Skip to content
GitLab
Menu
Projects
Groups
Snippets
Loading...
Help
Help
Support
Community forum
Keyboard shortcuts
?
Submit feedback
Contribute to GitLab
Sign in / Register
Toggle navigation
Menu
Open sidebar
gaoqiong
lm-evaluation-harness
Commits
5bae76d6
"git@developer.sourcefind.cn:gaoqiong/migraphx.git" did not exist on "026bc3537309902328eaa1ff16fffa9e25532e87"
Commit
5bae76d6
authored
Jul 24, 2024
by
root
Browse files
mmlu-pro/default: aligned with mmlu updates
parent
c1e63555
Changes
18
Hide whitespace changes
Inline
Side-by-side
Showing
18 changed files
with
54 additions
and
26 deletions
+54
-26
lm_eval/tasks/mmlu_pro/default/_mmlu_pro.yaml
lm_eval/tasks/mmlu_pro/default/_mmlu_pro.yaml
+5
-0
lm_eval/tasks/mmlu_pro/default/_mmlu_pro_humanities.yaml
lm_eval/tasks/mmlu_pro/default/_mmlu_pro_humanities.yaml
+9
-0
lm_eval/tasks/mmlu_pro/default/_mmlu_pro_other.yaml
lm_eval/tasks/mmlu_pro/default/_mmlu_pro_other.yaml
+9
-0
lm_eval/tasks/mmlu_pro/default/_mmlu_pro_social_sciences.yaml
...val/tasks/mmlu_pro/default/_mmlu_pro_social_sciences.yaml
+9
-0
lm_eval/tasks/mmlu_pro/default/_mmlu_pro_stem.yaml
lm_eval/tasks/mmlu_pro/default/_mmlu_pro_stem.yaml
+9
-0
lm_eval/tasks/mmlu_pro/default/mmlu_pro_biology.yaml
lm_eval/tasks/mmlu_pro/default/mmlu_pro_biology.yaml
+1
-2
lm_eval/tasks/mmlu_pro/default/mmlu_pro_business.yaml
lm_eval/tasks/mmlu_pro/default/mmlu_pro_business.yaml
+1
-2
lm_eval/tasks/mmlu_pro/default/mmlu_pro_chemistry.yaml
lm_eval/tasks/mmlu_pro/default/mmlu_pro_chemistry.yaml
+1
-2
lm_eval/tasks/mmlu_pro/default/mmlu_pro_computer_science.yaml
...val/tasks/mmlu_pro/default/mmlu_pro_computer_science.yaml
+1
-2
lm_eval/tasks/mmlu_pro/default/mmlu_pro_economics.yaml
lm_eval/tasks/mmlu_pro/default/mmlu_pro_economics.yaml
+1
-2
lm_eval/tasks/mmlu_pro/default/mmlu_pro_engineering.yaml
lm_eval/tasks/mmlu_pro/default/mmlu_pro_engineering.yaml
+1
-2
lm_eval/tasks/mmlu_pro/default/mmlu_pro_history.yaml
lm_eval/tasks/mmlu_pro/default/mmlu_pro_history.yaml
+1
-2
lm_eval/tasks/mmlu_pro/default/mmlu_pro_law.yaml
lm_eval/tasks/mmlu_pro/default/mmlu_pro_law.yaml
+1
-2
lm_eval/tasks/mmlu_pro/default/mmlu_pro_math.yaml
lm_eval/tasks/mmlu_pro/default/mmlu_pro_math.yaml
+1
-2
lm_eval/tasks/mmlu_pro/default/mmlu_pro_miscellaneous.yaml
lm_eval/tasks/mmlu_pro/default/mmlu_pro_miscellaneous.yaml
+1
-2
lm_eval/tasks/mmlu_pro/default/mmlu_pro_philosophy.yaml
lm_eval/tasks/mmlu_pro/default/mmlu_pro_philosophy.yaml
+1
-2
lm_eval/tasks/mmlu_pro/default/mmlu_pro_physics.yaml
lm_eval/tasks/mmlu_pro/default/mmlu_pro_physics.yaml
+1
-2
lm_eval/tasks/mmlu_pro/default/mmlu_pro_psychology.yaml
lm_eval/tasks/mmlu_pro/default/mmlu_pro_psychology.yaml
+1
-2
No files found.
lm_eval/tasks/mmlu_pro/default/_mmlu_pro.yaml
View file @
5bae76d6
...
...
@@ -4,3 +4,8 @@ task:
-
mmlu_pro_other
-
mmlu_pro_social_sciences
-
mmlu_pro_humanities
aggregate_metric_list
:
-
metric
:
acc
weight_by_size
:
True
metadata
:
version
:
1
lm_eval/tasks/mmlu_pro/default/_mmlu_pro_humanities.yaml
0 → 100644
View file @
5bae76d6
group
:
mmlu_pro_humanities
group_alias
:
humanities
task
:
-
mmlu_pro_humanities_tasks
aggregate_metric_list
:
-
metric
:
acc
weight_by_size
:
True
metadata
:
version
:
1
lm_eval/tasks/mmlu_pro/default/_mmlu_pro_other.yaml
0 → 100644
View file @
5bae76d6
group
:
mmlu_pro_other
group_alias
:
other
task
:
-
mmlu_pro_other_tasks
aggregate_metric_list
:
-
metric
:
acc
weight_by_size
:
True
metadata
:
version
:
1
lm_eval/tasks/mmlu_pro/default/_mmlu_pro_social_sciences.yaml
0 → 100644
View file @
5bae76d6
group
:
mmlu_pro_social_sciences
group_alias
:
social sciences
task
:
-
mmlu_pro_social_sciences_tasks
aggregate_metric_list
:
-
metric
:
acc
weight_by_size
:
True
metadata
:
version
:
1
lm_eval/tasks/mmlu_pro/default/_mmlu_pro_stem.yaml
0 → 100644
View file @
5bae76d6
group
:
mmlu_pro_stem
group_alias
:
stem
task
:
-
mmlu_pro_stem_tasks
aggregate_metric_list
:
-
metric
:
acc
weight_by_size
:
True
metadata
:
version
:
1
lm_eval/tasks/mmlu_pro/default/mmlu_pro_biology.yaml
View file @
5bae76d6
"
dataset_name"
:
"
biology"
"
description"
:
"
The
following
are
multiple
choice
questions
(with
answers)
about
biology.
\n\
\n
"
"
group"
:
"
mmlu_pro_stem"
"
group_alias"
:
"
stem"
"
tag"
:
"
mmlu_pro_stem_tasks"
"
include"
:
"
_default_template_yaml"
"
task"
:
"
mmlu_pro_biology"
"
task_alias"
:
"
biology"
lm_eval/tasks/mmlu_pro/default/mmlu_pro_business.yaml
View file @
5bae76d6
"
dataset_name"
:
"
business"
"
description"
:
"
The
following
are
multiple
choice
questions
(with
answers)
about
business.
\n\
\n
"
"
group"
:
"
mmlu_pro_other"
"
group_alias"
:
"
other"
"
tag"
:
"
mmlu_pro_other_tasks"
"
include"
:
"
_default_template_yaml"
"
task"
:
"
mmlu_pro_business"
"
task_alias"
:
"
business"
lm_eval/tasks/mmlu_pro/default/mmlu_pro_chemistry.yaml
View file @
5bae76d6
"
dataset_name"
:
"
math"
"
description"
:
"
The
following
are
multiple
choice
questions
(with
answers)
about
math.
\n\
\n
"
"
group"
:
"
mmlu_pro_stem"
"
group_alias"
:
"
stem"
"
tag"
:
"
mmlu_pro_stem_tasks"
"
include"
:
"
_default_template_yaml"
"
task"
:
"
mmlu_pro_math"
"
task_alias"
:
"
math"
lm_eval/tasks/mmlu_pro/default/mmlu_pro_computer_science.yaml
View file @
5bae76d6
"
dataset_name"
:
"
computer_science"
"
description"
:
"
The
following
are
multiple
choice
questions
(with
answers)
about
computer_science.
\n\
\n
"
"
group"
:
"
mmlu_pro_stem"
"
group_alias"
:
"
stem"
"
tag"
:
"
mmlu_pro_stem_tasks"
"
include"
:
"
_default_template_yaml"
"
task"
:
"
mmlu_pro_computer_science"
"
task_alias"
:
"
computer_science"
lm_eval/tasks/mmlu_pro/default/mmlu_pro_economics.yaml
View file @
5bae76d6
"
dataset_name"
:
"
economics"
"
description"
:
"
The
following
are
multiple
choice
questions
(with
answers)
about
economics.
\n\
\n
"
"
group"
:
"
mmlu_pro_social_sciences"
"
group_alias"
:
"
social_sciences"
"
tag"
:
"
mmlu_pro_social_sciences_tasks"
"
include"
:
"
_default_template_yaml"
"
task"
:
"
mmlu_pro_economics"
"
task_alias"
:
"
economics"
lm_eval/tasks/mmlu_pro/default/mmlu_pro_engineering.yaml
View file @
5bae76d6
"
dataset_name"
:
"
engineering"
"
description"
:
"
The
following
are
multiple
choice
questions
(with
answers)
about
engineering.
\n\
\n
"
"
group"
:
"
mmlu_pro_stem"
"
group_alias"
:
"
stem"
"
tag"
:
"
mmlu_pro_stem_tasks"
"
include"
:
"
_default_template_yaml"
"
task"
:
"
mmlu_pro_engineering"
"
task_alias"
:
"
engineering"
lm_eval/tasks/mmlu_pro/default/mmlu_pro_history.yaml
View file @
5bae76d6
"
dataset_name"
:
"
history"
"
description"
:
"
The
following
are
multiple
choice
questions
(with
answers)
about
history.
\n\
\n
"
"
group"
:
"
mmlu_pro_humanities"
"
group_alias"
:
"
humanities"
"
tag"
:
"
mmlu_pro_humanities_tasks"
"
include"
:
"
_default_template_yaml"
"
task"
:
"
mmlu_pro_history"
"
task_alias"
:
"
history"
lm_eval/tasks/mmlu_pro/default/mmlu_pro_law.yaml
View file @
5bae76d6
"
dataset_name"
:
"
law"
"
description"
:
"
The
following
are
multiple
choice
questions
(with
answers)
about
law.
\n\
\n
"
"
group"
:
"
mmlu_pro_humanities"
"
group_alias"
:
"
humanities"
"
tag"
:
"
mmlu_pro_humanities_tasks"
"
include"
:
"
_default_template_yaml"
"
task"
:
"
mmlu_pro_law"
"
task_alias"
:
"
law"
lm_eval/tasks/mmlu_pro/default/mmlu_pro_math.yaml
View file @
5bae76d6
"
dataset_name"
:
"
math"
"
description"
:
"
The
following
are
multiple
choice
questions
(with
answers)
about
math.
\n\
\n
"
"
group"
:
"
mmlu_pro_stem"
"
group_alias"
:
"
stem"
"
tag"
:
"
mmlu_pro_stem_tasks"
"
include"
:
"
_default_template_yaml"
"
task"
:
"
mmlu_pro_math"
"
task_alias"
:
"
math"
lm_eval/tasks/mmlu_pro/default/mmlu_pro_miscellaneous.yaml
View file @
5bae76d6
"
dataset_name"
:
"
other"
"
description"
:
"
The
following
are
multiple
choice
questions
(with
answers)
about
other.
\n\
\n
"
"
group"
:
"
mmlu_pro_other"
"
group_alias"
:
"
other"
"
tag"
:
"
mmlu_pro_other_tasks"
"
include"
:
"
_default_template_yaml"
"
task"
:
"
mmlu_pro_miscellaneous"
"
task_alias"
:
"
miscellaneous"
lm_eval/tasks/mmlu_pro/default/mmlu_pro_philosophy.yaml
View file @
5bae76d6
"
dataset_name"
:
"
philosophy"
"
description"
:
"
The
following
are
multiple
choice
questions
(with
answers)
about
philosophy.
\n\
\n
"
"
group"
:
"
mmlu_pro_humanities"
"
group_alias"
:
"
humanities"
"
tag"
:
"
mmlu_pro_humanities_tasks"
"
include"
:
"
_default_template_yaml"
"
task"
:
"
mmlu_pro_philosophy"
"
task_alias"
:
"
philosophy"
lm_eval/tasks/mmlu_pro/default/mmlu_pro_physics.yaml
View file @
5bae76d6
"
dataset_name"
:
"
physics"
"
description"
:
"
The
following
are
multiple
choice
questions
(with
answers)
about
physics.
\n\
\n
"
"
group"
:
"
mmlu_pro_stem"
"
group_alias"
:
"
stem"
"
tag"
:
"
mmlu_pro_stem_tasks"
"
include"
:
"
_default_template_yaml"
"
task"
:
"
mmlu_pro_physics"
"
task_alias"
:
"
physics"
lm_eval/tasks/mmlu_pro/default/mmlu_pro_psychology.yaml
View file @
5bae76d6
"
dataset_name"
:
"
psychology"
"
description"
:
"
The
following
are
multiple
choice
questions
(with
answers)
about
psychology.
\n\
\n
"
"
group"
:
"
mmlu_pro_social_sciences"
"
group_alias"
:
"
social_sciences"
"
tag"
:
"
mmlu_pro_social_sciences_tasks"
"
include"
:
"
_default_template_yaml"
"
task"
:
"
mmlu_pro_psychology"
"
task_alias"
:
"
psychology"
Write
Preview
Markdown
is supported
0%
Try again
or
attach a new file
.
Attach a file
Cancel
You are about to add
0
people
to the discussion. Proceed with caution.
Finish editing this message first!
Cancel
Please
register
or
sign in
to comment