Created by: sriniiyer
Fix the regex for model loading. Tested with
CUDA_VISIBLE_DEVICES=0,1 FSD=~/data/instruct-opt/prompt_data/allbenchmarks_io_streaming_after_dedup_v5_sorted python -m metaseq_internal.eval.gpt3_eval --tasks opti
--opti-eval-set test_text_completion__ni_v2__task215_rocstories_incorrect_answer_generation__prompt0 --model-name opt_instr_test --nb-few-shot-samples-values 0 --distributed-world-size 2 --fsdp --prediction
s-dump-dir ~/del_clf --n-eval-samples 10
Output:
Loaded model
model_loading_time=1177.1 seconds
model_loading_time_cuda=1177.1 seconds
Using max-tokens=2048
task=opti
eval_set=test_text_completion__ni_v2__task215_rocstories_incorrect_answer_generation__prompt0
eval language=en
train_set=None
train_lang=None
template=optinstructbenchmark
calibration_options=[]
scoring=sum
nb_few_shot_samples=0