Commit
•
e1bb252
1
Parent(s):
7785f02
Create val.sh
Browse files
val.sh
ADDED
@@ -0,0 +1,345 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
#!/bin/bash
|
2 |
+
#SBATCH --job-name=ckpts
|
3 |
+
#SBATCH --ntasks=1 # number of MP tasks
|
4 |
+
#SBATCH --nodes=1
|
5 |
+
#SBATCH --cpus-per-task=40 # number of cores per tasks
|
6 |
+
#SBATCH --hint=nomultithread # we get physical cores not logical
|
7 |
+
#SBATCH --time=20:00:00 # maximum execution time (HH:MM:SS)
|
8 |
+
#SBATCH --output=%x-%j.out # output file name
|
9 |
+
#SBATCH --account=ajs@cpu
|
10 |
+
#SBATCH --partition=cpu_p1
|
11 |
+
|
12 |
+
set -x -e
|
13 |
+
|
14 |
+
source $six_ALL_CCFRWORK/start-tr13f-6B3-ml-t0
|
15 |
+
export HF_DATASETS_OFFLINE=1
|
16 |
+
export TRANSFORMERS_OFFLINE=1
|
17 |
+
conda activate muennighoffmodelconv
|
18 |
+
|
19 |
+
CKPT_PATH=$six_ALL_CCFRSCRATCH/checkpoints/tr13f-6B3-ml-t0/checkpoints/tasky
|
20 |
+
#CKPT_PATH=$six_ALL_CCFRSCRATCH/checkpoints/tr13f-6B3-ml-t0/checkpoints/p31lossseq
|
21 |
+
|
22 |
+
CKPTS=(
|
23 |
+
global_step250
|
24 |
+
global_step500
|
25 |
+
global_step750
|
26 |
+
global_step1000
|
27 |
+
global_step1250
|
28 |
+
)
|
29 |
+
EXAMPLE_CKPT=$six_ALL_CCFRSCRATCH/commun/experiments/muennighoff/bloomckpt/6b3t0/tr13f-6b3-ml-t0-lmtoks341b-t0toks13b-xp3capmixnewcodelonglossseq
|
30 |
+
DUMP_PATH=$six_ALL_CCFRSCRATCH/commun/experiments/muennighoff/bloomckpt/6b3t0
|
31 |
+
OUT_PREFIX=tasky_
|
32 |
+
#OUT_PREFIX=p31lossseq
|
33 |
+
|
34 |
+
TP=1
|
35 |
+
|
36 |
+
### CONVERT ###
|
37 |
+
|
38 |
+
|
39 |
+
for i in {0..6}; do
|
40 |
+
CKPT=${CKPTS[$i]}
|
41 |
+
echo "$i"
|
42 |
+
echo "Running $CKPT"
|
43 |
+
|
44 |
+
OUTPUTCKPT=$DUMP_PATH/"$OUT_PREFIX$CKPT"
|
45 |
+
python $six_ALL_CCFRSCRATCH/commun/experiments/muennighoff/bloomckpt/transformers_clone/src/transformers/models/bloom/convert_bloom_original_checkpoint_to_pytorch.py --pytorch_dump_folder_path $OUTPUTCKPT --bloom_checkpoint_path $CKPT_PATH/$CKPT --pretraining_tp $TP --bloom_config_file $EXAMPLE_CKPT/config.json
|
46 |
+
|
47 |
+
# Copy tokenizer.json etc
|
48 |
+
cp -r $EXAMPLE_CKPT/*.json $OUTPUTCKPT/
|
49 |
+
|
50 |
+
eval_script="./eval_$i.slurm"
|
51 |
+
cat <<EOT > $eval_script
|
52 |
+
#!/bin/bash
|
53 |
+
#SBATCH --job-name=evaluate_t0
|
54 |
+
#SBATCH --nodes=1
|
55 |
+
#SBATCH --ntasks-per-node=1 # crucial - only 1 task per dist per node!
|
56 |
+
#SBATCH --cpus-per-task=8 # number of cores per tasks
|
57 |
+
#SBATCH --hint=nomultithread # we get physical cores not logical
|
58 |
+
#SBATCH --gres=gpu:1 # number of gpus
|
59 |
+
#SBATCH --constraint=a100
|
60 |
+
#SBATCH --time 5:00:00 # maximum execution time (HH:MM:SS)
|
61 |
+
#SBATCH --output=%x-%j.out # output file name
|
62 |
+
#SBATCH --account=ajs@a100
|
63 |
+
#SBATCH --array=0-168
|
64 |
+
|
65 |
+
set -x -e
|
66 |
+
|
67 |
+
source $six_ALL_CCFRWORK/start-py38-pt111
|
68 |
+
conda activate thomas_t_zero_evaluation
|
69 |
+
|
70 |
+
CHECKPOINT_PATH=$OUTPUTCKPT
|
71 |
+
|
72 |
+
WORKDIR=/gpfswork/rech/six/commun/code/tr13f-6B3-ml-t0
|
73 |
+
pushd "\$WORKDIR"
|
74 |
+
OUTPUT_DIR="\$CHECKPOINT_PATH/evaluation"
|
75 |
+
mkdir -p "\$OUTPUT_DIR"
|
76 |
+
|
77 |
+
# Validation
|
78 |
+
DATASETS_AND_CONFIGS_VAL=(
|
79 |
+
head_qa,en,en,"multiple_choice_q_and_a_index_with_context_en",validation
|
80 |
+
head_qa,en,en,"multiple_choice_q_and_a_en",validation
|
81 |
+
head_qa,en,en,"multiple_choice_q_and_a_index_en",validation
|
82 |
+
head_qa,en,en,"multiple_choice_a_and_q_with_context_en",validation
|
83 |
+
head_qa,en,en,"multiple_choice_a_and_q_en",validation
|
84 |
+
head_qa,es,en,"multiple_choice_q_and_a_index_with_context_en",validation
|
85 |
+
head_qa,es,en,"multiple_choice_q_and_a_en",validation
|
86 |
+
head_qa,es,en,"multiple_choice_q_and_a_index_en",validation
|
87 |
+
head_qa,es,en,"multiple_choice_a_and_q_with_context_en",validation
|
88 |
+
head_qa,es,en,"multiple_choice_a_and_q_en",validation
|
89 |
+
climate_fever,None,None,"first_evidence_and_claim_itemization",test
|
90 |
+
climate_fever,None,None,"claim_and_all_supporting_evidences",test
|
91 |
+
climate_fever,None,None,"fifth_evidence_and_claim_itemization",test
|
92 |
+
climate_fever,None,None,"third_evidence_claim_pair",test
|
93 |
+
climate_fever,None,None,"second_evidence_and_claim_itemization",test
|
94 |
+
codah,codah,None,"interrogative_instruction_after_sentence_and_choices",train
|
95 |
+
codah,codah,None,"affirmative_instruction_before_sentence_and_choices",train
|
96 |
+
codah,codah,None,"affirmative_instruction_after_sentence_and_choices",train
|
97 |
+
aqua_rat,raw,None,"select_the_best_option",validation
|
98 |
+
aqua_rat,raw,None,"answer_quiz",validation
|
99 |
+
aqua_rat,raw,None,"Answer questions from options",validation
|
100 |
+
commonsense_qa,None,None,"answer_given_question_without_options",validation
|
101 |
+
commonsense_qa,None,None,"question_answering",validation
|
102 |
+
commonsense_qa,None,None,"most_suitable_answer",validation
|
103 |
+
amazon_reviews_multi,en,en,"prompt_title_to_star",validation
|
104 |
+
amazon_reviews_multi,en,en,"prompt_review_to_star",validation
|
105 |
+
amazon_reviews_multi,en,en,"prompt_body_title_to_star",validation
|
106 |
+
amazon_reviews_multi,zh,en,"prompt_title_to_star",validation
|
107 |
+
amazon_reviews_multi,zh,en,"prompt_review_to_star",validation
|
108 |
+
amazon_reviews_multi,zh,en,"prompt_body_title_to_star",validation
|
109 |
+
amazon_reviews_multi,fr,en,"prompt_title_to_star",validation
|
110 |
+
amazon_reviews_multi,fr,en,"prompt_review_to_star",validation
|
111 |
+
amazon_reviews_multi,fr,en,"prompt_body_title_to_star",validation
|
112 |
+
amazon_reviews_multi,es,en,"prompt_title_to_star",validation
|
113 |
+
amazon_reviews_multi,es,en,"prompt_review_to_star",validation
|
114 |
+
amazon_reviews_multi,es,en,"prompt_body_title_to_star",validation
|
115 |
+
art,None,None,"choose_hypothesis_options",validation
|
116 |
+
art,None,None,"choose_hypothesis_believable",validation
|
117 |
+
art,None,None,"choose_hypothesis",validation
|
118 |
+
art,None,None,"choose_hypothesis_desc",validation
|
119 |
+
art,None,None,"choose_hypothesis_likely",validation
|
120 |
+
banking77,None,None,"help_page_topic",test
|
121 |
+
banking77,None,None,"direct_to_which_department",test
|
122 |
+
banking77,None,None,"rephrase_as_banking_term",test
|
123 |
+
blbooksgenre,title_genre_classifiction,None,"multi-choice",train
|
124 |
+
blbooksgenre,title_genre_classifiction,None,"premise_context_first",train
|
125 |
+
blbooksgenre,title_genre_classifiction,None,"classify",train
|
126 |
+
blimp,adjunct_island,None,"grammatical_between_1_2",train
|
127 |
+
blimp,adjunct_island,None,"grammatical_between_A_B",train
|
128 |
+
blimp,adjunct_island,None,"grammatical_which_one_1_2",train
|
129 |
+
blimp,adjunct_island,None,"single_sentence_bad_yes_no",train
|
130 |
+
blimp,adjunct_island,None,"single_sentence_good_yes_no",train
|
131 |
+
conv_ai_3,None,None,"clarification_needed",validation
|
132 |
+
conv_ai_3,None,None,"score_give_number",validation
|
133 |
+
conv_ai_3,None,None,"ambiguous",validation
|
134 |
+
conv_ai_3,None,None,"directly_answer",validation
|
135 |
+
conv_ai_3,None,None,"score_how_much",validation
|
136 |
+
craigslist_bargains,None,None,"good deal for seller no list price implicit",validation
|
137 |
+
craigslist_bargains,None,None,"good deal for seller no list price",validation
|
138 |
+
craigslist_bargains,None,None,"good deal for seller",validation
|
139 |
+
craigslist_bargains,None,None,"best deal",validation
|
140 |
+
ecthr_cases,alleged-violation-prediction,None,"implicit_advice_number",validation
|
141 |
+
ecthr_cases,alleged-violation-prediction,None,"ecthr_alleged_articles_declaration_at_end",validation
|
142 |
+
ecthr_cases,alleged-violation-prediction,None,"ecthr_alleged_articles_question_at_start",validation
|
143 |
+
ecthr_cases,alleged-violation-prediction,None,"implicit_judgment_paragraph",validation
|
144 |
+
ecthr_cases,alleged-violation-prediction,None,"confirm number of violated articles",validation
|
145 |
+
emo,None,None,"persons_describe",validation
|
146 |
+
emo,None,None,"final_message",validation
|
147 |
+
emo,None,None,"what_emotion_do_you_think",validation
|
148 |
+
emo,None,None,"emotional_state",validation
|
149 |
+
emo,None,None,"dialogue_between",validation
|
150 |
+
emotion,None,None,"choose_the_best_emotion_label",test
|
151 |
+
emotion,None,None,"reply_with_emoation_label",test
|
152 |
+
emotion,None,None,"answer_with_class_label",test
|
153 |
+
emotion,None,None,"answer_question_with_emotion_label",test
|
154 |
+
financial_phrasebank,sentences_allagree,None,"share_price_option",train
|
155 |
+
financial_phrasebank,sentences_allagree,None,"sentiment",train
|
156 |
+
financial_phrasebank,sentences_allagree,None,"word_comes_to_mind",train
|
157 |
+
financial_phrasebank,sentences_allagree,None,"complementary_industries",train
|
158 |
+
financial_phrasebank,sentences_allagree,None,"bullish_neutral_bearish",train
|
159 |
+
glue,cola,None,"Make sense yes no",validation
|
160 |
+
glue,cola,None,"is_this_correct",validation
|
161 |
+
glue,cola,None,"editing",validation
|
162 |
+
glue,cola,None,"Following sentence acceptable",validation
|
163 |
+
glue,cola,None,"Previous sentence acceptable",validation
|
164 |
+
glue,sst2,None,"positive negative after",validation
|
165 |
+
glue,sst2,None,"review",validation
|
166 |
+
glue,sst2,None,"said",validation
|
167 |
+
glue,sst2,None,"following positive negative",validation
|
168 |
+
glue,sst2,None,"happy or mad",validation
|
169 |
+
health_fact,None,None,"claim_veracity_classification_after_reading_I_believe",validation
|
170 |
+
health_fact,None,None,"claim_explanation_classification",validation
|
171 |
+
health_fact,None,None,"claim_veracity_classification_tell_me",validation
|
172 |
+
hlgd,None,None,"is_same_event_with_time_interrogative_related",validation
|
173 |
+
hlgd,None,None,"is_same_event_interrogative_talk",validation
|
174 |
+
hlgd,None,None,"is_same_event_with_time_interrogative_talk",validation
|
175 |
+
hlgd,None,None,"is_same_event_refer",validation
|
176 |
+
hlgd,None,None,"is_same_event_editor_asks",validation
|
177 |
+
hyperpartisan_news_detection,byarticle,None,"consider_does_it_follow_a_hyperpartisan_argumentation",train
|
178 |
+
hyperpartisan_news_detection,byarticle,None,"follows_hyperpartisan_argumentation",train
|
179 |
+
hyperpartisan_news_detection,byarticle,None,"consume_with_caution",train
|
180 |
+
hyperpartisan_news_detection,byarticle,None,"extreme_left_wing_or_right_wing",train
|
181 |
+
hyperpartisan_news_detection,byarticle,None,"consider_it_exhibits_extreme_one_sidedness",train
|
182 |
+
liar,None,None,"Given statement guess category",validation
|
183 |
+
lince,sa_spaeng,None,"original poster expressed sentiment",validation
|
184 |
+
lince,sa_spaeng,None,"sentiment trying to express",validation
|
185 |
+
lince,sa_spaeng,None,"express sentiment",validation
|
186 |
+
lince,sa_spaeng,None,"negation template",validation
|
187 |
+
lince,sa_spaeng,None,"the author seem",validation
|
188 |
+
math_qa,None,None,"choose_correct_og",test
|
189 |
+
math_qa,None,None,"pick_the_correct",test
|
190 |
+
math_qa,None,None,"first_choice_then_problem",test
|
191 |
+
math_qa,None,None,"problem_set_type",test
|
192 |
+
math_qa,None,None,"gre_problem",test
|
193 |
+
movie_rationales,None,None,"Standard binary sentiment analysis",validation
|
194 |
+
movie_rationales,None,None,"Evidences sentiment classification",validation
|
195 |
+
movie_rationales,None,None,"Evidences + review",validation
|
196 |
+
movie_rationales,None,None,"Generate evidences and sentiment",validation
|
197 |
+
mwsc,None,None,"in-the-sentence-question-first",validation
|
198 |
+
mwsc,None,None,"what-think",validation
|
199 |
+
mwsc,None,None,"in-the-sentence",validation
|
200 |
+
mwsc,None,None,"options-or",validation
|
201 |
+
mwsc,None,None,"is-correct",validation
|
202 |
+
poem_sentiment,None,None,"positive_or_negative_sentiment_variation_2",validation
|
203 |
+
poem_sentiment,None,None,"question_answer_format",validation
|
204 |
+
poem_sentiment,None,None,"guess_sentiment_without_options_variation_1",validation
|
205 |
+
poem_sentiment,None,None,"positive_or_negative_sentiment_variation_1",validation
|
206 |
+
poem_sentiment,None,None,"most_appropriate_sentiment",validation
|
207 |
+
onestop_english,None,None,"esl_context",train
|
208 |
+
onestop_english,None,None,"ara_context",train
|
209 |
+
onestop_english,None,None,"determine_reading_level_from_the_first_three_sentences",train
|
210 |
+
onestop_english,None,None,"esl_variation",train
|
211 |
+
onestop_english,None,None,"assess",train
|
212 |
+
pubmed_qa,pqa_labeled,None,"Long Answer to Final Decision",train
|
213 |
+
pubmed_qa,pqa_labeled,None,"Question Answering (Short)",train
|
214 |
+
riddle_sense,None,None,"most_suitable_answer",validation
|
215 |
+
riddle_sense,None,None,"answer_given_question_without_options",validation
|
216 |
+
riddle_sense,None,None,"question_to_answer_index",validation
|
217 |
+
riddle_sense,None,None,"question_answering",validation
|
218 |
+
scicite,None,None,"Classify intent w/section (select choice)",validation
|
219 |
+
scicite,None,None,"Classify intent (choices first)",validation
|
220 |
+
scicite,None,None,"Classify intent (select choice)",validation
|
221 |
+
scicite,None,None,"Classify intent",validation
|
222 |
+
scicite,None,None,"can_describe",validation
|
223 |
+
selqa,answer_selection_analysis,None,"is-he-talking-about",validation
|
224 |
+
selqa,answer_selection_analysis,None,"would-make-sense-qu-rand",validation
|
225 |
+
selqa,answer_selection_analysis,None,"make-sense-rand",validation
|
226 |
+
selqa,answer_selection_analysis,None,"which-answer-1st-vs-random",validation
|
227 |
+
snips_built_in_intents,None,None,"voice_intent",train
|
228 |
+
snips_built_in_intents,None,None,"categorize_query",train
|
229 |
+
snips_built_in_intents,None,None,"intent_query",train
|
230 |
+
snips_built_in_intents,None,None,"categorize_query_brief",train
|
231 |
+
snips_built_in_intents,None,None,"query_intent",train
|
232 |
+
)
|
233 |
+
|
234 |
+
DATASET_AND_CONFIG="\${DATASETS_AND_CONFIGS_VAL[\$SLURM_ARRAY_TASK_ID]}"
|
235 |
+
echo "\$ARGUMENT"
|
236 |
+
|
237 |
+
# Run T0 evaluation
|
238 |
+
# For PrefixLM add --prefixlm
|
239 |
+
IFS=',' read dataset_name dataset_config_name template_config_name template_name split <<< "\${DATASET_AND_CONFIG}"
|
240 |
+
python t-zero/evaluation/run_eval.py \
|
241 |
+
--dataset_name "\$dataset_name" \
|
242 |
+
--dataset_config_name "\$dataset_config_name" \
|
243 |
+
--template_config_name "\$template_config_name" \
|
244 |
+
--template_name "\$template_name" \
|
245 |
+
--split "\$split" \
|
246 |
+
--model_name_or_path "\$CHECKPOINT_PATH" \
|
247 |
+
--output_dir "\$OUTPUT_DIR" \
|
248 |
+
--per_device_eval_batch_size 4 \
|
249 |
+
--max_length 2048 \
|
250 |
+
--dtype float16
|
251 |
+
EOT
|
252 |
+
|
253 |
+
sbatch $eval_script
|
254 |
+
|
255 |
+
|
256 |
+
lm_eval_script="./lm_eval_$i.slurm"
|
257 |
+
cat <<EOT > $lm_eval_script
|
258 |
+
#!/bin/bash
|
259 |
+
#SBATCH --job-name=lmeval
|
260 |
+
#SBATCH --nodes=1
|
261 |
+
#SBATCH --ntasks-per-node=1 # crucial - only 1 task per dist per node!
|
262 |
+
#SBATCH --cpus-per-task=8 # number of cores per tasks
|
263 |
+
#SBATCH --hint=nomultithread # we get physical cores not logical
|
264 |
+
#SBATCH --gres=gpu:1 # number of gpus
|
265 |
+
#SBATCH --constraint=a100
|
266 |
+
#SBATCH --time 20:00:00 # maximum execution time (HH:MM:SS)
|
267 |
+
#SBATCH --output=%x-%j.out # output file name
|
268 |
+
#SBATCH --account=ajs@a100
|
269 |
+
#SBATCH --array=0-12
|
270 |
+
|
271 |
+
set -x -e
|
272 |
+
|
273 |
+
source $six_ALL_CCFRWORK/start-tr13f-6B3-ml-t0
|
274 |
+
conda activate muennighofflmevalgen
|
275 |
+
|
276 |
+
echo "START TIME: $(date)"
|
277 |
+
|
278 |
+
# defining the right environment variables
|
279 |
+
export TRANSFORMERS_CACHE=$six_ALL_CCFRWORK/models
|
280 |
+
export HF_DATASETS_CACHE=$six_ALL_CCFRWORK/datasets
|
281 |
+
export HF_MODULES_CACHE=$six_ALL_CCFRWORK/modules
|
282 |
+
export HF_METRICS_CACHE=$six_ALL_CCFRWORK/metrics
|
283 |
+
export HF_DATASETS_OFFLINE=1
|
284 |
+
export TRANSFORMERS_OFFLINE=1
|
285 |
+
export TOKENIZERS_PARALLELISM=false
|
286 |
+
|
287 |
+
# Converted transformer checkpoint
|
288 |
+
MODEL_CKPT=$OUTPUTCKPT
|
289 |
+
|
290 |
+
cd /gpfsscratch/rech/six/commun/experiments/muennighoff/lm-evaluation-harness
|
291 |
+
|
292 |
+
|
293 |
+
DATASETS_AND_CONFIGS=(
|
294 |
+
wmt14_fr_en,fr-en,"version-en-fr-target"
|
295 |
+
wmt14_fr_en,fr-en,"a_good_translation-en-fr-target"
|
296 |
+
wmt14_fr_en,fr-en,"a_good_translation-en-fr-source+target"
|
297 |
+
wmt14_fr_en,fr-en,"xglm-en-fr-target"
|
298 |
+
wmt14_fr_en,fr-en,"gpt3-en-fr"
|
299 |
+
wmt14_fr_en,fr-en,"version-fr-en-target"
|
300 |
+
wmt14_fr_en,fr-en,"a_good_translation-fr-en-target"
|
301 |
+
wmt14_fr_en,fr-en,"a_good_translation-fr-en-source+target"
|
302 |
+
wmt14_fr_en,fr-en,"xglm-fr-en-target"
|
303 |
+
wmt14_fr_en,fr-en,"gpt3-fr-en"
|
304 |
+
wmt14_hi_en,hi-en,"version-en-hi-target"
|
305 |
+
wmt14_hi_en,hi-en,"a_good_translation-en-hi-target"
|
306 |
+
wmt14_hi_en,hi-en,"a_good_translation-en-hi-source+target"
|
307 |
+
wmt14_hi_en,hi-en,"xglm-en-hi-target"
|
308 |
+
wmt14_hi_en,hi-en,"gpt-3-en-hi-target"
|
309 |
+
wmt14_hi_en,hi-en,"version-hi-en-target"
|
310 |
+
wmt14_hi_en,hi-en,"a_good_translation-hi-en-target"
|
311 |
+
wmt14_hi_en,hi-en,"a_good_translation-hi-en-source+target"
|
312 |
+
wmt14_hi_en,hi-en,"xglm-hi-en-target"
|
313 |
+
wmt14_hi_en,hi-en,"gpt-3-hi-en-target"
|
314 |
+
mlsum_es,"es","layman_summ_es"
|
315 |
+
mlsum_es,"es","palm_prompt"
|
316 |
+
mlsum_es,"es","summarise_this_in_es_few_sentences"
|
317 |
+
)
|
318 |
+
|
319 |
+
DATASET_AND_CONFIG="\${DATASETS_AND_CONFIGS[\$SLURM_ARRAY_TASK_ID]}"
|
320 |
+
echo "\$ARGUMENT"
|
321 |
+
|
322 |
+
IFS=',' read dataset_name lang template_name <<< "\${DATASET_AND_CONFIG}"
|
323 |
+
|
324 |
+
# Use this fork of lm-eval: https://github.com/bigscience-workshop/lm-evaluation-harness/pull/109
|
325 |
+
python main.py \
|
326 |
+
--model_api_name 'hf-causal' \
|
327 |
+
--model_args "pretrained=\$MODEL_CKPT,use_accelerate=True,tokenizer=\$MODEL_CKPT,dtype=float16" \
|
328 |
+
--device cuda \
|
329 |
+
--batch_size 16 \
|
330 |
+
--no_tracking \
|
331 |
+
--task_name "\$dataset_name" \
|
332 |
+
--template_names "\$template_name" \
|
333 |
+
--bootstrap_iters 10 \
|
334 |
+
--limit 3000
|
335 |
+
|
336 |
+
mkdir -p "$OUTPUTCKPT/evaluation/\$dataset_name"
|
337 |
+
mv "outputs/*$CKPT*\$dataset_name*" "$OUTPUTCKPT/evaluation/\$dataset_name/"
|
338 |
+
|
339 |
+
echo "END TIME: $(date)"
|
340 |
+
EOT
|
341 |
+
|
342 |
+
sbatch $lm_eval_script
|
343 |
+
|
344 |
+
|
345 |
+
done
|