Updates

Author: Apple Product Line

data/xp3/prepare_xp3_train.py

@@ -1131,9 +1131,6 @@ def write_to_jsonl_hub(ds, split="train"):
 
  for split in dataset_splits:
  for t_name in prompts.all_template_names:
- if not "zhmt" in t_name:
- print(f"Skipping {t_name}")
- continue
  print(f"Running {ds_name}/{subset_name}/{split}/{t_name}")
  if SKIP_PROMPTS.get(prompt_dataset_name, {}).get(split, False):
  if ("all" in SKIP_PROMPTS[prompt_dataset_name][split]) or (t_name in SKIP_PROMPTS[prompt_dataset_name][split]):
@@ -1183,23 +1180,9 @@ def write_to_jsonl_hub(ds, split="train"):
  out_ds.to_json(out_path, orient="records", lines=True, force_ascii=False)
 
 # Testing:
-TRAIN_DATASETS = [
- ('mlqa', 'mlqa.zh.ar'),
- ('mlqa', 'mlqa.zh.vi'),
- ('mlqa', 'mlqa.zh.es'),
- ('mlqa', 'mlqa.zh.en'),
- ('mlqa', 'mlqa.zh.hi'),
- ('paws-x', 'zh'),
- ('clue', 'c3'),
- ('clue', 'cmrc2018'),
- ('clue', 'csl'),
- ('clue', 'drcd'),
- ('clue', 'tnews'),
- ('pasinit/xlwic', "xlwic_en_zh"),
- ('GEM/xlsum', "chinese_simplified"),
- ('GEM/wiki_lingua', 'zh'),
-]
-
+#TRAIN_DATASETS = [
+# ('xquad', 'xquad.ar'),
+#]
 
 #for ds in TRAIN_DATASETS:
 # write_to_jsonl_hub(ds, split="train")

evaluation/results/tr13/tzeroeval/evaluate_t0.slurm

@@ -542,6 +542,24 @@ xcopa,zh,zh,"cause_effect_zhmt",validation
 xcopa,zh,zh,"plausible_alternatives_zhmt",validation 
 )
 
+DATASETS_AND_CONFIGS_ZHHT=(
+Muennighoff/xstory_cloze,zh,zh,"Story Continuation and Options_zhht",validation
+Muennighoff/xstory_cloze,zh,zh,"Answer Given options_zhht",validation
+Muennighoff/xstory_cloze,zh,zh,"Novel Correct Ending_zhht",validation
+Muennighoff/xstory_cloze,zh,zh,"Generate Ending_zhht",validation
+Muennighoff/xstory_cloze,zh,zh,"Choose Story Ending_zhht",validation
+Muennighoff/xwinograd,zh,zh,"underscore refer to_zhht",test
+Muennighoff/xwinograd,zh,zh,"Replace_zhht",test
+Muennighoff/xwinograd,zh,zh,"stand for_zhht",test
+Muennighoff/xwinograd,zh,zh,"does underscore refer to_zhht",test
+Muennighoff/xwinograd,zh,zh,"True or False_zhht",test
+xcopa,zh,zh,"best_option_zhht",validation
+xcopa,zh,zh,"C1 or C2? premise_zhht",validation
+xcopa,zh,zh,"i_am_hesitating_zhht",validation
+xcopa,zh,zh,"cause_effect_zhht",validation
+xcopa,zh,zh,"plausible_alternatives_zhht",validation 
+)
+
 DATASETS_AND_CONFIGS_XNLITR=(
 xnli,ar,ar,"guaranteed/possible/impossible_arht",validation
 xnli,ar,ar,"MNLI crowdsource_arht",validation

train/tr13-mtf/smaller_models/tr13e-760m-mtf-xp3capmixnewcodelonglossseq.slurm

@@ -1,5 +1,5 @@
 #!/bin/bash
-#SBATCH --job-name=xp3capmixnewcodelong
+#SBATCH --job-name=xp3capmixnewcodelong760m
 #SBATCH --qos=qos_gpu-t3
 #SBATCH --nodes=8
 #SBATCH --ntasks-per-node=1 # crucial - only 1 task per dist per node!
@@ -26,7 +26,7 @@ LOGS_PATH=$REPO_PATH/logs/$variant
 mkdir -p $LOGS_PATH
 mkdir -p $TENSORBOARD_PATH
 
-MEGATRON_DEEPSPEED_REPO=/gpfswork/rech/six/commun/code/tr13f-6B3-ml-t0/megdslossseqnew350m/Megatron-DeepSpeed
+MEGATRON_DEEPSPEED_REPO=/gpfswork/rech/six/commun/code/tr13f-6B3-ml-t0/megdslossseqnew/Megatron-DeepSpeed
 cd $MEGATRON_DEEPSPEED_REPO
 
 KILL_SWITCH_PATH=$MEGATRON_DEEPSPEED_REPO/kill-switch-tr13e-350M-mtf