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
0d03a9f3
Commit
0d03a9f3
authored
Nov 28, 2023
by
lintangsutawika
Browse files
add versions
parent
bf26d979
Changes
125
Hide whitespace changes
Inline
Side-by-side
Showing
20 changed files
with
40 additions
and
0 deletions
+40
-0
lm_eval/tasks/super_glue/rte/default.yaml
lm_eval/tasks/super_glue/rte/default.yaml
+2
-0
lm_eval/tasks/super_glue/rte/t5-prompt.yaml
lm_eval/tasks/super_glue/rte/t5-prompt.yaml
+2
-0
lm_eval/tasks/super_glue/wic/default.yaml
lm_eval/tasks/super_glue/wic/default.yaml
+2
-0
lm_eval/tasks/super_glue/wic/t5-prompt.yaml
lm_eval/tasks/super_glue/wic/t5-prompt.yaml
+2
-0
lm_eval/tasks/super_glue/wsc/t5-prompt.yaml
lm_eval/tasks/super_glue/wsc/t5-prompt.yaml
+2
-0
lm_eval/tasks/swag/swag.yaml
lm_eval/tasks/swag/swag.yaml
+2
-0
lm_eval/tasks/toxigen/toxigen.yaml
lm_eval/tasks/toxigen/toxigen.yaml
+2
-0
lm_eval/tasks/translation/wmt_common_yaml
lm_eval/tasks/translation/wmt_common_yaml
+2
-0
lm_eval/tasks/triviaqa/default.yaml
lm_eval/tasks/triviaqa/default.yaml
+2
-0
lm_eval/tasks/truthfulqa/truthfulqa_gen.yaml
lm_eval/tasks/truthfulqa/truthfulqa_gen.yaml
+2
-0
lm_eval/tasks/truthfulqa/truthfulqa_mc1.yaml
lm_eval/tasks/truthfulqa/truthfulqa_mc1.yaml
+2
-0
lm_eval/tasks/truthfulqa/truthfulqa_mc2.yaml
lm_eval/tasks/truthfulqa/truthfulqa_mc2.yaml
+2
-0
lm_eval/tasks/unscramble/anagrams1.yaml
lm_eval/tasks/unscramble/anagrams1.yaml
+2
-0
lm_eval/tasks/unscramble/anagrams2.yaml
lm_eval/tasks/unscramble/anagrams2.yaml
+2
-0
lm_eval/tasks/unscramble/cycle_letters.yaml
lm_eval/tasks/unscramble/cycle_letters.yaml
+2
-0
lm_eval/tasks/unscramble/random_insertion.yaml
lm_eval/tasks/unscramble/random_insertion.yaml
+2
-0
lm_eval/tasks/unscramble/reversed_words.yaml
lm_eval/tasks/unscramble/reversed_words.yaml
+2
-0
lm_eval/tasks/webqs/webqs.yaml
lm_eval/tasks/webqs/webqs.yaml
+2
-0
lm_eval/tasks/wikitext/wikitext.yaml
lm_eval/tasks/wikitext/wikitext.yaml
+2
-0
lm_eval/tasks/winogrande/default.yaml
lm_eval/tasks/winogrande/default.yaml
+2
-0
No files found.
lm_eval/tasks/super_glue/rte/default.yaml
View file @
0d03a9f3
...
...
@@ -11,3 +11,5 @@ doc_to_target: label
doc_to_choice
:
[
'
True'
,
'
False'
]
metric_list
:
-
metric
:
acc
metadata
:
-
version
:
0.0
\ No newline at end of file
lm_eval/tasks/super_glue/rte/t5-prompt.yaml
View file @
0d03a9f3
...
...
@@ -18,3 +18,5 @@ metric_list:
higher_is_better
:
true
ignore_case
:
true
ignore_punctuation
:
true
metadata
:
-
version
:
0.0
\ No newline at end of file
lm_eval/tasks/super_glue/wic/default.yaml
View file @
0d03a9f3
...
...
@@ -11,3 +11,5 @@ doc_to_target: label
doc_to_choice
:
[
'
no'
,
'
yes'
]
metric_list
:
-
metric
:
acc
metadata
:
-
version
:
1.0
\ No newline at end of file
lm_eval/tasks/super_glue/wic/t5-prompt.yaml
View file @
0d03a9f3
...
...
@@ -18,3 +18,5 @@ metric_list:
higher_is_better
:
true
ignore_case
:
true
ignore_punctuation
:
true
metadata
:
-
version
:
0.0
\ No newline at end of file
lm_eval/tasks/super_glue/wsc/t5-prompt.yaml
View file @
0d03a9f3
...
...
@@ -19,3 +19,5 @@ filter_list:
-
name
:
"
wsc_postprocessor"
filter
:
-
function
:
!function
t5_utils.WSCPostprocess
metadata
:
-
version
:
0.0
\ No newline at end of file
lm_eval/tasks/swag/swag.yaml
View file @
0d03a9f3
...
...
@@ -15,3 +15,5 @@ metric_list:
-
metric
:
acc_norm
aggregation
:
mean
higher_is_better
:
true
metadata
:
-
version
:
1.0
\ No newline at end of file
lm_eval/tasks/toxigen/toxigen.yaml
View file @
0d03a9f3
...
...
@@ -14,3 +14,5 @@ metric_list:
-
metric
:
acc_norm
aggregation
:
mean
higher_is_better
:
true
metadata
:
-
version
:
1.0
\ No newline at end of file
lm_eval/tasks/translation/wmt_common_yaml
View file @
0d03a9f3
...
...
@@ -13,3 +13,5 @@ generation_kwargs:
do_sample: false
temperature: 0.0
repeats: 1
metadata:
- version: 0.0
\ No newline at end of file
lm_eval/tasks/triviaqa/default.yaml
View file @
0d03a9f3
...
...
@@ -27,3 +27,5 @@ metric_list:
higher_is_better
:
true
ignore_case
:
true
ignore_punctuation
:
true
metadata
:
-
version
:
2.0
\ No newline at end of file
lm_eval/tasks/truthfulqa/truthfulqa_gen.yaml
View file @
0d03a9f3
...
...
@@ -75,3 +75,5 @@ metric_list:
-
metric
:
rougeL_diff
aggregation
:
mean
higher_is_better
:
true
metadata
:
-
version
:
2.0
\ No newline at end of file
lm_eval/tasks/truthfulqa/truthfulqa_mc1.yaml
View file @
0d03a9f3
...
...
@@ -32,3 +32,5 @@ metric_list:
-
metric
:
acc
aggregation
:
mean
higher_is_better
:
true
metadata
:
-
version
:
2.0
\ No newline at end of file
lm_eval/tasks/truthfulqa/truthfulqa_mc2.yaml
View file @
0d03a9f3
...
...
@@ -9,3 +9,5 @@ metric_list:
-
metric
:
acc
aggregation
:
mean
higher_is_better
:
true
metadata
:
-
version
:
2.0
\ No newline at end of file
lm_eval/tasks/unscramble/anagrams1.yaml
View file @
0d03a9f3
...
...
@@ -16,3 +16,5 @@ metric_list:
higher_is_better
:
true
ignore_case
:
false
ignore_punctuation
:
false
metadata
:
-
version
:
1.0
\ No newline at end of file
lm_eval/tasks/unscramble/anagrams2.yaml
View file @
0d03a9f3
...
...
@@ -16,3 +16,5 @@ metric_list:
higher_is_better
:
true
ignore_case
:
false
ignore_punctuation
:
false
metadata
:
-
version
:
1.0
\ No newline at end of file
lm_eval/tasks/unscramble/cycle_letters.yaml
View file @
0d03a9f3
...
...
@@ -16,3 +16,5 @@ metric_list:
higher_is_better
:
true
ignore_case
:
false
ignore_punctuation
:
false
metadata
:
-
version
:
1.0
\ No newline at end of file
lm_eval/tasks/unscramble/random_insertion.yaml
View file @
0d03a9f3
...
...
@@ -16,3 +16,5 @@ metric_list:
higher_is_better
:
true
ignore_case
:
false
ignore_punctuation
:
false
metadata
:
-
version
:
1.0
\ No newline at end of file
lm_eval/tasks/unscramble/reversed_words.yaml
View file @
0d03a9f3
...
...
@@ -16,3 +16,5 @@ metric_list:
higher_is_better
:
true
ignore_case
:
false
ignore_punctuation
:
false
metadata
:
-
version
:
1.0
\ No newline at end of file
lm_eval/tasks/webqs/webqs.yaml
View file @
0d03a9f3
...
...
@@ -16,3 +16,5 @@ metric_list:
-
metric
:
exact_match
aggregation
:
mean
higher_is_better
:
true
metadata
:
-
version
:
1.0
\ No newline at end of file
lm_eval/tasks/wikitext/wikitext.yaml
View file @
0d03a9f3
...
...
@@ -14,3 +14,5 @@ metric_list:
-
metric
:
word_perplexity
-
metric
:
byte_perplexity
-
metric
:
bits_per_byte
metadata
:
-
version
:
2.0
\ No newline at end of file
lm_eval/tasks/winogrande/default.yaml
View file @
0d03a9f3
...
...
@@ -13,3 +13,5 @@ metric_list:
-
metric
:
acc
aggregation
:
mean
higher_is_better
:
true
metadata
:
-
version
:
1.0
\ No newline at end of file
Prev
1
2
3
4
5
6
7
Next
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