] Engine 000: Avg prompt throughput: 604.3 tokens/s, Avg generation throughput: 3260.4 tokens/s, Running: 161 reqs, Waiting: 5880 reqs, GPU KV cache usage: 99.9%, Prefix cache hit rate: 59.1%
235
+ (APIServer pid=114640) INFO 07-17 07:37:25 [loggers.py:273] Engine 000: Avg prompt throughput: 609.6 tokens/s, Avg generation throughput: 4257.2 tokens/s, Running: 171 reqs, Waiting: 5748 reqs, GPU KV cache usage: 98.4%, Prefix cache hit rate: 59.5%
236
+ (APIServer pid=114640) INFO 07-17 07:37:35 [loggers.py:273] Engine 000: Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 3110.2 tokens/s, Running: 105 reqs, Waiting: 5752 reqs, GPU KV cache usage: 100.0%, Prefix cache hit rate: 59.5%
237
+ (APIServer pid=114640) INFO 07-17 07:37:45 [loggers.py:273] Engine 000: Avg prompt throughput: 745.0 tokens/s, Avg generation throughput: 3044.5 tokens/s, Running: 227 reqs, Waiting: 5540 reqs, GPU KV cache usage: 99.9%, Prefix cache hit rate: 60.6%
238
+ (APIServer pid=114640) INFO 07-17 07:37:55 [loggers.py:273] Engine 000: Avg prompt throughput: 589.8 tokens/s, Avg generation throughput: 4258.3 tokens/s, Running: 209 reqs, Waiting: 5440 reqs, GPU KV cache usage: 99.8%, Prefix cache hit rate: 59.7%
239
+ (APIServer pid=114640) INFO 07-17 07:38:05 [loggers.py:273] Engine 000: Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 3372.4 tokens/s, Running: 132 reqs, Waiting: 5428 reqs, GPU KV cache usage: 99.9%, Prefix cache hit rate: 59.7%
240
+ (APIServer pid=114640) INFO 07-17 07:38:15 [loggers.py:273] Engine 000: Avg prompt throughput: 809.5 tokens/s, Avg generation throughput: 3682.6 tokens/s, Running: 149 reqs, Waiting: 5309 reqs, GPU KV cache usage: 99.9%, Prefix cache hit rate: 58.5%
241
+ (APIServer pid=114640) INFO 07-17 07:38:25 [loggers.py:273] Engine 000: Avg prompt throughput: 364.4 tokens/s, Avg generation throughput: 3714.3 tokens/s, Running: 138 reqs, Waiting: 5232 reqs, GPU KV cache usage: 99.3%, Prefix cache hit rate: 58.8%
242
+ (APIServer pid=114640) INFO 07-17 07:38:35 [loggers.py:273] Engine 000: Avg prompt throughput: 147.2 tokens/s, Avg generation throughput: 3095.5 tokens/s, Running: 155 reqs, Waiting: 5136 reqs, GPU KV cache usage: 95.7%, Prefix cache hit rate: 58.2%
243
+ (APIServer pid=114640) INFO 07-17 07:38:45 [loggers.py:273] Engine 000: Avg prompt throughput: 531.5 tokens/s, Avg generation throughput: 3526.3 tokens/s, Running: 165 reqs, Waiting: 5047 reqs, GPU KV cache usage: 100.0%, Prefix cache hit rate: 57.6%
244
+ (APIServer pid=114640) INFO 07-17 07:38:55 [loggers.py:273] Engine 000: Avg prompt throughput: 433.7 tokens/s, Avg generation throughput: 3795.5 tokens/s, Running: 181 reqs, Waiting: 4948 reqs, GPU KV cache usage: 99.7%, Prefix cache hit rate: 57.9%
245
+ (APIServer pid=114640) INFO 07-17 07:39:05 [loggers.py:273] Engine 000: Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 3294.4 tokens/s, Running: 117 reqs, Waiting: 4941 reqs, GPU KV cache usage: 98.4%, Prefix cache hit rate: 57.9%
246
+ (APIServer pid=114640) INFO 07-17 07:39:15 [loggers.py:273] Engine 000: Avg prompt throughput: 581.5 tokens/s, Avg generation throughput: 2992.8 tokens/s, Running: 170 reqs, Waiting: 4824 reqs, GPU KV cache usage: 99.5%, Prefix cache hit rate: 57.0%
247
+ (APIServer pid=114640) INFO 07-17 07:39:25 [loggers.py:273] Engine 000: Avg prompt throughput: 774.5 tokens/s, Avg generation throughput: 3083.7 tokens/s, Running: 215 reqs, Waiting: 4706 reqs, GPU KV cache usage: 98.6%, Prefix cache hit rate: 58.6%
248
+ (APIServer pid=114640) INFO 07-17 07:39:35 [loggers.py:273] Engine 000: Avg prompt throughput: 26.1 tokens/s, Avg generation throughput: 3475.0 tokens/s, Running: 113 reqs, Waiting: 4762 reqs, GPU KV cache usage: 97.3%, Prefix cache hit rate: 58.7%
249
+ (APIServer pid=114640) INFO 07-17 07:39:45 [loggers.py:273] Engine 000: Avg prompt throughput: 288.4 tokens/s, Avg generation throughput: 2824.0 tokens/s, Running: 174 reqs, Waiting: 4617 reqs, GPU KV cache usage: 95.2%, Prefix cache hit rate: 59.3%
250
+ (APIServer pid=114640) INFO 07-17 07:39:55 [loggers.py:273] Engine 000: Avg prompt throughput: 321.3 tokens/s, Avg generation throughput: 3403.7 tokens/s, Running: 112 reqs, Waiting: 4622 reqs, GPU KV cache usage: 99.9%, Prefix cache hit rate: 58.9%
251
+ (APIServer pid=114640) INFO 07-17 07:40:05 [loggers.py:273] Engine 000: Avg prompt throughput: 445.0 tokens/s, Avg generation throughput: 3473.8 tokens/s, Running: 133 reqs, Waiting: 4524 reqs, GPU KV cache usage: 99.6%, Prefix cache hit rate: 60.3%
252
+ (APIServer pid=114640) INFO 07-17 07:40:15 [loggers.py:273] Engine 000: Avg prompt throughput: 416.0 tokens/s, Avg generation throughput: 3166.8 tokens/s, Running: 139 reqs, Waiting: 4432 reqs, GPU KV cache usage: 99.8%, Prefix cache hit rate: 61.7%
253
+ (APIServer pid=114640) INFO 07-17 07:40:25 [loggers.py:273] Engine 000: Avg prompt throughput: 283.6 tokens/s, Avg generation throughput: 2985.8 tokens/s, Running: 149 reqs, Waiting: 4360 reqs, GPU KV cache usage: 99.1%, Prefix cache hit rate: 62.4%
254
+ (APIServer pid=114640) INFO 07-17 07:40:35 [loggers.py:273] Engine 000: Avg prompt throughput: 927.0 tokens/s, Avg generation throughput: 3649.9 tokens/s, Running: 200 reqs, Waiting: 4218 reqs, GPU KV cache usage: 99.8%, Prefix cache hit rate: 63.1%
255
+ (APIServer pid=114640) INFO 07-17 07:40:45 [loggers.py:273] Engine 000: Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 3314.7 tokens/s, Running: 113 reqs, Waiting: 4253 reqs, GPU KV cache usage: 99.7%, Prefix cache hit rate: 63.1%
256
+ (APIServer pid=114640) INFO 07-17 07:40:55 [loggers.py:273] Engine 000: Avg prompt throughput: 229.0 tokens/s, Avg generation throughput: 2389.3 tokens/s, Running: 168 reqs, Waiting: 4140 reqs, GPU KV cache usage: 99.7%, Prefix cache hit rate: 63.8%
257
+ (APIServer pid=114640) INFO 07-17 07:41:05 [loggers.py:273] Engine 000: Avg prompt throughput: 627.0 tokens/s, Avg generation throughput: 3971.3 tokens/s, Running: 224 reqs, Waiting: 3968 reqs, GPU KV cache usage: 97.3%, Prefix cache hit rate: 63.8%
258
+ (APIServer pid=114640) INFO 07-17 07:41:15 [loggers.py:273] Engine 000: Avg prompt throughput: 295.9 tokens/s, Avg generation throughput: 3754.0 tokens/s, Running: 117 reqs, Waiting: 4011 reqs, GPU KV cache usage: 98.9%, Prefix cache hit rate: 64.7%
259
+ (APIServer pid=114640) INFO 07-17 07:41:25 [loggers.py:273] Engine 000: Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 2510.2 tokens/s, Running: 98 reqs, Waiting: 3961 reqs, GPU KV cache usage: 99.9%, Prefix cache hit rate: 64.7%
260
+ (APIServer pid=114640) INFO 07-17 07:41:35 [loggers.py:273] Engine 000: Avg prompt throughput: 996.4 tokens/s, Avg generation throughput: 3622.1 tokens/s, Running: 256 reqs, Waiting: 3712 reqs, GPU KV cache usage: 90.6%, Prefix cache hit rate: 64.8%
261
+ (APIServer pid=114640) INFO 07-17 07:41:45 [loggers.py:273] Engine 000: Avg prompt throughput: 78.2 tokens/s, Avg generation throughput: 4041.9 tokens/s, Running: 123 reqs, Waiting: 3769 reqs, GPU KV cache usage: 99.9%, Prefix cache hit rate: 64.8%
262
+ (APIServer pid=114640) INFO 07-17 07:41:55 [loggers.py:273] Engine 000: Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 2440.4 tokens/s, Running: 124 reqs, Waiting: 3700 reqs, GPU KV cache usage: 98.6%, Prefix cache hit rate: 64.8%
263
+ (APIServer pid=114640) INFO 07-17 07:42:05 [loggers.py:273] Engine 000: Avg prompt throughput: 1071.1 tokens/s, Avg generation throughput: 3537.6 tokens/s, Running: 256 reqs, Waiting: 3475 reqs, GPU KV cache usage: 98.5%, Prefix cache hit rate: 64.2%
264
+ (APIServer pid=114640) INFO 07-17 07:42:15 [loggers.py:273] Engine 000: Avg prompt throughput: 185.1 tokens/s, Avg generation throughput: 4191.2 tokens/s, Running: 148 reqs, Waiting: 3484 reqs, GPU KV cache usage: 98.8%, Prefix cache hit rate: 64.1%
265
+ (APIServer pid=114640) INFO 07-17 07:42:25 [loggers.py:273] Engine 000: Avg prompt throughput: 91.3 tokens/s, Avg generation throughput: 2920.9 tokens/s, Running: 128 reqs, Waiting: 3428 reqs, GPU KV cache usage: 98.9%, Prefix cache hit rate: 62.7%
266
+ (APIServer pid=114640) INFO 07-17 07:42:35 [loggers.py:273] Engine 000: Avg prompt throughput: 551.2 tokens/s, Avg generation throughput: 3084.0 tokens/s, Running: 150 reqs, Waiting: 3340 reqs, GPU KV cache usage: 98.5%, Prefix cache hit rate: 61.3%
267
+ (APIServer pid=114640) INFO 07-17 07:42:45 [loggers.py:273] Engine 000: Avg prompt throughput: 714.3 tokens/s, Avg generation throughput: 4758.7 tokens/s, Running: 166 reqs, Waiting: 3204 reqs, GPU KV cache usage: 99.2%, Prefix cache hit rate: 61.7%
268
+ (APIServer pid=114640) INFO 07-17 07:42:55 [loggers.py:273] Engine 000: Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 2982.7 tokens/s, Running: 110 reqs, Waiting: 3198 reqs, GPU KV cache usage: 99.9%, Prefix cache hit rate: 61.7%
269
+ (APIServer pid=114640) INFO 07-17 07:43:05 [loggers.py:273] Engine 000: Avg prompt throughput: 308.8 tokens/s, Avg generation throughput: 3099.8 tokens/s, Running: 134 reqs, Waiting: 3096 reqs, GPU KV cache usage: 99.7%, Prefix cache hit rate: 60.0%
270
+ (APIServer pid=114640) INFO 07-17 07:43:15 [loggers.py:273] Engine 000: Avg prompt throughput: 709.3 tokens/s, Avg generation throughput: 3840.7 tokens/s, Running: 171 reqs, Waiting: 2983 reqs, GPU KV cache usage: 99.9%, Prefix cache hit rate: 60.2%
271
+ (APIServer pid=114640) INFO 07-17 07:43:25 [loggers.py:273] Engine 000: Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 3094.4 tokens/s, Running: 128 reqs, Waiting: 2956 reqs, GPU KV cache usage: 99.9%, Prefix cache hit rate: 60.2%
272
+ (APIServer pid=114640) INFO 07-17 07:43:35 [loggers.py:273] Engine 000: Avg prompt throughput: 690.5 tokens/s, Avg generation throughput: 3310.5 tokens/s, Running: 194 reqs, Waiting: 2800 reqs, GPU KV cache usage: 99.4%, Prefix cache hit rate: 58.5%
273
+ (APIServer pid=114640) INFO 07-17 07:43:45 [loggers.py:273] Engine 000: Avg prompt throughput: 299.3 tokens/s, Avg generation throughput: 3494.8 tokens/s, Running: 211 reqs, Waiting: 2692 reqs, GPU KV cache usage: 96.9%, Prefix cache hit rate: 59.0%
274
+ (APIServer pid=114640) INFO 07-17 07:43:55 [loggers.py:273] Engine 000: Avg prompt throughput: 213.8 tokens/s, Avg generation throughput: 3481.7 tokens/s, Running: 106 reqs, Waiting: 2725 reqs, GPU KV cache usage: 99.7%, Prefix cache hit rate: 59.6%
275
+ (APIServer pid=114640) INFO 07-17 07:44:05 [loggers.py:273] Engine 000: Avg prompt throughput: 412.8 tokens/s, Avg generation throughput: 3237.1 tokens/s, Running: 158 reqs, Waiting: 2589 reqs, GPU KV cache usage: 99.9%, Prefix cache hit rate: 59.5%
276
+ (APIServer pid=114640) INFO 07-17 07:44:15 [loggers.py:273] Engine 000: Avg prompt throughput: 447.9 tokens/s, Avg generation throughput: 3791.4 tokens/s, Running: 183 reqs, Waiting: 2476 reqs, GPU KV cache usage: 100.0%, Prefix cache hit rate: 61.7%
277
+ (APIServer pid=114640) INFO 07-17 07:44:25 [loggers.py:273] Engine 000: Avg prompt throughput: 204.7 tokens/s, Avg generation throughput: 3402.3 tokens/s, Running: 166 reqs, Waiting: 2405 reqs, GPU KV cache usage: 100.0%, Prefix cache hit rate: 62.1%
278
+ (APIServer pid=114640) INFO 07-17 07:44:35 [loggers.py:273] Engine 000: Avg prompt throughput: 264.5 tokens/s, Avg generation throughput: 3002.4 tokens/s, Running: 162 reqs, Waiting: 2344 reqs, GPU KV cache usage: 100.0%, Prefix cache hit rate: 61.3%
279
+ (APIServer pid=114640) INFO 07-17 07:44:45 [loggers.py:273] Engine 000: Avg prompt throughput: 744.4 tokens/s, Avg generation throughput: 3536.9 tokens/s, Running: 253 reqs, Waiting: 2159 reqs, GPU KV cache usage: 91.2%, Prefix cache hit rate: 62.3%
280
+ (APIServer pid=114640) INFO 07-17 07:44:55 [loggers.py:273] Engine 000: Avg prompt throughput: 24.8 tokens/s, Avg generation throughput: 3870.2 tokens/s, Running: 147 reqs, Waiting: 2207 reqs, GPU KV cache usage: 99.9%, Prefix cache hit rate: 62.3%
281
+ (APIServer pid=114640) INFO 07-17 07:45:05 [loggers.py:273] Engine 000: Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 2952.7 tokens/s, Running: 119 reqs, Waiting: 2172 reqs, GPU KV cache usage: 99.7%, Prefix cache hit rate: 62.3%
282
+ (APIServer pid=114640) INFO 07-17 07:45:15 [loggers.py:273] Engine 000: Avg prompt throughput: 1057.3 tokens/s, Avg generation throughput: 4199.4 tokens/s, Running: 188 reqs, Waiting: 2005 reqs, GPU KV cache usage: 99.8%, Prefix cache hit rate: 58.5%
283
+ (APIServer pid=114640) INFO 07-17 07:45:25 [loggers.py:273] Engine 000: Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 3416.4 tokens/s, Running: 114 reqs, Waiting: 2000 reqs, GPU KV cache usage: 99.2%, Prefix cache hit rate: 58.5%
284
+ (APIServer pid=114640) INFO 07-17 07:45:35 [loggers.py:273] Engine 000: Avg prompt throughput: 459.9 tokens/s, Avg generation throughput: 3120.7 tokens/s, Running: 155 reqs, Waiting: 1887 reqs, GPU KV cache usage: 100.0%, Prefix cache hit rate: 59.7%
285
+ (APIServer pid=114640) INFO 07-17 07:45:45 [loggers.py:273] Engine 000: Avg prompt throughput: 690.5 tokens/s, Avg generation throughput: 3359.8 tokens/s, Running: 215 reqs, Waiting: 1741 reqs, GPU KV cache usage: 97.2%, Prefix cache hit rate: 60.0%
286
+ (APIServer pid=114640) INFO 07-17 07:45:55 [loggers.py:273] Engine 000: Avg prompt throughput: 14.2 tokens/s, Avg generation throughput: 3588.6 tokens/s, Running: 120 reqs, Waiting: 1766 reqs, GPU KV cache usage: 99.9%, Prefix cache hit rate: 59.8%
287
+ (APIServer pid=114640) INFO 07-17 07:46:05 [loggers.py:273] Engine 000: Avg prompt throughput: 579.5 tokens/s, Avg generation throughput: 3361.0 tokens/s, Running: 157 reqs, Waiting: 1641 reqs, GPU KV cache usage: 99.7%, Prefix cache hit rate: 60.4%
288
+ (APIServer pid=114640) INFO 07-17 07:46:15 [loggers.py:273] Engine 000: Avg prompt throughput: 154.0 tokens/s, Avg generation throughput: 3570.0 tokens/s, Running: 129 reqs, Waiting: 1594 reqs, GPU KV cache usage: 99.9%, Prefix cache hit rate: 59.6%
289
+ (APIServer pid=114640) INFO 07-17 07:46:25 [loggers.py:273] Engine 000: Avg prompt throughput: 356.4 tokens/s, Avg generation throughput: 3209.2 tokens/s, Running: 134 reqs, Waiting: 1512 reqs, GPU KV cache usage: 99.9%, Prefix cache hit rate: 59.9%
290
+ (APIServer pid=114640) INFO 07-17 07:46:35 [loggers.py:273] Engine 000: Avg prompt throughput: 572.7 tokens/s, Avg generation throughput: 3611.6 tokens/s, Running: 176 reqs, Waiting: 1364 reqs, GPU KV cache usage: 98.5%, Prefix cache hit rate: 60.6%
291
+ (APIServer pid=114640) INFO 07-17 07:46:45 [loggers.py:273] Engine 000: Avg prompt throughput: 302.9 tokens/s, Avg generation throughput: 3532.0 tokens/s, Running: 116 reqs, Waiting: 1355 reqs, GPU KV cache usage: 99.7%, Prefix cache hit rate: 60.1%
292
+ (APIServer pid=114640) INFO 07-17 07:46:55 [loggers.py:273] Engine 000: Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 2623.2 tokens/s, Running: 97 reqs, Waiting: 1315 reqs, GPU KV cache usage: 99.9%, Prefix cache hit rate: 60.1%
293
+ (APIServer pid=114640) INFO 07-17 07:47:05 [loggers.py:273] Engine 000: Avg prompt throughput: 928.9 tokens/s, Avg generation throughput: 3949.0 tokens/s, Running: 256 reqs, Waiting: 1040 reqs, GPU KV cache usage: 97.6%, Prefix cache hit rate: 64.3%
294
+ (APIServer pid=114640) INFO 07-17 07:47:15 [loggers.py:273] Engine 000: Avg prompt throughput: 188.8 tokens/s, Avg generation throughput: 4374.0 tokens/s, Running: 133 reqs, Waiting: 1067 reqs, GPU KV cache usage: 99.1%, Prefix cache hit rate: 64.7%
295
+ (APIServer pid=114640) INFO 07-17 07:47:25 [loggers.py:273] Engine 000: Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 2592.8 tokens/s, Running: 112 reqs, Waiting: 1017 reqs, GPU KV cache usage: 100.0%, Prefix cache hit rate: 64.7%
296
+ (APIServer pid=114640) INFO 07-17 07:47:35 [loggers.py:273] Engine 000: Avg prompt throughput: 1018.3 tokens/s, Avg generation throughput: 3506.5 tokens/s, Running: 254 reqs, Waiting: 746 reqs, GPU KV cache usage: 74.3%, Prefix cache hit rate: 64.8%
297
+ (APIServer pid=114640) INFO 07-17 07:47:45 [loggers.py:273] Engine 000: Avg prompt throughput: 97.9 tokens/s, Avg generation throughput: 4570.2 tokens/s, Running: 138 reqs, Waiting: 795 reqs, GPU KV cache usage: 99.7%, Prefix cache hit rate: 64.6%
298
+ (APIServer pid=114640) INFO 07-17 07:47:55 [loggers.py:273] Engine 000: Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 2673.9 tokens/s, Running: 123 reqs, Waiting: 741 reqs, GPU KV cache usage: 99.9%, Prefix cache hit rate: 64.6%
299
+ (APIServer pid=114640) INFO 07-17 07:48:05 [loggers.py:273] Engine 000: Avg prompt throughput: 896.7 tokens/s, Avg generation throughput: 3841.0 tokens/s, Running: 169 reqs, Waiting: 605 reqs, GPU KV cache usage: 99.2%, Prefix cache hit rate: 65.5%
300
+ (APIServer pid=114640) INFO 07-17 07:48:15 [loggers.py:273] Engine 000: Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 3061.3 tokens/s, Running: 107 reqs, Waiting: 616 reqs, GPU KV cache usage: 100.0%, Prefix cache hit rate: 65.5%
301
+ (APIServer pid=114640) INFO 07-17 07:48:25 [loggers.py:273] Engine 000: Avg prompt throughput: 212.6 tokens/s, Avg generation throughput: 2727.9 tokens/s, Running: 169 reqs, Waiting: 491 reqs, GPU KV cache usage: 100.0%, Prefix cache hit rate: 65.6%
302
+ (APIServer pid=114640) INFO 07-17 07:48:35 [loggers.py:273] Engine 000: Avg prompt throughput: 818.1 tokens/s, Avg generation throughput: 3571.1 tokens/s, Running: 256 reqs, Waiting: 304 reqs, GPU KV cache usage: 92.7%, Prefix cache hit rate: 66.6%
303
+ (APIServer pid=114640) INFO 07-17 07:48:45 [loggers.py:273] Engine 000: Avg prompt throughput: 38.2 tokens/s, Avg generation throughput: 3975.5 tokens/s, Running: 163 reqs, Waiting: 308 reqs, GPU KV cache usage: 99.8%, Prefix cache hit rate: 66.6%
304
+ (APIServer pid=114640) INFO 07-17 07:48:55 [loggers.py:273] Engine 000: Avg prompt throughput: 201.8 tokens/s, Avg generation throughput: 2881.4 tokens/s, Running: 147 reqs, Waiting: 256 reqs, GPU KV cache usage: 94.9%, Prefix cache hit rate: 66.0%
305
+ (APIServer pid=114640) INFO 07-17 07:49:05 [loggers.py:273] Engine 000: Avg prompt throughput: 356.0 tokens/s, Avg generation throughput: 3223.9 tokens/s, Running: 109 reqs, Waiting: 238 reqs, GPU KV cache usage: 99.8%, Prefix cache hit rate: 65.9%
306
+ (APIServer pid=114640) INFO 07-17 07:49:15 [loggers.py:273] Engine 000: Avg prompt throughput: 779.1 tokens/s, Avg generation throughput: 4190.1 tokens/s, Running: 155 reqs, Waiting: 79 reqs, GPU KV cache usage: 99.2%, Prefix cache hit rate: 66.2%
307
+ (APIServer pid=114640) INFO 07-17 07:49:25 [loggers.py:273] Engine 000: Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 2792.5 tokens/s, Running: 90 reqs, Waiting: 83 reqs, GPU KV cache usage: 100.0%, Prefix cache hit rate: 66.2%
308
+ (APIServer pid=114640) INFO 07-17 07:49:35 [loggers.py:273] Engine 000: Avg prompt throughput: 46.4 tokens/s, Avg generation throughput: 2781.4 tokens/s, Running: 77 reqs, Waiting: 0 reqs, GPU KV cache usage: 62.1%, Prefix cache hit rate: 66.6%
309
+ (APIServer pid=114640) INFO 07-17 07:49:45 [loggers.py:273] Engine 000: Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 1590.5 tokens/s, Running: 8 reqs, Waiting: 0 reqs, GPU KV cache usage: 11.2%, Prefix cache hit rate: 66.6%
310
+ (APIServer pid=114640) INFO: 127.0.0.1:55130 - "POST /v1/completions HTTP/1.1" 200 OK
311
+ (EngineCore pid=114758) INFO 07-17 07:49:49 [core.py:1214] [shutdown] EngineCore: trigger received signal=SIGTERM
312
+ (APIServer pid=114640) INFO 07-17 07:49:49 [launcher.py:100] [shutdown] API server: shutdown triggered
313
+ (APIServer pid=114640) INFO 07-17 07:49:49 [launcher.py:116] [shutdown] API server: stopping engine client mode=abort timeout=0s
314
+ (EngineCore pid=114758) INFO 07-17 07:49:49 [core.py:1333] [shutdown] EngineCore: start mode=abort timeout=0s
315
+ (EngineCore pid=114758) INFO 07-17 07:49:49 [core.py:1364] [shutdown] EngineCore: request processing complete; starting resource teardown
316
+ (EngineCore pid=114758) INFO 07-17 07:49:49 [core.py:1227] [shutdown] EngineCore: exiting busy loop
317
+ (APIServer pid=114640) INFO 07-17 07:49:49 [core_client.py:655] [shutdown] MPClient: start timeout=0s
318
+ (APIServer pid=114640) INFO 07-17 07:49:49 [core_client.py:657] [shutdown] MPClient: stopping engine manager
319
+ (APIServer pid=114640) WARNING 07-17 07:49:49 [utils.py:626] [shutdown] Process manager: force killing remaining processes count=1
320
+ (APIServer pid=114640) INFO: Shutting down
321
+ (APIServer pid=114640) INFO 07-17 07:49:49 [core_client.py:659] [shutdown] MPClient: engine manager stopped
322
+ (APIServer pid=114640) INFO 07-17 07:49:49 [core_client.py:660] [shutdown] MPClient: cleaning up background resources
323
+ (APIServer pid=114640) INFO 07-17 07:49:49 [core_client.py:662] [shutdown] MPClient: complete
324
+ (APIServer pid=114640) INFO 07-17 07:49:49 [launcher.py:125] [shutdown] API server: engine client stopped
325
+ (APIServer pid=114640) INFO 07-17 07:49:49 [launcher.py:128] [shutdown] API server: signalling HTTP server shutdown
326
+ (APIServer pid=114640) INFO 07-17 07:49:49 [launcher.py:149] [shutdown] API server: shutting down FastAPI HTTP server
327
+ (APIServer pid=114640) INFO: Shutting down
328
+ (APIServer pid=114640) INFO: Waiting for application shutdown.
329
+ (APIServer pid=114640) INFO: Application shutdown complete.
330
+ /home/henry/.local/share/uv/python/cpython-3.12.12-linux-x86_64-gnu/lib/python3.12/multiprocessing/resource_tracker.py:279: UserWarning: resource_tracker: There appear to be 1 leaked semaphore objects to clean up at shutdown
331
+ warnings.warn('resource_tracker: There appear to be %d '
teacher_trajectories/top128_score.log ADDED
@@ -0,0 +1,783 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ reconstructing exact chat prompts ...
2
+ loading teacher on cuda:0 ...
3
+
4
+ shard 00000: 128/12115 records | 73,031 tokens | 9.34 records/s
5
+ shard 00001: 256/12115 records | 142,706 tokens | 9.67 records/s
6
+ shard 00002: 384/12115 records | 221,592 tokens | 9.64 records/s
7
+ shard 00003: 512/12115 records | 290,966 tokens | 9.70 records/s
8
+ shard 00004: 640/12115 records | 362,190 tokens | 9.72 records/s
9
+ shard 00005: 768/12115 records | 432,885 tokens | 9.73 records/s
10
+ shard 00006: 896/12115 records | 507,356 tokens | 9.74 records/s
11
+ shard 00007: 1024/12115 records | 578,914 tokens | 9.75 records/s
12
+ shard 00008: 1152/12115 records | 646,302 tokens | 9.78 records/s
13
+ shard 00009: 1280/12115 records | 715,522 tokens | 9.78 records/s
14
+ shard 00010: 1408/12115 records | 789,398 tokens | 9.77 records/s
15
+ shard 00011: 1536/12115 records | 859,905 tokens | 9.77 records/s
16
+ shard 00012: 1664/12115 records | 929,957 tokens | 9.79 records/s
17
+ shard 00013: 1792/12115 records | 1,002,897 tokens | 9.79 records/s
18
+ shard 00014: 1920/12115 records | 1,076,939 tokens | 9.79 records/s
19
+ shard 00015: 2048/12115 records | 1,152,867 tokens | 9.78 records/s
20
+ shard 00016: 2176/12115 records | 1,226,568 tokens | 9.78 records/s
21
+ shard 00017: 2304/12115 records | 1,295,943 tokens | 9.79 records/s
22
+ shard 00018: 2432/12115 records | 1,367,163 tokens | 9.79 records/s
23
+ shard 00019: 2560/12115 records | 1,444,451 tokens | 9.78 records/s
24
+ shard 00020: 2688/12115 records | 1,517,881 tokens | 9.79 records/s
25
+ shard 00021: 2816/12115 records | 1,586,866 tokens | 9.80 records/s
26
+ shard 00022: 2944/12115 records | 1,661,501 tokens | 9.80 records/s
27
+ shard 00023: 3072/12115 records | 1,733,775 tokens | 9.79 records/s
28
+ shard 00024: 3200/12115 records | 1,806,409 tokens | 9.80 records/s
29
+ shard 00025: 3328/12115 records | 1,882,641 tokens | 9.80 records/s
30
+ shard 00026: 3456/12115 records | 1,947,920 tokens | 9.80 records/s
31
+ shard 00027: 3584/12115 records | 2,010,837 tokens | 9.82 records/s
32
+ shard 00028: 3712/12115 records | 2,069,317 tokens | 9.84 records/s
33
+ shard 00029: 3840/12115 records | 2,128,231 tokens | 9.85 records/s
34
+ shard 00030: 3968/12115 records | 2,190,763 tokens | 9.87 records/s
35
+ shard 00031: 4096/12115 records | 2,252,493 tokens | 9.88 records/s
36
+ shard 00032: 4224/12115 records | 2,312,761 tokens | 9.90 records/s
37
+ shard 00033: 4352/12115 records | 2,375,766 tokens | 9.91 records/s
38
+ shard 00034: 4480/12115 records | 2,439,594 tokens | 9.92 records/s
39
+ shard 00035: 4608/12115 records | 2,495,617 tokens | 9.94 records/s
40
+ shard 00036: 4736/12115 records | 2,558,881 tokens | 9.95 records/s
41
+ shard 00037: 4864/12115 records | 2,618,211 tokens | 9.96 records/s
42
+ shard 00038: 4992/12115 records | 2,677,399 tokens | 9.97 records/s
43
+ shard 00039: 5120/12115 records | 2,735,990 tokens | 9.98 records/s
44
+ shard 00040: 5248/12115 records | 2,797,207 tokens | 9.99 records/s
45
+ shard 00041: 5376/12115 records | 2,854,573 tokens | 10.00 records/s
46
+ shard 00042: 5504/12115 records | 2,915,874 tokens | 10.01 records/s
47
+ shard 00043: 5632/12115 records | 2,979,417 tokens | 10.01 records/s
48
+ shard 00044: 5760/12115 records | 3,038,136 tokens | 10.02 records/s
49
+ shard 00045: 5888/12115 records | 3,097,503 tokens | 10.03 records/s
50
+ shard 00046: 6016/12115 records | 3,152,821 tokens | 10.04 records/s
51
+ shard 00047: 6144/12115 records | 3,217,264 tokens | 10.05 records/s
52
+ shard 00048: 6272/12115 records | 3,277,016 tokens | 10.05 records/s
53
+ shard 00049: 6400/12115 records | 3,341,336 tokens | 10.05 records/s
54
+ shard 00050: 6528/12115 records | 3,405,709 tokens | 10.06 records/s
55
+ shard 00051: 6656/12115 records | 3,460,422 tokens | 10.07 records/s
56
+ shard 00052: 6784/12115 records | 3,522,884 tokens | 10.07 records/s
57
+ shard 00053: 6912/12115 records | 3,582,024 tokens | 10.07 records/s
58
+ shard 00054: 7040/12115 records | 3,648,314 tokens | 10.07 records/s
59
+ shard 00055: 7168/12115 records | 3,714,896 tokens | 10.08 records/s
60
+ shard 00056: 7296/12115 records | 3,777,946 tokens | 10.09 records/s
61
+ shard 00057: 7424/12115 records | 3,836,597 tokens | 10.09 records/s
62
+ shard 00058: 7552/12115 records | 3,896,186 tokens | 10.10 records/s
63
+ shard 00059: 7680/12115 records | 3,956,340 tokens | 10.11 records/s
64
+ shard 00060: 7808/12115 records | 4,012,509 tokens | 10.11 records/s
65
+ shard 00061: 7936/12115 records | 4,076,475 tokens | 10.11 records/s
66
+ shard 00062: 8064/12115 records | 4,151,986 tokens | 10.10 records/s
67
+ shard 00063: 8192/12115 records | 4,232,690 tokens | 10.09 records/s
68
+ shard 00064: 8320/12115 records | 4,307,965 tokens | 10.08 records/s
69
+ shard 00065: 8448/12115 records | 4,380,815 tokens | 10.07 records/s
70
+ shard 00066: 8576/12115 records | 4,453,096 tokens | 10.06 records/s
71
+ shard 00067: 8704/12115 records | 4,527,660 tokens | 10.04 records/s
72
+ shard 00068: 8832/12115 records | 4,605,858 tokens | 10.03 records/s
73
+ shard 00069: 8960/12115 records | 4,680,546 tokens | 10.02 records/s
74
+ shard 00070: 9088/12115 records | 4,753,994 tokens | 10.02 records/s
75
+ shard 00071: 9216/12115 records | 4,827,678 tokens | 10.01 records/s
76
+ shard 00072: 9344/12115 records | 4,903,825 tokens | 10.00 records/s
77
+ shard 00073: 9472/12115 records | 4,977,432 tokens | 10.00 records/s
78
+ shard 00074: 9600/12115 records | 5,054,090 tokens | 9.99 records/s
79
+ shard 00075: 9728/12115 records | 5,125,142 tokens | 9.99 records/s
80
+ shard 00076: 9856/12115 records | 5,196,981 tokens | 9.99 records/s
81
+ shard 00077: 9984/12115 records | 5,261,629 tokens | 9.99 records/s
82
+ shard 00078: 10112/12115 records | 5,331,065 tokens | 9.99 records/s
83
+ shard 00079: 10240/12115 records | 5,403,259 tokens | 9.99 records/s
84
+ shard 00080: 10368/12115 records | 5,477,552 tokens | 9.99 records/s
85
+ shard 00081: 10496/12115 records | 5,550,170 tokens | 9.98 records/s
86
+ shard 00082: 10624/12115 records | 5,625,121 tokens | 9.98 records/s
87
+ shard 00083: 10752/12115 records | 5,691,110 tokens | 9.98 records/s
88
+ shard 00084: 10880/12115 records | 5,761,375 tokens | 9.98 records/s
89
+ shard 00085: 11008/12115 records | 5,833,732 tokens | 9.98 records/s
90
+ shard 00086: 11136/12115 records | 5,904,113 tokens | 9.98 records/s
91
+ shard 00087: 11264/12115 records | 5,967,914 tokens | 9.99 records/s
92
+ shard 00088: 11392/12115 records | 6,040,652 tokens | 9.99 records/s
93
+ shard 00089: 11520/12115 records | 6,114,847 tokens | 9.98 records/s
94
+ shard 00090: 11648/12115 records | 6,193,540 tokens | 9.98 records/s
95
+ shard 00091: 11776/12115 records | 6,269,234 tokens | 9.98 records/s
96
+ shard 00092: 11904/12115 records | 6,349,496 tokens | 9.97 records/s
97
+ shard 00093: 12032/12115 records | 6,429,742 tokens | 9.97 records/s
98
+ shard 00094: 12115/12115 records | 6,476,634 tokens | 9.97 records/s
99
+ {
100
+ "version": 1,
101
+ "complete": true,
102
+ "teacher": "allenai/OLMoE-1B-7B-0125-Instruct",
103
+ "tokenizer": "allenai/OLMoE-1B-7B-0125-Instruct",
104
+ "source": "/home/henry/Documents/PythonProjects/variable-reap/outputs/teacher_trajectories/dolci_math_curated_opd.jsonl",
105
+ "source_size": 57436902,
106
+ "frame": "chat",
107
+ "topk": 128,
108
+ "max_prompt_len": 1024,
109
+ "max_seq_len": 2048,
110
+ "pad_token_id": 50280,
111
+ "eos_token_id": 50279,
112
+ "total_records": 12115,
113
+ "total_tokens": 6476634,
114
+ "shards": [
115
+ {
116
+ "path": "shard_00000.pt",
117
+ "records": 128,
118
+ "tokens": 73031,
119
+ "first_record": 0,
120
+ "last_record": 127
121
+ },
122
+ {
123
+ "path": "shard_00001.pt",
124
+ "records": 128,
125
+ "tokens": 69675,
126
+ "first_record": 128,
127
+ "last_record": 255
128
+ },
129
+ {
130
+ "path": "shard_00002.pt",
131
+ "records": 128,
132
+ "tokens": 78886,
133
+ "first_record": 256,
134
+ "last_record": 383
135
+ },
136
+ {
137
+ "path": "shard_00003.pt",
138
+ "records": 128,
139
+ "tokens": 69374,
140
+ "first_record": 384,
141
+ "last_record": 511
142
+ },
143
+ {
144
+ "path": "shard_00004.pt",
145
+ "records": 128,
146
+ "tokens": 71224,
147
+ "first_record": 512,
148
+ "last_record": 639
149
+ },
150
+ {
151
+ "path": "shard_00005.pt",
152
+ "records": 128,
153
+ "tokens": 70695,
154
+ "first_record": 640,
155
+ "last_record": 767
156
+ },
157
+ {
158
+ "path": "shard_00006.pt",
159
+ "records": 128,
160
+ "tokens": 74471,
161
+ "first_record": 768,
162
+ "last_record": 895
163
+ },
164
+ {
165
+ "path": "shard_00007.pt",
166
+ "records": 128,
167
+ "tokens": 71558,
168
+ "first_record": 896,
169
+ "last_record": 1023
170
+ },
171
+ {
172
+ "path": "shard_00008.pt",
173
+ "records": 128,
174
+ "tokens": 67388,
175
+ "first_record": 1024,
176
+ "last_record": 1151
177
+ },
178
+ {
179
+ "path": "shard_00009.pt",
180
+ "records": 128,
181
+ "tokens": 69220,
182
+ "first_record": 1152,
183
+ "last_record": 1279
184
+ },
185
+ {
186
+ "path": "shard_00010.pt",
187
+ "records": 128,
188
+ "tokens": 73876,
189
+ "first_record": 1280,
190
+ "last_record": 1407
191
+ },
192
+ {
193
+ "path": "shard_00011.pt",
194
+ "records": 128,
195
+ "tokens": 70507,
196
+ "first_record": 1408,
197
+ "last_record": 1535
198
+ },
199
+ {
200
+ "path": "shard_00012.pt",
201
+ "records": 128,
202
+ "tokens": 70052,
203
+ "first_record": 1536,
204
+ "last_record": 1663
205
+ },
206
+ {
207
+ "path": "shard_00013.pt",
208
+ "records": 128,
209
+ "tokens": 72940,
210
+ "first_record": 1664,
211
+ "last_record": 1791
212
+ },
213
+ {
214
+ "path": "shard_00014.pt",
215
+ "records": 128,
216
+ "tokens": 74042,
217
+ "first_record": 1792,
218
+ "last_record": 1919
219
+ },
220
+ {
221
+ "path": "shard_00015.pt",
222
+ "records": 128,
223
+ "tokens": 75928,
224
+ "first_record": 1920,
225
+ "last_record": 2047
226
+ },
227
+ {
228
+ "path": "shard_00016.pt",
229
+ "records": 128,
230
+ "tokens": 73701,
231
+ "first_record": 2048,
232
+ "last_record": 2175
233
+ },
234
+ {
235
+ "path": "shard_00017.pt",
236
+ "records": 128,
237
+ "tokens": 69375,
238
+ "first_record": 2176,
239
+ "last_record": 2303
240
+ },
241
+ {
242
+ "path": "shard_00018.pt",
243
+ "records": 128,
244
+ "tokens": 71220,
245
+ "first_record": 2304,
246
+ "last_record": 2431
247
+ },
248
+ {
249
+ "path": "shard_00019.pt",
250
+ "records": 128,
251
+ "tokens": 77288,
252
+ "first_record": 2432,
253
+ "last_record": 2559
254
+ },
255
+ {
256
+ "path": "shard_00020.pt",
257
+ "records": 128,
258
+ "tokens": 73430,
259
+ "first_record": 2560,
260
+ "last_record": 2687
261
+ },
262
+ {
263
+ "path": "shard_00021.pt",
264
+ "records": 128,
265
+ "tokens": 68985,
266
+ "first_record": 2688,
267
+ "last_record": 2815
268
+ },
269
+ {
270
+ "path": "shard_00022.pt",
271
+ "records": 128,
272
+ "tokens": 74635,
273
+ "first_record": 2816,
274
+ "last_record": 2943
275
+ },
276
+ {
277
+ "path": "shard_00023.pt",
278
+ "records": 128,
279
+ "tokens": 72274,
280
+ "first_record": 2944,
281
+ "last_record": 3071
282
+ },
283
+ {
284
+ "path": "shard_00024.pt",
285
+ "records": 128,
286
+ "tokens": 72634,
287
+ "first_record": 3072,
288
+ "last_record": 3199
289
+ },
290
+ {
291
+ "path": "shard_00025.pt",
292
+ "records": 128,
293
+ "tokens": 76232,
294
+ "first_record": 3200,
295
+ "last_record": 3327
296
+ },
297
+ {
298
+ "path": "shard_00026.pt",
299
+ "records": 128,
300
+ "tokens": 65279,
301
+ "first_record": 3328,
302
+ "last_record": 3455
303
+ },
304
+ {
305
+ "path": "shard_00027.pt",
306
+ "records": 128,
307
+ "tokens": 62917,
308
+ "first_record": 3456,
309
+ "last_record": 3583
310
+ },
311
+ {
312
+ "path": "shard_00028.pt",
313
+ "records": 128,
314
+ "tokens": 58480,
315
+ "first_record": 3584,
316
+ "last_record": 3711
317
+ },
318
+ {
319
+ "path": "shard_00029.pt",
320
+ "records": 128,
321
+ "tokens": 58914,
322
+ "first_record": 3712,
323
+ "last_record": 3839
324
+ },
325
+ {
326
+ "path": "shard_00030.pt",
327
+ "records": 128,
328
+ "tokens": 62532,
329
+ "first_record": 3840,
330
+ "last_record": 3967
331
+ },
332
+ {
333
+ "path": "shard_00031.pt",
334
+ "records": 128,
335
+ "tokens": 61730,
336
+ "first_record": 3968,
337
+ "last_record": 4095
338
+ },
339
+ {
340
+ "path": "shard_00032.pt",
341
+ "records": 128,
342
+ "tokens": 60268,
343
+ "first_record": 4096,
344
+ "last_record": 4223
345
+ },
346
+ {
347
+ "path": "shard_00033.pt",
348
+ "records": 128,
349
+ "tokens": 63005,
350
+ "first_record": 4224,
351
+ "last_record": 4351
352
+ },
353
+ {
354
+ "path": "shard_00034.pt",
355
+ "records": 128,
356
+ "tokens": 63828,
357
+ "first_record": 4352,
358
+ "last_record": 4479
359
+ },
360
+ {
361
+ "path": "shard_00035.pt",
362
+ "records": 128,
363
+ "tokens": 56023,
364
+ "first_record": 4480,
365
+ "last_record": 4607
366
+ },
367
+ {
368
+ "path": "shard_00036.pt",
369
+ "records": 128,
370
+ "tokens": 63264,
371
+ "first_record": 4608,
372
+ "last_record": 4735
373
+ },
374
+ {
375
+ "path": "shard_00037.pt",
376
+ "records": 128,
377
+ "tokens": 59330,
378
+ "first_record": 4736,
379
+ "last_record": 4863
380
+ },
381
+ {
382
+ "path": "shard_00038.pt",
383
+ "records": 128,
384
+ "tokens": 59188,
385
+ "first_record": 4864,
386
+ "last_record": 4991
387
+ },
388
+ {
389
+ "path": "shard_00039.pt",
390
+ "records": 128,
391
+ "tokens": 58591,
392
+ "first_record": 4992,
393
+ "last_record": 5119
394
+ },
395
+ {
396
+ "path": "shard_00040.pt",
397
+ "records": 128,
398
+ "tokens": 61217,
399
+ "first_record": 5120,
400
+ "last_record": 5247
401
+ },
402
+ {
403
+ "path": "shard_00041.pt",
404
+ "records": 128,
405
+ "tokens": 57366,
406
+ "first_record": 5248,
407
+ "last_record": 5375
408
+ },
409
+ {
410
+ "path": "shard_00042.pt",
411
+ "records": 128,
412
+ "tokens": 61301,
413
+ "first_record": 5376,
414
+ "last_record": 5503
415
+ },
416
+ {
417
+ "path": "shard_00043.pt",
418
+ "records": 128,
419
+ "tokens": 63543,
420
+ "first_record": 5504,
421
+ "last_record": 5631
422
+ },
423
+ {
424
+ "path": "shard_00044.pt",
425
+ "records": 128,
426
+ "tokens": 58719,
427
+ "first_record": 5632,
428
+ "last_record": 5759
429
+ },
430
+ {
431
+ "path": "shard_00045.pt",
432
+ "records": 128,
433
+ "tokens": 59367,
434
+ "first_record": 5760,
435
+ "last_record": 5887
436
+ },
437
+ {
438
+ "path": "shard_00046.pt",
439
+ "records": 128,
440
+ "tokens": 55318,
441
+ "first_record": 5888,
442
+ "last_record": 6015
443
+ },
444
+ {
445
+ "path": "shard_00047.pt",
446
+ "records": 128,
447
+ "tokens": 64443,
448
+ "first_record": 6016,
449
+ "last_record": 6143
450
+ },
451
+ {
452
+ "path": "shard_00048.pt",
453
+ "records": 128,
454
+ "tokens": 59752,
455
+ "first_record": 6144,
456
+ "last_record": 6271
457
+ },
458
+ {
459
+ "path": "shard_00049.pt",
460
+ "records": 128,
461
+ "tokens": 64320,
462
+ "first_record": 6272,
463
+ "last_record": 6399
464
+ },
465
+ {
466
+ "path": "shard_00050.pt",
467
+ "records": 128,
468
+ "tokens": 64373,
469
+ "first_record": 6400,
470
+ "last_record": 6527
471
+ },
472
+ {
473
+ "path": "shard_00051.pt",
474
+ "records": 128,
475
+ "tokens": 54713,
476
+ "first_record": 6528,
477
+ "last_record": 6655
478
+ },
479
+ {
480
+ "path": "shard_00052.pt",
481
+ "records": 128,
482
+ "tokens": 62462,
483
+ "first_record": 6656,
484
+ "last_record": 6783
485
+ },
486
+ {
487
+ "path": "shard_00053.pt",
488
+ "records": 128,
489
+ "tokens": 59140,
490
+ "first_record": 6784,
491
+ "last_record": 6911
492
+ },
493
+ {
494
+ "path": "shard_00054.pt",
495
+ "records": 128,
496
+ "tokens": 66290,
497
+ "first_record": 6912,
498
+ "last_record": 7039
499
+ },
500
+ {
501
+ "path": "shard_00055.pt",
502
+ "records": 128,
503
+ "tokens": 66582,
504
+ "first_record": 7040,
505
+ "last_record": 7167
506
+ },
507
+ {
508
+ "path": "shard_00056.pt",
509
+ "records": 128,
510
+ "tokens": 63050,
511
+ "first_record": 7168,
512
+ "last_record": 7295
513
+ },
514
+ {
515
+ "path": "shard_00057.pt",
516
+ "records": 128,
517
+ "tokens": 58651,
518
+ "first_record": 7296,
519
+ "last_record": 7423
520
+ },
521
+ {
522
+ "path": "shard_00058.pt",
523
+ "records": 128,
524
+ "tokens": 59589,
525
+ "first_record": 7424,
526
+ "last_record": 7551
527
+ },
528
+ {
529
+ "path": "shard_00059.pt",
530
+ "records": 128,
531
+ "tokens": 60154,
532
+ "first_record": 7552,
533
+ "last_record": 7679
534
+ },
535
+ {
536
+ "path": "shard_00060.pt",
537
+ "records": 128,
538
+ "tokens": 56169,
539
+ "first_record": 7680,
540
+ "last_record": 7807
541
+ },
542
+ {
543
+ "path": "shard_00061.pt",
544
+ "records": 128,
545
+ "tokens": 63966,
546
+ "first_record": 7808,
547
+ "last_record": 7935
548
+ },
549
+ {
550
+ "path": "shard_00062.pt",
551
+ "records": 128,
552
+ "tokens": 75511,
553
+ "first_record": 7936,
554
+ "last_record": 8063
555
+ },
556
+ {
557
+ "path": "shard_00063.pt",
558
+ "records": 128,
559
+ "tokens": 80704,
560
+ "first_record": 8064,
561
+ "last_record": 8191
562
+ },
563
+ {
564
+ "path": "shard_00064.pt",
565
+ "records": 128,
566
+ "tokens": 75275,
567
+ "first_record": 8192,
568
+ "last_record": 8319
569
+ },
570
+ {
571
+ "path": "shard_00065.pt",
572
+ "records": 128,
573
+ "tokens": 72850,
574
+ "first_record": 8320,
575
+ "last_record": 8447
576
+ },
577
+ {
578
+ "path": "shard_00066.pt",
579
+ "records": 128,
580
+ "tokens": 72281,
581
+ "first_record": 8448,
582
+ "last_record": 8575
583
+ },
584
+ {
585
+ "path": "shard_00067.pt",
586
+ "records": 128,
587
+ "tokens": 74564,
588
+ "first_record": 8576,
589
+ "last_record": 8703
590
+ },
591
+ {
592
+ "path": "shard_00068.pt",
593
+ "records": 128,
594
+ "tokens": 78198,
595
+ "first_record": 8704,
596
+ "last_record": 8831
597
+ },
598
+ {
599
+ "path": "shard_00069.pt",
600
+ "records": 128,
601
+ "tokens": 74688,
602
+ "first_record": 8832,
603
+ "last_record": 8959
604
+ },
605
+ {
606
+ "path": "shard_00070.pt",
607
+ "records": 128,
608
+ "tokens": 73448,
609
+ "first_record": 8960,
610
+ "last_record": 9087
611
+ },
612
+ {
613
+ "path": "shard_00071.pt",
614
+ "records": 128,
615
+ "tokens": 73684,
616
+ "first_record": 9088,
617
+ "last_record": 9215
618
+ },
619
+ {
620
+ "path": "shard_00072.pt",
621
+ "records": 128,
622
+ "tokens": 76147,
623
+ "first_record": 9216,
624
+ "last_record": 9343
625
+ },
626
+ {
627
+ "path": "shard_00073.pt",
628
+ "records": 128,
629
+ "tokens": 73607,
630
+ "first_record": 9344,
631
+ "last_record": 9471
632
+ },
633
+ {
634
+ "path": "shard_00074.pt",
635
+ "records": 128,
636
+ "tokens": 76658,
637
+ "first_record": 9472,
638
+ "last_record": 9599
639
+ },
640
+ {
641
+ "path": "shard_00075.pt",
642
+ "records": 128,
643
+ "tokens": 71052,
644
+ "first_record": 9600,
645
+ "last_record": 9727
646
+ },
647
+ {
648
+ "path": "shard_00076.pt",
649
+ "records": 128,
650
+ "tokens": 71839,
651
+ "first_record": 9728,
652
+ "last_record": 9855
653
+ },
654
+ {
655
+ "path": "shard_00077.pt",
656
+ "records": 128,
657
+ "tokens": 64648,
658
+ "first_record": 9856,
659
+ "last_record": 9983
660
+ },
661
+ {
662
+ "path": "shard_00078.pt",
663
+ "records": 128,
664
+ "tokens": 69436,
665
+ "first_record": 9984,
666
+ "last_record": 10111
667
+ },
668
+ {
669
+ "path": "shard_00079.pt",
670
+ "records": 128,
671
+ "tokens": 72194,
672
+ "first_record": 10112,
673
+ "last_record": 10239
674
+ },
675
+ {
676
+ "path": "shard_00080.pt",
677
+ "records": 128,
678
+ "tokens": 74293,
679
+ "first_record": 10240,
680
+ "last_record": 10367
681
+ },
682
+ {
683
+ "path": "shard_00081.pt",
684
+ "records": 128,
685
+ "tokens": 72618,
686
+ "first_record": 10368,
687
+ "last_record": 10495
688
+ },
689
+ {
690
+ "path": "shard_00082.pt",
691
+ "records": 128,
692
+ "tokens": 74951,
693
+ "first_record": 10496,
694
+ "last_record": 10623
695
+ },
696
+ {
697
+ "path": "shard_00083.pt",
698
+ "records": 128,
699
+ "tokens": 65989,
700
+ "first_record": 10624,
701
+ "last_record": 10751
702
+ },
703
+ {
704
+ "path": "shard_00084.pt",
705
+ "records": 128,
706
+ "tokens": 70265,
707
+ "first_record": 10752,
708
+ "last_record": 10879
709
+ },
710
+ {
711
+ "path": "shard_00085.pt",
712
+ "records": 128,
713
+ "tokens": 72357,
714
+ "first_record": 10880,
715
+ "last_record": 11007
716
+ },
717
+ {
718
+ "path": "shard_00086.pt",
719
+ "records": 128,
720
+ "tokens": 70381,
721
+ "first_record": 11008,
722
+ "last_record": 11135
723
+ },
724
+ {
725
+ "path": "shard_00087.pt",
726
+ "records": 128,
727
+ "tokens": 63801,
728
+ "first_record": 11136,
729
+ "last_record": 11263
730
+ },
731
+ {
732
+ "path": "shard_00088.pt",
733
+ "records": 128,
734
+ "tokens": 72738,
735
+ "first_record": 11264,
736
+ "last_record": 11391
737
+ },
738
+ {
739
+ "path": "shard_00089.pt",
740
+ "records": 128,
741
+ "tokens": 74195,
742
+ "first_record": 11392,
743
+ "last_record": 11519
744
+ },
745
+ {
746
+ "path": "shard_00090.pt",
747
+ "records": 128,
748
+ "tokens": 78693,
749
+ "first_record": 11520,
750
+ "last_record": 11647
751
+ },
752
+ {
753
+ "path": "shard_00091.pt",
754
+ "records": 128,
755
+ "tokens": 75694,
756
+ "first_record": 11648,
757
+ "last_record": 11775
758
+ },
759
+ {
760
+ "path": "shard_00092.pt",
761
+ "records": 128,
762
+ "tokens": 80262,
763
+ "first_record": 11776,
764
+ "last_record": 11903
765
+ },
766
+ {
767
+ "path": "shard_00093.pt",
768
+ "records": 128,
769
+ "tokens": 80246,
770
+ "first_record": 11904,
771
+ "last_record": 12031
772
+ },
773
+ {
774
+ "path": "shard_00094.pt",
775
+ "records": 83,
776
+ "tokens": 46892,
777
+ "first_record": 12032,
778
+ "last_record": 12114
779
+ }
780
+ ],
781
+ "requested_records": 12115,
782
+ "source_records": 12115
783
+ }