model_id
stringlengths
19
121
model_role
stringclasses
5 values
provenance_family
stringclasses
35 values
parents
stringclasses
328 values
rung
stringclasses
108 values
operator
stringclasses
52 values
experiment
stringclasses
17 values
benchmark
stringclasses
84 values
metric
stringclasses
13 values
value
float64
-5.09
348
chance
float64
0
50
n
float64
273
67k
source_dataset
stringclasses
7 values
notes
stringclasses
92 values
metric_kind
stringclasses
6 values
unit
stringclasses
7 values
ahn1376/Ties___Code-Instruction_Tuned
merged
Llama-2-7B (AIM release)
Llama-2 experts: Code + Instruction_Tuned
aim=0
Ties
aim_published_outcomes
MBPP
accuracy_pct
23.6
0
null
Mergeability-2/aim-activation-informed-merging
without AIM (baseline merge)
accuracy
percent_0_100
ahn1376/Ties___Code-Instruction_Tuned___AIM
merged
Llama-2-7B (AIM release)
Llama-2 experts: Code + Instruction_Tuned
aim=1
Ties
aim_published_outcomes
MBPP
accuracy_pct
24.2
0
null
Mergeability-2/aim-activation-informed-merging
with AIM (activation-informed merging applied post hoc)
accuracy
percent_0_100
ahn1376/Ties___Code-Math
merged
Llama-2-7B (AIM release)
Llama-2 experts: Code + Math
aim=0
Ties
aim_published_outcomes
MBPP
accuracy_pct
26.8
0
null
Mergeability-2/aim-activation-informed-merging
without AIM (baseline merge)
accuracy
percent_0_100
ahn1376/Ties___Code-Math-Instruction_Tuned
merged
Llama-2-7B (AIM release)
Llama-2 experts: Code + Math + Instruction_Tuned
aim=0
Ties
aim_published_outcomes
MBPP
accuracy_pct
29.2
0
null
Mergeability-2/aim-activation-informed-merging
without AIM (baseline merge)
accuracy
percent_0_100
ahn1376/Ties___Code-Math-Instruction_Tuned___AIM
merged
Llama-2-7B (AIM release)
Llama-2 experts: Code + Math + Instruction_Tuned
aim=1
Ties
aim_published_outcomes
MBPP
accuracy_pct
29.2
0
null
Mergeability-2/aim-activation-informed-merging
with AIM (activation-informed merging applied post hoc)
accuracy
percent_0_100
ahn1376/Ties___Code-Math___AIM
merged
Llama-2-7B (AIM release)
Llama-2 experts: Code + Math
aim=1
Ties
aim_published_outcomes
MBPP
accuracy_pct
28.6
0
null
Mergeability-2/aim-activation-informed-merging
with AIM (activation-informed merging applied post hoc)
accuracy
percent_0_100
ahn1376/Ties___Math-Instruction_Tuned
merged
Llama-2-7B (AIM release)
Llama-2 experts: Math + Instruction_Tuned
aim=0
Ties
aim_published_outcomes
MBPP
accuracy_pct
34.6
0
null
Mergeability-2/aim-activation-informed-merging
without AIM (baseline merge)
accuracy
percent_0_100
ahn1376/Ties___Math-Instruction_Tuned___AIM
merged
Llama-2-7B (AIM release)
Llama-2 experts: Math + Instruction_Tuned
aim=1
Ties
aim_published_outcomes
MBPP
accuracy_pct
35
0
null
Mergeability-2/aim-activation-informed-merging
with AIM (activation-informed merging applied post hoc)
accuracy
percent_0_100
ahn1376/WIDEN___Code-Instruction_Tuned
merged
Llama-2-7B (AIM release)
Llama-2 experts: Code + Instruction_Tuned
aim=0
WIDEN
aim_published_outcomes
MBPP
accuracy_pct
35.6
0
null
Mergeability-2/aim-activation-informed-merging
without AIM (baseline merge)
accuracy
percent_0_100
ahn1376/WIDEN___Code-Instruction_Tuned___AIM
merged
Llama-2-7B (AIM release)
Llama-2 experts: Code + Instruction_Tuned
aim=1
WIDEN
aim_published_outcomes
MBPP
accuracy_pct
34.6
0
null
Mergeability-2/aim-activation-informed-merging
with AIM (activation-informed merging applied post hoc)
accuracy
percent_0_100
ahn1376/WIDEN___Code-Math
merged
Llama-2-7B (AIM release)
Llama-2 experts: Code + Math
aim=0
WIDEN
aim_published_outcomes
MBPP
accuracy_pct
29.4
0
null
Mergeability-2/aim-activation-informed-merging
without AIM (baseline merge)
accuracy
percent_0_100
ahn1376/WIDEN___Code-Math-Instruction_Tuned
merged
Llama-2-7B (AIM release)
Llama-2 experts: Code + Math + Instruction_Tuned
aim=0
WIDEN
aim_published_outcomes
MBPP
accuracy_pct
33.2
0
null
Mergeability-2/aim-activation-informed-merging
without AIM (baseline merge)
accuracy
percent_0_100
ahn1376/WIDEN___Code-Math-Instruction_Tuned___AIM
merged
Llama-2-7B (AIM release)
Llama-2 experts: Code + Math + Instruction_Tuned
aim=1
WIDEN
aim_published_outcomes
MBPP
accuracy_pct
32.8
0
null
Mergeability-2/aim-activation-informed-merging
with AIM (activation-informed merging applied post hoc)
accuracy
percent_0_100
ahn1376/WIDEN___Code-Math___AIM
merged
Llama-2-7B (AIM release)
Llama-2 experts: Code + Math
aim=1
WIDEN
aim_published_outcomes
MBPP
accuracy_pct
29.6
0
null
Mergeability-2/aim-activation-informed-merging
with AIM (activation-informed merging applied post hoc)
accuracy
percent_0_100
ahn1376/WIDEN___Math-Instruction_Tuned
merged
Llama-2-7B (AIM release)
Llama-2 experts: Math + Instruction_Tuned
aim=0
WIDEN
aim_published_outcomes
MBPP
accuracy_pct
30.4
0
null
Mergeability-2/aim-activation-informed-merging
without AIM (baseline merge)
accuracy
percent_0_100
ahn1376/WIDEN___Math-Instruction_Tuned___AIM
merged
Llama-2-7B (AIM release)
Llama-2 experts: Math + Instruction_Tuned
aim=1
WIDEN
aim_published_outcomes
MBPP
accuracy_pct
32
0
null
Mergeability-2/aim-activation-informed-merging
with AIM (activation-informed merging applied post hoc)
accuracy
percent_0_100
aim:_base|Base|aim0
reference
Llama-2-7B (AIM release)
null
aim=0
null
aim_published_outcomes
MBPP
accuracy_pct
27.8
0
null
Mergeability-2/aim-activation-informed-merging
base / single-expert reference checkpoint
accuracy
percent_0_100
aim:_base|Code|aim0
reference
Llama-2-7B (AIM release)
null
aim=0
null
aim_published_outcomes
MBPP
accuracy_pct
31.6
0
null
Mergeability-2/aim-activation-informed-merging
base / single-expert reference checkpoint
accuracy
percent_0_100
aim:_base|Instruction_Tuned|aim0
reference
Llama-2-7B (AIM release)
null
aim=0
null
aim_published_outcomes
MBPP
accuracy_pct
34.8
0
null
Mergeability-2/aim-activation-informed-merging
base / single-expert reference checkpoint
accuracy
percent_0_100
aim:_base|Math|aim0
reference
Llama-2-7B (AIM release)
null
aim=0
null
aim_published_outcomes
MBPP
accuracy_pct
27.6
0
null
Mergeability-2/aim-activation-informed-merging
base / single-expert reference checkpoint
accuracy
percent_0_100
ahn1376/DARETaskArithmetic___Code-Instruction_Tuned
merged
Llama-2-7B (AIM release)
Llama-2 experts: Code + Instruction_Tuned
aim=0
DARETaskArithmetic
aim_published_outcomes
MMLU
accuracy_pct
53.53
25
null
Mergeability-2/aim-activation-informed-merging
without AIM (baseline merge)
accuracy
percent_0_100
ahn1376/DARETaskArithmetic___Code-Instruction_Tuned___AIM
merged
Llama-2-7B (AIM release)
Llama-2 experts: Code + Instruction_Tuned
aim=1
DARETaskArithmetic
aim_published_outcomes
MMLU
accuracy_pct
54.18
25
null
Mergeability-2/aim-activation-informed-merging
with AIM (activation-informed merging applied post hoc)
accuracy
percent_0_100
ahn1376/DARETaskArithmetic___Code-Math
merged
Llama-2-7B (AIM release)
Llama-2 experts: Code + Math
aim=0
DARETaskArithmetic
aim_published_outcomes
MMLU
accuracy_pct
51.96
25
null
Mergeability-2/aim-activation-informed-merging
without AIM (baseline merge)
accuracy
percent_0_100
ahn1376/DARETaskArithmetic___Code-Math-Instruction_Tuned
merged
Llama-2-7B (AIM release)
Llama-2 experts: Code + Math + Instruction_Tuned
aim=0
DARETaskArithmetic
aim_published_outcomes
MMLU
accuracy_pct
50.89
25
null
Mergeability-2/aim-activation-informed-merging
without AIM (baseline merge)
accuracy
percent_0_100
ahn1376/DARETaskArithmetic___Code-Math-Instruction_Tuned___AIM
merged
Llama-2-7B (AIM release)
Llama-2 experts: Code + Math + Instruction_Tuned
aim=1
DARETaskArithmetic
aim_published_outcomes
MMLU
accuracy_pct
52.59
25
null
Mergeability-2/aim-activation-informed-merging
with AIM (activation-informed merging applied post hoc)
accuracy
percent_0_100
ahn1376/DARETaskArithmetic___Code-Math___AIM
merged
Llama-2-7B (AIM release)
Llama-2 experts: Code + Math
aim=1
DARETaskArithmetic
aim_published_outcomes
MMLU
accuracy_pct
52.5
25
null
Mergeability-2/aim-activation-informed-merging
with AIM (activation-informed merging applied post hoc)
accuracy
percent_0_100
ahn1376/DARETaskArithmetic___Math-Instruction_Tuned
merged
Llama-2-7B (AIM release)
Llama-2 experts: Math + Instruction_Tuned
aim=0
DARETaskArithmetic
aim_published_outcomes
MMLU
accuracy_pct
51.08
25
null
Mergeability-2/aim-activation-informed-merging
without AIM (baseline merge)
accuracy
percent_0_100
ahn1376/DARETaskArithmetic___Math-Instruction_Tuned___AIM
merged
Llama-2-7B (AIM release)
Llama-2 experts: Math + Instruction_Tuned
aim=1
DARETaskArithmetic
aim_published_outcomes
MMLU
accuracy_pct
52.72
25
null
Mergeability-2/aim-activation-informed-merging
with AIM (activation-informed merging applied post hoc)
accuracy
percent_0_100
ahn1376/DARETies___Code-Instruction_Tuned
merged
Llama-2-7B (AIM release)
Llama-2 experts: Code + Instruction_Tuned
aim=0
DARETies
aim_published_outcomes
MMLU
accuracy_pct
53.4
25
null
Mergeability-2/aim-activation-informed-merging
without AIM (baseline merge)
accuracy
percent_0_100
ahn1376/DARETies___Code-Instruction_Tuned___AIM
merged
Llama-2-7B (AIM release)
Llama-2 experts: Code + Instruction_Tuned
aim=1
DARETies
aim_published_outcomes
MMLU
accuracy_pct
54.02
25
null
Mergeability-2/aim-activation-informed-merging
with AIM (activation-informed merging applied post hoc)
accuracy
percent_0_100
ahn1376/DARETies___Code-Math
merged
Llama-2-7B (AIM release)
Llama-2 experts: Code + Math
aim=0
DARETies
aim_published_outcomes
MMLU
accuracy_pct
51.92
25
null
Mergeability-2/aim-activation-informed-merging
without AIM (baseline merge)
accuracy
percent_0_100
ahn1376/DARETies___Code-Math-Instruction_Tuned
merged
Llama-2-7B (AIM release)
Llama-2 experts: Code + Math + Instruction_Tuned
aim=0
DARETies
aim_published_outcomes
MMLU
accuracy_pct
51.15
25
null
Mergeability-2/aim-activation-informed-merging
without AIM (baseline merge)
accuracy
percent_0_100
ahn1376/DARETies___Code-Math-Instruction_Tuned___AIM
merged
Llama-2-7B (AIM release)
Llama-2 experts: Code + Math + Instruction_Tuned
aim=1
DARETies
aim_published_outcomes
MMLU
accuracy_pct
52.63
25
null
Mergeability-2/aim-activation-informed-merging
with AIM (activation-informed merging applied post hoc)
accuracy
percent_0_100
ahn1376/DARETies___Code-Math___AIM
merged
Llama-2-7B (AIM release)
Llama-2 experts: Code + Math
aim=1
DARETies
aim_published_outcomes
MMLU
accuracy_pct
52.61
25
null
Mergeability-2/aim-activation-informed-merging
with AIM (activation-informed merging applied post hoc)
accuracy
percent_0_100
ahn1376/DARETies___Math-Instruction_Tuned
merged
Llama-2-7B (AIM release)
Llama-2 experts: Math + Instruction_Tuned
aim=0
DARETies
aim_published_outcomes
MMLU
accuracy_pct
51.39
25
null
Mergeability-2/aim-activation-informed-merging
without AIM (baseline merge)
accuracy
percent_0_100
ahn1376/DARETies___Math-Instruction_Tuned___AIM
merged
Llama-2-7B (AIM release)
Llama-2 experts: Math + Instruction_Tuned
aim=1
DARETies
aim_published_outcomes
MMLU
accuracy_pct
52.89
25
null
Mergeability-2/aim-activation-informed-merging
with AIM (activation-informed merging applied post hoc)
accuracy
percent_0_100
ahn1376/TaskArithmetic___Code-Instruction_Tuned
merged
Llama-2-7B (AIM release)
Llama-2 experts: Code + Instruction_Tuned
aim=0
TaskArithmetic
aim_published_outcomes
MMLU
accuracy_pct
53.44
25
null
Mergeability-2/aim-activation-informed-merging
without AIM (baseline merge)
accuracy
percent_0_100
ahn1376/TaskArithmetic___Code-Instruction_Tuned___AIM
merged
Llama-2-7B (AIM release)
Llama-2 experts: Code + Instruction_Tuned
aim=1
TaskArithmetic
aim_published_outcomes
MMLU
accuracy_pct
54.12
25
null
Mergeability-2/aim-activation-informed-merging
with AIM (activation-informed merging applied post hoc)
accuracy
percent_0_100
ahn1376/TaskArithmetic___Code-Math
merged
Llama-2-7B (AIM release)
Llama-2 experts: Code + Math
aim=0
TaskArithmetic
aim_published_outcomes
MMLU
accuracy_pct
52.1
25
null
Mergeability-2/aim-activation-informed-merging
without AIM (baseline merge)
accuracy
percent_0_100
ahn1376/TaskArithmetic___Code-Math-Instruction_Tuned
merged
Llama-2-7B (AIM release)
Llama-2 experts: Code + Math + Instruction_Tuned
aim=0
TaskArithmetic
aim_published_outcomes
MMLU
accuracy_pct
51.2
25
null
Mergeability-2/aim-activation-informed-merging
without AIM (baseline merge)
accuracy
percent_0_100
ahn1376/TaskArithmetic___Code-Math-Instruction_Tuned___AIM
merged
Llama-2-7B (AIM release)
Llama-2 experts: Code + Math + Instruction_Tuned
aim=1
TaskArithmetic
aim_published_outcomes
MMLU
accuracy_pct
52.63
25
null
Mergeability-2/aim-activation-informed-merging
with AIM (activation-informed merging applied post hoc)
accuracy
percent_0_100
ahn1376/TaskArithmetic___Code-Math___AIM
merged
Llama-2-7B (AIM release)
Llama-2 experts: Code + Math
aim=1
TaskArithmetic
aim_published_outcomes
MMLU
accuracy_pct
52.52
25
null
Mergeability-2/aim-activation-informed-merging
with AIM (activation-informed merging applied post hoc)
accuracy
percent_0_100
ahn1376/TaskArithmetic___Math-Instruction_Tuned
merged
Llama-2-7B (AIM release)
Llama-2 experts: Math + Instruction_Tuned
aim=0
TaskArithmetic
aim_published_outcomes
MMLU
accuracy_pct
51.5
25
null
Mergeability-2/aim-activation-informed-merging
without AIM (baseline merge)
accuracy
percent_0_100
ahn1376/TaskArithmetic___Math-Instruction_Tuned___AIM
merged
Llama-2-7B (AIM release)
Llama-2 experts: Math + Instruction_Tuned
aim=1
TaskArithmetic
aim_published_outcomes
MMLU
accuracy_pct
52.83
25
null
Mergeability-2/aim-activation-informed-merging
with AIM (activation-informed merging applied post hoc)
accuracy
percent_0_100
ahn1376/Ties___Code-Instruction_Tuned
merged
Llama-2-7B (AIM release)
Llama-2 experts: Code + Instruction_Tuned
aim=0
Ties
aim_published_outcomes
MMLU
accuracy_pct
52.7
25
null
Mergeability-2/aim-activation-informed-merging
without AIM (baseline merge)
accuracy
percent_0_100
ahn1376/Ties___Code-Instruction_Tuned___AIM
merged
Llama-2-7B (AIM release)
Llama-2 experts: Code + Instruction_Tuned
aim=1
Ties
aim_published_outcomes
MMLU
accuracy_pct
53.15
25
null
Mergeability-2/aim-activation-informed-merging
with AIM (activation-informed merging applied post hoc)
accuracy
percent_0_100
ahn1376/Ties___Code-Math
merged
Llama-2-7B (AIM release)
Llama-2 experts: Code + Math
aim=0
Ties
aim_published_outcomes
MMLU
accuracy_pct
51.86
25
null
Mergeability-2/aim-activation-informed-merging
without AIM (baseline merge)
accuracy
percent_0_100
ahn1376/Ties___Code-Math-Instruction_Tuned
merged
Llama-2-7B (AIM release)
Llama-2 experts: Code + Math + Instruction_Tuned
aim=0
Ties
aim_published_outcomes
MMLU
accuracy_pct
53.97
25
null
Mergeability-2/aim-activation-informed-merging
without AIM (baseline merge)
accuracy
percent_0_100
ahn1376/Ties___Code-Math-Instruction_Tuned___AIM
merged
Llama-2-7B (AIM release)
Llama-2 experts: Code + Math + Instruction_Tuned
aim=1
Ties
aim_published_outcomes
MMLU
accuracy_pct
54.46
25
null
Mergeability-2/aim-activation-informed-merging
with AIM (activation-informed merging applied post hoc)
accuracy
percent_0_100
ahn1376/Ties___Code-Math___AIM
merged
Llama-2-7B (AIM release)
Llama-2 experts: Code + Math
aim=1
Ties
aim_published_outcomes
MMLU
accuracy_pct
52.29
25
null
Mergeability-2/aim-activation-informed-merging
with AIM (activation-informed merging applied post hoc)
accuracy
percent_0_100
ahn1376/Ties___Math-Instruction_Tuned
merged
Llama-2-7B (AIM release)
Llama-2 experts: Math + Instruction_Tuned
aim=0
Ties
aim_published_outcomes
MMLU
accuracy_pct
54.45
25
null
Mergeability-2/aim-activation-informed-merging
without AIM (baseline merge)
accuracy
percent_0_100
ahn1376/Ties___Math-Instruction_Tuned___AIM
merged
Llama-2-7B (AIM release)
Llama-2 experts: Math + Instruction_Tuned
aim=1
Ties
aim_published_outcomes
MMLU
accuracy_pct
54.74
25
null
Mergeability-2/aim-activation-informed-merging
with AIM (activation-informed merging applied post hoc)
accuracy
percent_0_100
ahn1376/WIDEN___Code-Instruction_Tuned
merged
Llama-2-7B (AIM release)
Llama-2 experts: Code + Instruction_Tuned
aim=0
WIDEN
aim_published_outcomes
MMLU
accuracy_pct
54.9
25
null
Mergeability-2/aim-activation-informed-merging
without AIM (baseline merge)
accuracy
percent_0_100
ahn1376/WIDEN___Code-Instruction_Tuned___AIM
merged
Llama-2-7B (AIM release)
Llama-2 experts: Code + Instruction_Tuned
aim=1
WIDEN
aim_published_outcomes
MMLU
accuracy_pct
54.97
25
null
Mergeability-2/aim-activation-informed-merging
with AIM (activation-informed merging applied post hoc)
accuracy
percent_0_100
ahn1376/WIDEN___Code-Math
merged
Llama-2-7B (AIM release)
Llama-2 experts: Code + Math
aim=0
WIDEN
aim_published_outcomes
MMLU
accuracy_pct
53.35
25
null
Mergeability-2/aim-activation-informed-merging
without AIM (baseline merge)
accuracy
percent_0_100
ahn1376/WIDEN___Code-Math-Instruction_Tuned
merged
Llama-2-7B (AIM release)
Llama-2 experts: Code + Math + Instruction_Tuned
aim=0
WIDEN
aim_published_outcomes
MMLU
accuracy_pct
54.58
25
null
Mergeability-2/aim-activation-informed-merging
without AIM (baseline merge)
accuracy
percent_0_100
ahn1376/WIDEN___Code-Math-Instruction_Tuned___AIM
merged
Llama-2-7B (AIM release)
Llama-2 experts: Code + Math + Instruction_Tuned
aim=1
WIDEN
aim_published_outcomes
MMLU
accuracy_pct
54.98
25
null
Mergeability-2/aim-activation-informed-merging
with AIM (activation-informed merging applied post hoc)
accuracy
percent_0_100
ahn1376/WIDEN___Code-Math___AIM
merged
Llama-2-7B (AIM release)
Llama-2 experts: Code + Math
aim=1
WIDEN
aim_published_outcomes
MMLU
accuracy_pct
53.36
25
null
Mergeability-2/aim-activation-informed-merging
with AIM (activation-informed merging applied post hoc)
accuracy
percent_0_100
ahn1376/WIDEN___Math-Instruction_Tuned
merged
Llama-2-7B (AIM release)
Llama-2 experts: Math + Instruction_Tuned
aim=0
WIDEN
aim_published_outcomes
MMLU
accuracy_pct
54.2
25
null
Mergeability-2/aim-activation-informed-merging
without AIM (baseline merge)
accuracy
percent_0_100
ahn1376/WIDEN___Math-Instruction_Tuned___AIM
merged
Llama-2-7B (AIM release)
Llama-2 experts: Math + Instruction_Tuned
aim=1
WIDEN
aim_published_outcomes
MMLU
accuracy_pct
54.69
25
null
Mergeability-2/aim-activation-informed-merging
with AIM (activation-informed merging applied post hoc)
accuracy
percent_0_100
aim:_base|Base|aim0
reference
Llama-2-7B (AIM release)
null
aim=0
null
aim_published_outcomes
MMLU
accuracy_pct
52.18
25
null
Mergeability-2/aim-activation-informed-merging
base / single-expert reference checkpoint
accuracy
percent_0_100
aim:_base|Code|aim0
reference
Llama-2-7B (AIM release)
null
aim=0
null
aim_published_outcomes
MMLU
accuracy_pct
52.91
25
null
Mergeability-2/aim-activation-informed-merging
base / single-expert reference checkpoint
accuracy
percent_0_100
aim:_base|Instruction_Tuned|aim0
reference
Llama-2-7B (AIM release)
null
aim=0
null
aim_published_outcomes
MMLU
accuracy_pct
53.41
25
null
Mergeability-2/aim-activation-informed-merging
base / single-expert reference checkpoint
accuracy
percent_0_100
aim:_base|Math|aim0
reference
Llama-2-7B (AIM release)
null
aim=0
null
aim_published_outcomes
MMLU
accuracy_pct
51.89
25
null
Mergeability-2/aim-activation-informed-merging
base / single-expert reference checkpoint
accuracy
percent_0_100
ahn1376/DARETaskArithmetic___Code-Instruction_Tuned
merged
Llama-2-7B (AIM release)
Llama-2 experts: Code + Instruction_Tuned
aim=0
DARETaskArithmetic
aim_published_outcomes
HV
harmfulness_score
0.27
null
null
Mergeability-2/aim-activation-informed-merging
without AIM (baseline merge) | HV is a harmfulness score, not an accuracy
other
score_0_1
ahn1376/DARETaskArithmetic___Code-Instruction_Tuned___AIM
merged
Llama-2-7B (AIM release)
Llama-2 experts: Code + Instruction_Tuned
aim=1
DARETaskArithmetic
aim_published_outcomes
HV
harmfulness_score
0.28
null
null
Mergeability-2/aim-activation-informed-merging
with AIM (activation-informed merging applied post hoc) | HV is a harmfulness score, not an accuracy
other
score_0_1
ahn1376/DARETaskArithmetic___Code-Math
merged
Llama-2-7B (AIM release)
Llama-2 experts: Code + Math
aim=0
DARETaskArithmetic
aim_published_outcomes
HV
harmfulness_score
0.23
null
null
Mergeability-2/aim-activation-informed-merging
without AIM (baseline merge) | HV is a harmfulness score, not an accuracy
other
score_0_1
ahn1376/DARETaskArithmetic___Code-Math-Instruction_Tuned
merged
Llama-2-7B (AIM release)
Llama-2 experts: Code + Math + Instruction_Tuned
aim=0
DARETaskArithmetic
aim_published_outcomes
HV
harmfulness_score
0.16
null
null
Mergeability-2/aim-activation-informed-merging
without AIM (baseline merge) | HV is a harmfulness score, not an accuracy
other
score_0_1
ahn1376/DARETaskArithmetic___Code-Math-Instruction_Tuned___AIM
merged
Llama-2-7B (AIM release)
Llama-2 experts: Code + Math + Instruction_Tuned
aim=1
DARETaskArithmetic
aim_published_outcomes
HV
harmfulness_score
0.23
null
null
Mergeability-2/aim-activation-informed-merging
with AIM (activation-informed merging applied post hoc) | HV is a harmfulness score, not an accuracy
other
score_0_1
ahn1376/DARETaskArithmetic___Code-Math___AIM
merged
Llama-2-7B (AIM release)
Llama-2 experts: Code + Math
aim=1
DARETaskArithmetic
aim_published_outcomes
HV
harmfulness_score
0.23
null
null
Mergeability-2/aim-activation-informed-merging
with AIM (activation-informed merging applied post hoc) | HV is a harmfulness score, not an accuracy
other
score_0_1
ahn1376/DARETaskArithmetic___Math-Instruction_Tuned
merged
Llama-2-7B (AIM release)
Llama-2 experts: Math + Instruction_Tuned
aim=0
DARETaskArithmetic
aim_published_outcomes
HV
harmfulness_score
0.18
null
null
Mergeability-2/aim-activation-informed-merging
without AIM (baseline merge) | HV is a harmfulness score, not an accuracy
other
score_0_1
ahn1376/DARETaskArithmetic___Math-Instruction_Tuned___AIM
merged
Llama-2-7B (AIM release)
Llama-2 experts: Math + Instruction_Tuned
aim=1
DARETaskArithmetic
aim_published_outcomes
HV
harmfulness_score
0.26
null
null
Mergeability-2/aim-activation-informed-merging
with AIM (activation-informed merging applied post hoc) | HV is a harmfulness score, not an accuracy
other
score_0_1
ahn1376/DARETies___Code-Instruction_Tuned
merged
Llama-2-7B (AIM release)
Llama-2 experts: Code + Instruction_Tuned
aim=0
DARETies
aim_published_outcomes
HV
harmfulness_score
0.28
null
null
Mergeability-2/aim-activation-informed-merging
without AIM (baseline merge) | HV is a harmfulness score, not an accuracy
other
score_0_1
ahn1376/DARETies___Code-Instruction_Tuned___AIM
merged
Llama-2-7B (AIM release)
Llama-2 experts: Code + Instruction_Tuned
aim=1
DARETies
aim_published_outcomes
HV
harmfulness_score
0.29
null
null
Mergeability-2/aim-activation-informed-merging
with AIM (activation-informed merging applied post hoc) | HV is a harmfulness score, not an accuracy
other
score_0_1
ahn1376/DARETies___Code-Math
merged
Llama-2-7B (AIM release)
Llama-2 experts: Code + Math
aim=0
DARETies
aim_published_outcomes
HV
harmfulness_score
0.23
null
null
Mergeability-2/aim-activation-informed-merging
without AIM (baseline merge) | HV is a harmfulness score, not an accuracy
other
score_0_1
ahn1376/DARETies___Code-Math-Instruction_Tuned
merged
Llama-2-7B (AIM release)
Llama-2 experts: Code + Math + Instruction_Tuned
aim=0
DARETies
aim_published_outcomes
HV
harmfulness_score
0.17
null
null
Mergeability-2/aim-activation-informed-merging
without AIM (baseline merge) | HV is a harmfulness score, not an accuracy
other
score_0_1
ahn1376/DARETies___Code-Math-Instruction_Tuned___AIM
merged
Llama-2-7B (AIM release)
Llama-2 experts: Code + Math + Instruction_Tuned
aim=1
DARETies
aim_published_outcomes
HV
harmfulness_score
0.24
null
null
Mergeability-2/aim-activation-informed-merging
with AIM (activation-informed merging applied post hoc) | HV is a harmfulness score, not an accuracy
other
score_0_1
ahn1376/DARETies___Code-Math___AIM
merged
Llama-2-7B (AIM release)
Llama-2 experts: Code + Math
aim=1
DARETies
aim_published_outcomes
HV
harmfulness_score
0.24
null
null
Mergeability-2/aim-activation-informed-merging
with AIM (activation-informed merging applied post hoc) | HV is a harmfulness score, not an accuracy
other
score_0_1
ahn1376/DARETies___Math-Instruction_Tuned
merged
Llama-2-7B (AIM release)
Llama-2 experts: Math + Instruction_Tuned
aim=0
DARETies
aim_published_outcomes
HV
harmfulness_score
0.2
null
null
Mergeability-2/aim-activation-informed-merging
without AIM (baseline merge) | HV is a harmfulness score, not an accuracy
other
score_0_1
ahn1376/DARETies___Math-Instruction_Tuned___AIM
merged
Llama-2-7B (AIM release)
Llama-2 experts: Math + Instruction_Tuned
aim=1
DARETies
aim_published_outcomes
HV
harmfulness_score
0.26
null
null
Mergeability-2/aim-activation-informed-merging
with AIM (activation-informed merging applied post hoc) | HV is a harmfulness score, not an accuracy
other
score_0_1
ahn1376/TaskArithmetic___Code-Instruction_Tuned
merged
Llama-2-7B (AIM release)
Llama-2 experts: Code + Instruction_Tuned
aim=0
TaskArithmetic
aim_published_outcomes
HV
harmfulness_score
0.28
null
null
Mergeability-2/aim-activation-informed-merging
without AIM (baseline merge) | HV is a harmfulness score, not an accuracy
other
score_0_1
ahn1376/TaskArithmetic___Code-Instruction_Tuned___AIM
merged
Llama-2-7B (AIM release)
Llama-2 experts: Code + Instruction_Tuned
aim=1
TaskArithmetic
aim_published_outcomes
HV
harmfulness_score
0.28
null
null
Mergeability-2/aim-activation-informed-merging
with AIM (activation-informed merging applied post hoc) | HV is a harmfulness score, not an accuracy
other
score_0_1
ahn1376/TaskArithmetic___Code-Math
merged
Llama-2-7B (AIM release)
Llama-2 experts: Code + Math
aim=0
TaskArithmetic
aim_published_outcomes
HV
harmfulness_score
0.24
null
null
Mergeability-2/aim-activation-informed-merging
without AIM (baseline merge) | HV is a harmfulness score, not an accuracy
other
score_0_1
ahn1376/TaskArithmetic___Code-Math-Instruction_Tuned
merged
Llama-2-7B (AIM release)
Llama-2 experts: Code + Math + Instruction_Tuned
aim=0
TaskArithmetic
aim_published_outcomes
HV
harmfulness_score
0.16
null
null
Mergeability-2/aim-activation-informed-merging
without AIM (baseline merge) | HV is a harmfulness score, not an accuracy
other
score_0_1
ahn1376/TaskArithmetic___Code-Math-Instruction_Tuned___AIM
merged
Llama-2-7B (AIM release)
Llama-2 experts: Code + Math + Instruction_Tuned
aim=1
TaskArithmetic
aim_published_outcomes
HV
harmfulness_score
0.22
null
null
Mergeability-2/aim-activation-informed-merging
with AIM (activation-informed merging applied post hoc) | HV is a harmfulness score, not an accuracy
other
score_0_1
ahn1376/TaskArithmetic___Code-Math___AIM
merged
Llama-2-7B (AIM release)
Llama-2 experts: Code + Math
aim=1
TaskArithmetic
aim_published_outcomes
HV
harmfulness_score
0.24
null
null
Mergeability-2/aim-activation-informed-merging
with AIM (activation-informed merging applied post hoc) | HV is a harmfulness score, not an accuracy
other
score_0_1
ahn1376/TaskArithmetic___Math-Instruction_Tuned
merged
Llama-2-7B (AIM release)
Llama-2 experts: Math + Instruction_Tuned
aim=0
TaskArithmetic
aim_published_outcomes
HV
harmfulness_score
0.18
null
null
Mergeability-2/aim-activation-informed-merging
without AIM (baseline merge) | HV is a harmfulness score, not an accuracy
other
score_0_1
ahn1376/TaskArithmetic___Math-Instruction_Tuned___AIM
merged
Llama-2-7B (AIM release)
Llama-2 experts: Math + Instruction_Tuned
aim=1
TaskArithmetic
aim_published_outcomes
HV
harmfulness_score
0.24
null
null
Mergeability-2/aim-activation-informed-merging
with AIM (activation-informed merging applied post hoc) | HV is a harmfulness score, not an accuracy
other
score_0_1
ahn1376/Ties___Code-Instruction_Tuned
merged
Llama-2-7B (AIM release)
Llama-2 experts: Code + Instruction_Tuned
aim=0
Ties
aim_published_outcomes
HV
harmfulness_score
0
null
null
Mergeability-2/aim-activation-informed-merging
without AIM (baseline merge) | HV is a harmfulness score, not an accuracy
other
score_0_1
ahn1376/Ties___Code-Instruction_Tuned___AIM
merged
Llama-2-7B (AIM release)
Llama-2 experts: Code + Instruction_Tuned
aim=1
Ties
aim_published_outcomes
HV
harmfulness_score
0.05
null
null
Mergeability-2/aim-activation-informed-merging
with AIM (activation-informed merging applied post hoc) | HV is a harmfulness score, not an accuracy
other
score_0_1
ahn1376/Ties___Code-Math
merged
Llama-2-7B (AIM release)
Llama-2 experts: Code + Math
aim=0
Ties
aim_published_outcomes
HV
harmfulness_score
0.2
null
null
Mergeability-2/aim-activation-informed-merging
without AIM (baseline merge) | HV is a harmfulness score, not an accuracy
other
score_0_1
ahn1376/Ties___Code-Math-Instruction_Tuned
merged
Llama-2-7B (AIM release)
Llama-2 experts: Code + Math + Instruction_Tuned
aim=0
Ties
aim_published_outcomes
HV
harmfulness_score
0.11
null
null
Mergeability-2/aim-activation-informed-merging
without AIM (baseline merge) | HV is a harmfulness score, not an accuracy
other
score_0_1
ahn1376/Ties___Code-Math-Instruction_Tuned___AIM
merged
Llama-2-7B (AIM release)
Llama-2 experts: Code + Math + Instruction_Tuned
aim=1
Ties
aim_published_outcomes
HV
harmfulness_score
0.11
null
null
Mergeability-2/aim-activation-informed-merging
with AIM (activation-informed merging applied post hoc) | HV is a harmfulness score, not an accuracy
other
score_0_1
ahn1376/Ties___Code-Math___AIM
merged
Llama-2-7B (AIM release)
Llama-2 experts: Code + Math
aim=1
Ties
aim_published_outcomes
HV
harmfulness_score
0.23
null
null
Mergeability-2/aim-activation-informed-merging
with AIM (activation-informed merging applied post hoc) | HV is a harmfulness score, not an accuracy
other
score_0_1
ahn1376/Ties___Math-Instruction_Tuned
merged
Llama-2-7B (AIM release)
Llama-2 experts: Math + Instruction_Tuned
aim=0
Ties
aim_published_outcomes
HV
harmfulness_score
0.23
null
null
Mergeability-2/aim-activation-informed-merging
without AIM (baseline merge) | HV is a harmfulness score, not an accuracy
other
score_0_1
ahn1376/Ties___Math-Instruction_Tuned___AIM
merged
Llama-2-7B (AIM release)
Llama-2 experts: Math + Instruction_Tuned
aim=1
Ties
aim_published_outcomes
HV
harmfulness_score
0.25
null
null
Mergeability-2/aim-activation-informed-merging
with AIM (activation-informed merging applied post hoc) | HV is a harmfulness score, not an accuracy
other
score_0_1
ahn1376/WIDEN___Code-Instruction_Tuned
merged
Llama-2-7B (AIM release)
Llama-2 experts: Code + Instruction_Tuned
aim=0
WIDEN
aim_published_outcomes
HV
harmfulness_score
0.27
null
null
Mergeability-2/aim-activation-informed-merging
without AIM (baseline merge) | HV is a harmfulness score, not an accuracy
other
score_0_1
ahn1376/WIDEN___Code-Instruction_Tuned___AIM
merged
Llama-2-7B (AIM release)
Llama-2 experts: Code + Instruction_Tuned
aim=1
WIDEN
aim_published_outcomes
HV
harmfulness_score
0.26
null
null
Mergeability-2/aim-activation-informed-merging
with AIM (activation-informed merging applied post hoc) | HV is a harmfulness score, not an accuracy
other
score_0_1
ahn1376/WIDEN___Code-Math
merged
Llama-2-7B (AIM release)
Llama-2 experts: Code + Math
aim=0
WIDEN
aim_published_outcomes
HV
harmfulness_score
0.24
null
null
Mergeability-2/aim-activation-informed-merging
without AIM (baseline merge) | HV is a harmfulness score, not an accuracy
other
score_0_1
ahn1376/WIDEN___Code-Math-Instruction_Tuned
merged
Llama-2-7B (AIM release)
Llama-2 experts: Code + Math + Instruction_Tuned
aim=0
WIDEN
aim_published_outcomes
HV
harmfulness_score
0.29
null
null
Mergeability-2/aim-activation-informed-merging
without AIM (baseline merge) | HV is a harmfulness score, not an accuracy
other
score_0_1