AlexWortega commited on
Commit
3d5cbef
·
verified ·
1 Parent(s): 9cba221

add results/latent_mlp_35b.json

Browse files
Files changed (1) hide show
  1. results/latent_mlp_35b.json +114 -0
results/latent_mlp_35b.json ADDED
@@ -0,0 +1,114 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "args": {
3
+ "cmd": "train",
4
+ "latents": "/mnt/qwen_nli_ckpt/latents_35b",
5
+ "out": "results/latent_mlp_35b.json",
6
+ "tasks": [
7
+ "gpqa",
8
+ "mmlu",
9
+ "arc_easy",
10
+ "arc_challenge",
11
+ "winogrande",
12
+ "chess",
13
+ "hellaswag",
14
+ "gsm8k_mc4",
15
+ "gsm8k_mc10"
16
+ ],
17
+ "eps": 0.1,
18
+ "use_nli": false,
19
+ "hidden": 512,
20
+ "dropout": 0.1,
21
+ "lr": 0.001,
22
+ "wd": 0.01,
23
+ "bs": 512,
24
+ "epochs": 60,
25
+ "patience": 8,
26
+ "seed": 0,
27
+ "train_dir": null,
28
+ "frac": 1.0,
29
+ "save_dir": "mlp_heads_35b"
30
+ },
31
+ "results": {
32
+ "gpqa": {
33
+ "n_test_q": 198,
34
+ "n_train_q": 250,
35
+ "random": 0.25,
36
+ "nli_rerank": 0.2878787878787879,
37
+ "mlp_per_task": 0.3838383838383838,
38
+ "mlp_per_task_val": 0.52,
39
+ "mlp_joint": 0.4090909090909091
40
+ },
41
+ "mmlu": {
42
+ "n_test_q": 14042,
43
+ "n_train_q": 1816,
44
+ "random": 0.25,
45
+ "nli_rerank": 0.4981484119071357,
46
+ "mlp_per_task": 0.7238997293832787,
47
+ "mlp_per_task_val": 0.7624309392265194,
48
+ "mlp_joint": 0.6977638513032332
49
+ },
50
+ "arc_easy": {
51
+ "n_test_q": 2376,
52
+ "n_train_q": 2251,
53
+ "random": 0.25016133557800224,
54
+ "nli_rerank": 0.7748316498316499,
55
+ "mlp_per_task": 0.9654882154882155,
56
+ "mlp_per_task_val": 0.9688888888888889,
57
+ "mlp_joint": 0.944023569023569
58
+ },
59
+ "arc_challenge": {
60
+ "n_test_q": 1172,
61
+ "n_train_q": 1119,
62
+ "random": 0.25015642775881686,
63
+ "nli_rerank": 0.7005119453924915,
64
+ "mlp_per_task": 0.9172354948805461,
65
+ "mlp_per_task_val": 0.9459459459459459,
66
+ "mlp_joint": 0.8907849829351536
67
+ },
68
+ "winogrande": {
69
+ "n_test_q": 1267,
70
+ "n_train_q": 8000,
71
+ "random": 0.5,
72
+ "nli_rerank": 0.6985003946329913,
73
+ "mlp_per_task": 0.8121546961325967,
74
+ "mlp_per_task_val": 0.95,
75
+ "mlp_joint": 0.8113654301499605
76
+ },
77
+ "chess": {
78
+ "n_test_q": 500,
79
+ "n_train_q": 2000,
80
+ "random": 0.25,
81
+ "nli_rerank": 0.2,
82
+ "mlp_per_task": 0.896,
83
+ "mlp_per_task_val": 0.905,
84
+ "mlp_joint": 0.876
85
+ },
86
+ "hellaswag": {
87
+ "n_test_q": 10042,
88
+ "n_train_q": 8000,
89
+ "random": 0.25,
90
+ "nli_rerank": 0.43596893049193386,
91
+ "mlp_per_task": 0.9428400716988647,
92
+ "mlp_per_task_val": 0.92875,
93
+ "mlp_joint": 0.9214299940250946
94
+ },
95
+ "gsm8k_mc4": {
96
+ "n_test_q": 1319,
97
+ "n_train_q": 7473,
98
+ "random": 0.25,
99
+ "nli_rerank": 0.45489006823351025,
100
+ "mlp_per_task": 0.7232752084912812,
101
+ "mlp_per_task_val": 0.7764390896921017,
102
+ "mlp_joint": 0.6944655041698257
103
+ },
104
+ "gsm8k_mc10": {
105
+ "n_test_q": 1319,
106
+ "n_train_q": 7473,
107
+ "random": 0.09999999999999998,
108
+ "nli_rerank": 0.30856709628506446,
109
+ "mlp_per_task": 0.6034874905231236,
110
+ "mlp_per_task_val": 0.6599732262382865,
111
+ "mlp_joint": 0.577710386656558
112
+ }
113
+ }
114
+ }