File size: 1,853 Bytes
5463614
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
out: bench/results/audits
defaults:
  suite: lite
  dtype: bfloat16
  seed: 0
  limit: 30
audits:
- name: null-identity-s0
  kind: null identity
  control:
    kind: identity
    seed: 0
  seed: 0
  base: Qwen/Qwen2.5-0.5B-Instruct
- name: null-identity-s1
  kind: null identity
  control:
    kind: identity
    seed: 1
  seed: 1
  base: Qwen/Qwen2.5-0.5B-Instruct
- name: null-micro-noise-s0
  kind: null micro noise
  control:
    kind: noise
    strength: 0.0001
    seed: 0
  expect: no_change
  seed: 0
  base: Qwen/Qwen2.5-0.5B-Instruct
- name: null-micro-noise-s1
  kind: null micro noise
  control:
    kind: noise
    strength: 0.0001
    seed: 1
  expect: no_change
  seed: 1
  base: Qwen/Qwen2.5-0.5B-Instruct
- name: planted-noise-0.10
  kind: planted noise
  control:
    kind: noise
    strength: 0.1
  base: Qwen/Qwen2.5-0.5B-Instruct
- name: planted-noise-0.40
  kind: planted noise
  control:
    kind: noise
    strength: 0.4
  base: Qwen/Qwen2.5-0.5B-Instruct
- name: planted-prune-0.30
  kind: planted prune
  control:
    kind: prune
    strength: 0.3
  base: Qwen/Qwen2.5-0.5B-Instruct
- name: planted-zero-last-mlp
  kind: planted zero
  control:
    kind: zero
    match: layers\.23\.mlp\.down_proj\.weight$
  base: Qwen/Qwen2.5-0.5B-Instruct
- name: real-shortthink-sft
  kind: sft
  base: Qwen/Qwen3-1.7B
  derivative: AmareshHebbar/ShortThink_Qwen3_1.7B_SFT
  expect: change
- name: real-toolsure-orpo
  kind: orpo
  base: Qwen/Qwen2.5-1.5B-Instruct
  derivative: AmareshHebbar/ToolSure_Qwen2.5_1.5B_ORPO
  expect: change
- name: real-instruct-qwen2.5-0.5b
  kind: instruction tuning
  base: Qwen/Qwen2.5-0.5B
  derivative: Qwen/Qwen2.5-0.5B-Instruct
  expect: change
- name: real-bitnarrow-qwen7b
  kind: weight edit
  base: Qwen/Qwen2.5-7B-Instruct
  derivative: /workspace/bitnarrow/exports/qwen7b_narrow
  expect: change