karantonis commited on
Commit
e2a6dbe
·
verified ·
1 Parent(s): 4fa7b27

judge-imitation combos: qprior_own

Browse files
This view is limited to 50 files because it contains too many changes.   See raw diff
Files changed (50) hide show
  1. judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/aime/L12_last.npy +3 -0
  2. judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/aime/L12_mean.npy +3 -0
  3. judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/aime/L16_last.npy +3 -0
  4. judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/aime/L16_mean.npy +3 -0
  5. judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/aime/L20_last.npy +3 -0
  6. judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/aime/L20_mean.npy +3 -0
  7. judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/aime/L24_last.npy +3 -0
  8. judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/aime/L24_mean.npy +3 -0
  9. judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/aime/L28_last.npy +3 -0
  10. judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/aime/L28_mean.npy +3 -0
  11. judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/aime/L32_last.npy +3 -0
  12. judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/aime/L32_mean.npy +3 -0
  13. judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/aime/L8_last.npy +3 -0
  14. judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/aime/L8_mean.npy +3 -0
  15. judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/aime/dump.provenance.json +100 -0
  16. judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/dump_aime.log +3 -0
  17. judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/dump_m500.log +3 -0
  18. judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/dump_math_train.log +27 -0
  19. judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/dump_mth.log +9 -0
  20. judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/m500/L12_last.npy +3 -0
  21. judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/m500/L12_mean.npy +3 -0
  22. judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/m500/L16_last.npy +3 -0
  23. judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/m500/L16_mean.npy +3 -0
  24. judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/m500/L20_last.npy +3 -0
  25. judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/m500/L20_mean.npy +3 -0
  26. judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/m500/L24_last.npy +3 -0
  27. judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/m500/L24_mean.npy +3 -0
  28. judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/m500/L28_last.npy +3 -0
  29. judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/m500/L28_mean.npy +3 -0
  30. judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/m500/L32_last.npy +3 -0
  31. judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/m500/L32_mean.npy +3 -0
  32. judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/m500/L8_last.npy +3 -0
  33. judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/m500/L8_mean.npy +3 -0
  34. judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/m500/dump.provenance.json +100 -0
  35. judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/math_train/L12_last.npy +3 -0
  36. judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/math_train/L12_mean.npy +3 -0
  37. judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/math_train/L16_last.npy +3 -0
  38. judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/math_train/L16_mean.npy +3 -0
  39. judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/math_train/L20_last.npy +3 -0
  40. judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/math_train/L20_mean.npy +3 -0
  41. judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/math_train/L24_last.npy +3 -0
  42. judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/math_train/L24_mean.npy +3 -0
  43. judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/math_train/L28_last.npy +3 -0
  44. judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/math_train/L28_mean.npy +3 -0
  45. judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/math_train/L32_last.npy +3 -0
  46. judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/math_train/L32_mean.npy +3 -0
  47. judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/math_train/L8_last.npy +3 -0
  48. judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/math_train/L8_mean.npy +3 -0
  49. judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/math_train/dump.provenance.json +100 -0
  50. judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/mth/L12_last.npy +3 -0
judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/aime/L12_last.npy ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:92fb6e49b30f340839071f6f1b8513d308652af277a69e64d8ff4868b811a75f
3
+ size 491648
judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/aime/L12_mean.npy ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7c057d42d9edac4d470272338db659df34c340df6c0170d9bc77aa32214c0a70
3
+ size 491648
judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/aime/L16_last.npy ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:36f0448c9f965c00622f13846c0b8d8966ba583032195a7092b05baed4878ea2
3
+ size 491648
judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/aime/L16_mean.npy ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:51192ae8c4e5466f2f5733700f14c2c4d6a84230cd89cf31f2b7c04215eb6f8a
3
+ size 491648
judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/aime/L20_last.npy ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7470b16c5d976a20dd63f46e1774e274b538724dcfe8f1c64b799b93c9e91c88
3
+ size 491648
judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/aime/L20_mean.npy ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:cd42f695b40695aef56a8c38107ad68659911000df17f91acb5cfc72bf0d9a58
3
+ size 491648
judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/aime/L24_last.npy ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:820f689886c22acc001b039882edd5834e572cd8582292e5ed6f515f9d5c0729
3
+ size 491648
judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/aime/L24_mean.npy ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c6495030bcf6a8bc2369204f38e4d66538b2a0dea4c722ac4a0d5ecd5e73ab39
3
+ size 491648
judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/aime/L28_last.npy ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b412c59e56c848798de6427bffadc8abf252fa3d8e6ea0c7808fc368c75ab21f
3
+ size 491648
judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/aime/L28_mean.npy ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:49ad2ee12e032ba5eb4fa1dc1518b2cfaee98fb0760f81514468e2d06557b5ac
3
+ size 491648
judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/aime/L32_last.npy ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:dc14589339203748621be11a51196c2d51418ebb3f2d7057fc2f5a90087f29e6
3
+ size 491648
judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/aime/L32_mean.npy ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:69c60064265c0d50fd3254b342f0bfaf6d4cbddeb2fa203e5d27d24c391b33f2
3
+ size 491648
judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/aime/L8_last.npy ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:eb84bcd8330ad17fa0a63956eae3e5c4ba2bcc890391a5d9a2956734dd5cad54
3
+ size 491648
judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/aime/L8_mean.npy ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2994a1f47e751fcd7aba0c2f1872ffab640dc2fdc24d1403f92921ad0862e713
3
+ size 491648
judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/aime/dump.provenance.json ADDED
@@ -0,0 +1,100 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "started_at": "2026-10-26T00:19:23+04:00",
3
+ "host": "ws006600",
4
+ "user": "vlad.smirnov",
5
+ "cwd": "/mnt/data/users/vlad.smirnov/tmp/wt_3fa3116d_own",
6
+ "argv": [
7
+ "/mnt/data/users/vlad.smirnov/tmp/wt_3fa3116d_own/rl_controller/scripts/question_prior.py",
8
+ "dump",
9
+ "--pool-dir",
10
+ "workdir/selector/pool_aime_k8_chat",
11
+ "--model",
12
+ "meta-llama/Llama-3.1-8B-Instruct",
13
+ "--raw",
14
+ "--layers",
15
+ "8",
16
+ "12",
17
+ "16",
18
+ "20",
19
+ "24",
20
+ "28",
21
+ "32",
22
+ "--out",
23
+ "workdir/qprior_own/llama31_8b/aime"
24
+ ],
25
+ "git": {
26
+ "commit": "3fa3116dbacf9db790f59be53eb037b859f4fce7",
27
+ "branch": "HEAD",
28
+ "subject": "Lane OWN: a base model's own MATH-train labels and prefill features, for own-probe vs transferred-probe",
29
+ "dirty": true,
30
+ "dirty_files": [
31
+ "M scripts/judge_imitation_lanes/plabels/laneOWN.sh"
32
+ ]
33
+ },
34
+ "gpus": {
35
+ "visible": "0",
36
+ "devices": [
37
+ {
38
+ "index": "0",
39
+ "name": "NVIDIA RTX 5880 Ada Generation",
40
+ "memory": "49140 MiB",
41
+ "driver": "580.178.04",
42
+ "uuid": "GPU-a15d9536-8962-7d07-d199-3fb785e80c83"
43
+ },
44
+ {
45
+ "index": "1",
46
+ "name": "NVIDIA RTX 5880 Ada Generation",
47
+ "memory": "49140 MiB",
48
+ "driver": "580.178.04",
49
+ "uuid": "GPU-168c29de-060d-b509-1ebd-6be5535c9d6a"
50
+ }
51
+ ]
52
+ },
53
+ "packages": {
54
+ "python": "3.11.15",
55
+ "torch": "2.9.1",
56
+ "vllm": "0.15.1",
57
+ "transformers": "4.57.6",
58
+ "speculators": "0.4.0.dev49",
59
+ "peft": null,
60
+ "accelerate": "1.13.0",
61
+ "numpy": "2.2.6",
62
+ "huggingface_hub": "0.36.2",
63
+ "openai": "2.32.0",
64
+ "wandb": "0.25.1",
65
+ "torch_cuda": "12.8"
66
+ },
67
+ "env": {
68
+ "CUDA_VISIBLE_DEVICES": "0",
69
+ "HF_HOME": "/mnt/data/users/vlad.smirnov/hf_cache",
70
+ "HF_HUB_OFFLINE": "1",
71
+ "TRANSFORMERS_OFFLINE": "1",
72
+ "PYTORCH_CUDA_ALLOC_CONF": "expandable_segments:True"
73
+ },
74
+ "models": [
75
+ {
76
+ "repo": "meta-llama/Llama-3.1-8B-Instruct",
77
+ "revision": "0e9e39f249a16976918f6564b8830bc894c89659",
78
+ "path": "/mnt/data/users/vlad.smirnov/hf_cache/hub/models--meta-llama--Llama-3.1-8B-Instruct/snapshots/0e9e39f249a16976918f6564b8830bc894c89659"
79
+ }
80
+ ],
81
+ "files": {
82
+ "questions": {
83
+ "path": "/mnt/data/users/vlad.smirnov/rl-uhead-ood/workdir/selector/pool_aime_k8_chat/pool_questions.json",
84
+ "bytes": 26788,
85
+ "sha256": "1b5ddb6fc10823368bdd82f269dcdef68f744f458fe9771b1d3abf555caa4d4f"
86
+ }
87
+ },
88
+ "finished_at": "2026-10-26T00:19:23+04:00",
89
+ "wall_secs": 0.0,
90
+ "n": 60,
91
+ "layers": [
92
+ 8,
93
+ 12,
94
+ 16,
95
+ 20,
96
+ 24,
97
+ 28,
98
+ 32
99
+ ]
100
+ }
judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/dump_aime.log ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ `torch_dtype` is deprecated! Use `dtype` instead!
2
+
3
+ [W1026 00:19:23.611483187 AllocatorConfig.cpp:28] Warning: PYTORCH_CUDA_ALLOC_CONF is deprecated, use PYTORCH_ALLOC_CONF instead (function operator())
judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/dump_m500.log ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ `torch_dtype` is deprecated! Use `dtype` instead!
2
+
3
+ [W1026 00:19:15.961921749 AllocatorConfig.cpp:28] Warning: PYTORCH_CUDA_ALLOC_CONF is deprecated, use PYTORCH_ALLOC_CONF instead (function operator())
judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/dump_math_train.log ADDED
@@ -0,0 +1,27 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ `torch_dtype` is deprecated! Use `dtype` instead!
2
+
3
+ [qprior] 200/4869 (5s)
4
+ [qprior] 400/4869 (11s)
5
+ [qprior] 600/4869 (16s)
6
+ [qprior] 800/4869 (21s)
7
+ [qprior] 1000/4869 (26s)
8
+ [qprior] 1200/4869 (32s)
9
+ [qprior] 1400/4869 (37s)
10
+ [qprior] 1600/4869 (42s)
11
+ [qprior] 1800/4869 (48s)
12
+ [qprior] 2000/4869 (53s)
13
+ [qprior] 2200/4869 (59s)
14
+ [qprior] 2400/4869 (64s)
15
+ [qprior] 2600/4869 (69s)
16
+ [qprior] 2800/4869 (74s)
17
+ [qprior] 3000/4869 (80s)
18
+ [qprior] 3200/4869 (85s)
19
+ [qprior] 3400/4869 (90s)
20
+ [qprior] 3600/4869 (96s)
21
+ [qprior] 3800/4869 (101s)
22
+ [qprior] 4000/4869 (107s)
23
+ [qprior] 4200/4869 (112s)
24
+ [qprior] 4400/4869 (118s)
25
+ [qprior] 4600/4869 (123s)
26
+ [qprior] 4800/4869 (128s)
27
+ [W1026 00:18:21.948905071 AllocatorConfig.cpp:28] Warning: PYTORCH_CUDA_ALLOC_CONF is deprecated, use PYTORCH_ALLOC_CONF instead (function operator())
judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/dump_mth.log ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ `torch_dtype` is deprecated! Use `dtype` instead!
2
+
3
+ [qprior] 200/1361 (6s)
4
+ [qprior] 400/1361 (11s)
5
+ [qprior] 600/1361 (16s)
6
+ [qprior] 800/1361 (22s)
7
+ [qprior] 1000/1361 (27s)
8
+ [qprior] 1200/1361 (32s)
9
+ [W1026 00:19:04.055689814 AllocatorConfig.cpp:28] Warning: PYTORCH_CUDA_ALLOC_CONF is deprecated, use PYTORCH_ALLOC_CONF instead (function operator())
judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/m500/L12_last.npy ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:66577328481b0b9f0c8ffaaff27610dec5eb9b01d79607dbd25d0cb319b5cadc
3
+ size 1310848
judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/m500/L12_mean.npy ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:071756fd23c0c1d4794bd96447b1f867ba6ffa5106ba9300d24dd5b81cd69461
3
+ size 1310848
judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/m500/L16_last.npy ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:fbee56c03eba17689d7e39fe68da7856f7d17b93a91ef2a59409cf02a5fefbd1
3
+ size 1310848
judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/m500/L16_mean.npy ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6a4d6da15abab4b4c4939de1f2ef4b2c3d824d47c55d413c47cf504fd3ae3cbc
3
+ size 1310848
judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/m500/L20_last.npy ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:abd9a29c9c5f1dc76eb8f5249a5f0e088655c308abe2d8cfc17b905905f32652
3
+ size 1310848
judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/m500/L20_mean.npy ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:47ba4dca971d06a72ea759c0b20b2ac304088a1e1694aec64931596e46ff88cc
3
+ size 1310848
judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/m500/L24_last.npy ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7db919d45c842c4195660f005950e65ea60de394ab1dabe922aa19d635966285
3
+ size 1310848
judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/m500/L24_mean.npy ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:86c5c08f264f53cac82bc9f7eb5c166faec40c065dacc6019c1e3273d6442a8a
3
+ size 1310848
judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/m500/L28_last.npy ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:40c6b21851bb7dd8898f85418e290ba76480b073d538135d3daad9028faba9ed
3
+ size 1310848
judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/m500/L28_mean.npy ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:cb3cb2adef1f64c5a13a1ce34ef106e07fa53fd96be9746f94fa34acb244d84a
3
+ size 1310848
judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/m500/L32_last.npy ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0c403b25d0f750940c2a353a737449f31d704cd23aecbdcd4c1efeb8123d9875
3
+ size 1310848
judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/m500/L32_mean.npy ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3c52aad926268ca09e7399cd0ea8a8a2fbdb1251e0b05c8986c7d39bba319ab2
3
+ size 1310848
judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/m500/L8_last.npy ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2b1653e102983c2cfb4e85226d2ff797dddb2b8d251c73061e125224d3a1a519
3
+ size 1310848
judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/m500/L8_mean.npy ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8f2881cf72650cbf6ec9d28cd0e3cb7508d238b5c2d73adb84e6a5669b0c0a15
3
+ size 1310848
judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/m500/dump.provenance.json ADDED
@@ -0,0 +1,100 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "started_at": "2026-10-26T00:19:14+04:00",
3
+ "host": "ws006600",
4
+ "user": "vlad.smirnov",
5
+ "cwd": "/mnt/data/users/vlad.smirnov/tmp/wt_3fa3116d_own",
6
+ "argv": [
7
+ "/mnt/data/users/vlad.smirnov/tmp/wt_3fa3116d_own/rl_controller/scripts/question_prior.py",
8
+ "dump",
9
+ "--pool-dir",
10
+ "workdir/selector/pool_m500_k8_chat",
11
+ "--model",
12
+ "meta-llama/Llama-3.1-8B-Instruct",
13
+ "--raw",
14
+ "--layers",
15
+ "8",
16
+ "12",
17
+ "16",
18
+ "20",
19
+ "24",
20
+ "28",
21
+ "32",
22
+ "--out",
23
+ "workdir/qprior_own/llama31_8b/m500"
24
+ ],
25
+ "git": {
26
+ "commit": "3fa3116dbacf9db790f59be53eb037b859f4fce7",
27
+ "branch": "HEAD",
28
+ "subject": "Lane OWN: a base model's own MATH-train labels and prefill features, for own-probe vs transferred-probe",
29
+ "dirty": true,
30
+ "dirty_files": [
31
+ "M scripts/judge_imitation_lanes/plabels/laneOWN.sh"
32
+ ]
33
+ },
34
+ "gpus": {
35
+ "visible": "0",
36
+ "devices": [
37
+ {
38
+ "index": "0",
39
+ "name": "NVIDIA RTX 5880 Ada Generation",
40
+ "memory": "49140 MiB",
41
+ "driver": "580.178.04",
42
+ "uuid": "GPU-a15d9536-8962-7d07-d199-3fb785e80c83"
43
+ },
44
+ {
45
+ "index": "1",
46
+ "name": "NVIDIA RTX 5880 Ada Generation",
47
+ "memory": "49140 MiB",
48
+ "driver": "580.178.04",
49
+ "uuid": "GPU-168c29de-060d-b509-1ebd-6be5535c9d6a"
50
+ }
51
+ ]
52
+ },
53
+ "packages": {
54
+ "python": "3.11.15",
55
+ "torch": "2.9.1",
56
+ "vllm": "0.15.1",
57
+ "transformers": "4.57.6",
58
+ "speculators": "0.4.0.dev49",
59
+ "peft": null,
60
+ "accelerate": "1.13.0",
61
+ "numpy": "2.2.6",
62
+ "huggingface_hub": "0.36.2",
63
+ "openai": "2.32.0",
64
+ "wandb": "0.25.1",
65
+ "torch_cuda": "12.8"
66
+ },
67
+ "env": {
68
+ "CUDA_VISIBLE_DEVICES": "0",
69
+ "HF_HOME": "/mnt/data/users/vlad.smirnov/hf_cache",
70
+ "HF_HUB_OFFLINE": "1",
71
+ "TRANSFORMERS_OFFLINE": "1",
72
+ "PYTORCH_CUDA_ALLOC_CONF": "expandable_segments:True"
73
+ },
74
+ "models": [
75
+ {
76
+ "repo": "meta-llama/Llama-3.1-8B-Instruct",
77
+ "revision": "0e9e39f249a16976918f6564b8830bc894c89659",
78
+ "path": "/mnt/data/users/vlad.smirnov/hf_cache/hub/models--meta-llama--Llama-3.1-8B-Instruct/snapshots/0e9e39f249a16976918f6564b8830bc894c89659"
79
+ }
80
+ ],
81
+ "files": {
82
+ "questions": {
83
+ "path": "/mnt/data/users/vlad.smirnov/rl-uhead-ood/workdir/selector/pool_m500_k8_chat/pool_questions.json",
84
+ "bytes": 37108,
85
+ "sha256": "2b9a3cd99cf3e1988a34d24fe1716dd55ddcd043c9c5e680c1954741c2e96505"
86
+ }
87
+ },
88
+ "finished_at": "2026-10-26T00:19:14+04:00",
89
+ "wall_secs": 0.0,
90
+ "n": 160,
91
+ "layers": [
92
+ 8,
93
+ 12,
94
+ 16,
95
+ 20,
96
+ 24,
97
+ 28,
98
+ 32
99
+ ]
100
+ }
judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/math_train/L12_last.npy ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f1ccbbd9dd21b1cbeb5825d08b42fd2a65c46f46d3826ee3b4a3ccd2c72e3973
3
+ size 39886976
judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/math_train/L12_mean.npy ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1a30de65553b0d079859dcd2763bfbb5defc69a270e0acc57d672e59f00d89b3
3
+ size 39886976
judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/math_train/L16_last.npy ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:447f4ca6f0f254c1ec225e0184c86d698b8585c3bfbd20218cd0857b531ad67e
3
+ size 39886976
judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/math_train/L16_mean.npy ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:29ab1ac6b2cc21ce41969abf6f080b5501db3e8c246ed43a5e5ad126055a4e98
3
+ size 39886976
judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/math_train/L20_last.npy ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:02e36ab22c559ad9f12863444d0b790481d47931fd7c5ee714699a351b7af848
3
+ size 39886976
judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/math_train/L20_mean.npy ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2ab1b69fc7f6a1f6a884acd0441c41171a718c8e1be9a72f7fa075570ced9296
3
+ size 39886976
judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/math_train/L24_last.npy ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:75aa3807c6410ca853c3916534706d6cb0c2a3d75c0c3fa3bce0151cd3b082c2
3
+ size 39886976
judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/math_train/L24_mean.npy ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:74209d12d0b58d4b85cae083ca135571a747903799aefb59f047c00628d1790e
3
+ size 39886976
judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/math_train/L28_last.npy ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:15f762aec43d9c7db5d879bf7d163a727a9d6be96e2ea285a2e9521b6d358308
3
+ size 39886976
judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/math_train/L28_mean.npy ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5374c006039d6405c8dc50c54de60d8e7c4c9b24d78ee7ed8db2be8de815485f
3
+ size 39886976
judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/math_train/L32_last.npy ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:701279928650214f0513d3778789ffea988d53653712c8a148161724c75a8f7d
3
+ size 39886976
judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/math_train/L32_mean.npy ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d79ecef6bca7158346b398426a50817c16524a766ca2e0cdb99cc421ba7a04b5
3
+ size 39886976
judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/math_train/L8_last.npy ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8572af04f63497f824dec28b038105f8e0f4bebf455a33397edf46ae8fe711ea
3
+ size 39886976
judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/math_train/L8_mean.npy ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6d0527ef3fbded49a7be23a4d6b13eaa68b87fe1b175c84c6ebde8a956040bb7
3
+ size 39886976
judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/math_train/dump.provenance.json ADDED
@@ -0,0 +1,100 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "started_at": "2026-10-26T00:18:20+04:00",
3
+ "host": "ws006600",
4
+ "user": "vlad.smirnov",
5
+ "cwd": "/mnt/data/users/vlad.smirnov/tmp/wt_3fa3116d_own",
6
+ "argv": [
7
+ "/mnt/data/users/vlad.smirnov/tmp/wt_3fa3116d_own/rl_controller/scripts/question_prior.py",
8
+ "dump",
9
+ "--pool-dir",
10
+ "workdir/plabels/math_train",
11
+ "--model",
12
+ "meta-llama/Llama-3.1-8B-Instruct",
13
+ "--raw",
14
+ "--layers",
15
+ "8",
16
+ "12",
17
+ "16",
18
+ "20",
19
+ "24",
20
+ "28",
21
+ "32",
22
+ "--out",
23
+ "workdir/qprior_own/llama31_8b/math_train"
24
+ ],
25
+ "git": {
26
+ "commit": "3fa3116dbacf9db790f59be53eb037b859f4fce7",
27
+ "branch": "HEAD",
28
+ "subject": "Lane OWN: a base model's own MATH-train labels and prefill features, for own-probe vs transferred-probe",
29
+ "dirty": true,
30
+ "dirty_files": [
31
+ "M scripts/judge_imitation_lanes/plabels/laneOWN.sh"
32
+ ]
33
+ },
34
+ "gpus": {
35
+ "visible": "0",
36
+ "devices": [
37
+ {
38
+ "index": "0",
39
+ "name": "NVIDIA RTX 5880 Ada Generation",
40
+ "memory": "49140 MiB",
41
+ "driver": "580.178.04",
42
+ "uuid": "GPU-a15d9536-8962-7d07-d199-3fb785e80c83"
43
+ },
44
+ {
45
+ "index": "1",
46
+ "name": "NVIDIA RTX 5880 Ada Generation",
47
+ "memory": "49140 MiB",
48
+ "driver": "580.178.04",
49
+ "uuid": "GPU-168c29de-060d-b509-1ebd-6be5535c9d6a"
50
+ }
51
+ ]
52
+ },
53
+ "packages": {
54
+ "python": "3.11.15",
55
+ "torch": "2.9.1",
56
+ "vllm": "0.15.1",
57
+ "transformers": "4.57.6",
58
+ "speculators": "0.4.0.dev49",
59
+ "peft": null,
60
+ "accelerate": "1.13.0",
61
+ "numpy": "2.2.6",
62
+ "huggingface_hub": "0.36.2",
63
+ "openai": "2.32.0",
64
+ "wandb": "0.25.1",
65
+ "torch_cuda": "12.8"
66
+ },
67
+ "env": {
68
+ "CUDA_VISIBLE_DEVICES": "0",
69
+ "HF_HOME": "/mnt/data/users/vlad.smirnov/hf_cache",
70
+ "HF_HUB_OFFLINE": "1",
71
+ "TRANSFORMERS_OFFLINE": "1",
72
+ "PYTORCH_CUDA_ALLOC_CONF": "expandable_segments:True"
73
+ },
74
+ "models": [
75
+ {
76
+ "repo": "meta-llama/Llama-3.1-8B-Instruct",
77
+ "revision": "0e9e39f249a16976918f6564b8830bc894c89659",
78
+ "path": "/mnt/data/users/vlad.smirnov/hf_cache/hub/models--meta-llama--Llama-3.1-8B-Instruct/snapshots/0e9e39f249a16976918f6564b8830bc894c89659"
79
+ }
80
+ ],
81
+ "files": {
82
+ "questions": {
83
+ "path": "/mnt/data/users/vlad.smirnov/rl-uhead-ood/workdir/plabels/math_train/pool_questions.json",
84
+ "bytes": 1014551,
85
+ "sha256": "b2a60649d9b6c29f149054fa54efef51c985a7294286e8d0ea76a5afe7cbf575"
86
+ }
87
+ },
88
+ "finished_at": "2026-10-26T00:18:20+04:00",
89
+ "wall_secs": 0.1,
90
+ "n": 4869,
91
+ "layers": [
92
+ 8,
93
+ 12,
94
+ 16,
95
+ 20,
96
+ 24,
97
+ 28,
98
+ 32
99
+ ]
100
+ }
judge-imitation/artem4/rl-uhead-combos-workdir/qprior_own/llama31_8b/mth/L12_last.npy ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b11f1dd6db2d3e986ffe5d4bf1d64820786b59da9bca5dbc9ad21103084d6b95
3
+ size 11149440