Upload 14 files

latest

@@ -0,0 +1 @@
+global_step20000

mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a410c9d6442304a0fbb141405c427517e0cf55f96f5d6192750272c0343c9069
+size 495168939

pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5aa41201cb775f65fe22499dca0cd86ddc1632fb12cc52b76ca8adb254167fa2
+size 495110123

rng_state_0.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8f74e147aacb7ede7afb2141f4aef7ae4b6c0a742c531081fd1f0df0d445d38c
+size 14503

rng_state_1.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:16f601207794489f0fc427d825751ea256b27df0c9f487e03ec34ab0e00df113
+size 14503

special_tokens_map.json

@@ -0,0 +1 @@
+{"eos_token": "</s>", "unk_token": "<unk>", "pad_token": "<pad>", "additional_special_tokens": ["<extra_id_0>", "<extra_id_1>", "<extra_id_2>", "<extra_id_3>", "<extra_id_4>", "<extra_id_5>", "<extra_id_6>", "<extra_id_7>", "<extra_id_8>", "<extra_id_9>", "<extra_id_10>", "<extra_id_11>", "<extra_id_12>", "<extra_id_13>", "<extra_id_14>", "<extra_id_15>", "<extra_id_16>", "<extra_id_17>", "<extra_id_18>", "<extra_id_19>", "<extra_id_20>", "<extra_id_21>", "<extra_id_22>", "<extra_id_23>", "<extra_id_24>", "<extra_id_25>", "<extra_id_26>", "<extra_id_27>", "<extra_id_28>", "<extra_id_29>", "<extra_id_30>", "<extra_id_31>", "<extra_id_32>", "<extra_id_33>", "<extra_id_34>", "<extra_id_35>", "<extra_id_36>", "<extra_id_37>", "<extra_id_38>", "<extra_id_39>", "<extra_id_40>", "<extra_id_41>", "<extra_id_42>", "<extra_id_43>", "<extra_id_44>", "<extra_id_45>", "<extra_id_46>", "<extra_id_47>", "<extra_id_48>", "<extra_id_49>", "<extra_id_50>", "<extra_id_51>", "<extra_id_52>", "<extra_id_53>", "<extra_id_54>", "<extra_id_55>", "<extra_id_56>", "<extra_id_57>", "<extra_id_58>", "<extra_id_59>", "<extra_id_60>", "<extra_id_61>", "<extra_id_62>", "<extra_id_63>", "<extra_id_64>", "<extra_id_65>", "<extra_id_66>", "<extra_id_67>", "<extra_id_68>", "<extra_id_69>", "<extra_id_70>", "<extra_id_71>", "<extra_id_72>", "<extra_id_73>", "<extra_id_74>", "<extra_id_75>", "<extra_id_76>", "<extra_id_77>", "<extra_id_78>", "<extra_id_79>", "<extra_id_80>", "<extra_id_81>", "<extra_id_82>", "<extra_id_83>", "<extra_id_84>", "<extra_id_85>", "<extra_id_86>", "<extra_id_87>", "<extra_id_88>", "<extra_id_89>", "<extra_id_90>", "<extra_id_91>", "<extra_id_92>", "<extra_id_93>", "<extra_id_94>", "<extra_id_95>", "<extra_id_96>", "<extra_id_97>", "<extra_id_98>", "<extra_id_99>"]}

spiece.model

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d60acb128cf7b7f2536e8f38a5b18a05535c9e14c7a355904270e15b0945ea86
+size 791656

tokenizer_config.json

@@ -0,0 +1 @@
+{"eos_token": "</s>", "unk_token": "<unk>", "pad_token": "<pad>", "extra_ids": 100, "additional_special_tokens": ["<extra_id_0>", "<extra_id_1>", "<extra_id_2>", "<extra_id_3>", "<extra_id_4>", "<extra_id_5>", "<extra_id_6>", "<extra_id_7>", "<extra_id_8>", "<extra_id_9>", "<extra_id_10>", "<extra_id_11>", "<extra_id_12>", "<extra_id_13>", "<extra_id_14>", "<extra_id_15>", "<extra_id_16>", "<extra_id_17>", "<extra_id_18>", "<extra_id_19>", "<extra_id_20>", "<extra_id_21>", "<extra_id_22>", "<extra_id_23>", "<extra_id_24>", "<extra_id_25>", "<extra_id_26>", "<extra_id_27>", "<extra_id_28>", "<extra_id_29>", "<extra_id_30>", "<extra_id_31>", "<extra_id_32>", "<extra_id_33>", "<extra_id_34>", "<extra_id_35>", "<extra_id_36>", "<extra_id_37>", "<extra_id_38>", "<extra_id_39>", "<extra_id_40>", "<extra_id_41>", "<extra_id_42>", "<extra_id_43>", "<extra_id_44>", "<extra_id_45>", "<extra_id_46>", "<extra_id_47>", "<extra_id_48>", "<extra_id_49>", "<extra_id_50>", "<extra_id_51>", "<extra_id_52>", "<extra_id_53>", "<extra_id_54>", "<extra_id_55>", "<extra_id_56>", "<extra_id_57>", "<extra_id_58>", "<extra_id_59>", "<extra_id_60>", "<extra_id_61>", "<extra_id_62>", "<extra_id_63>", "<extra_id_64>", "<extra_id_65>", "<extra_id_66>", "<extra_id_67>", "<extra_id_68>", "<extra_id_69>", "<extra_id_70>", "<extra_id_71>", "<extra_id_72>", "<extra_id_73>", "<extra_id_74>", "<extra_id_75>", "<extra_id_76>", "<extra_id_77>", "<extra_id_78>", "<extra_id_79>", "<extra_id_80>", "<extra_id_81>", "<extra_id_82>", "<extra_id_83>", "<extra_id_84>", "<extra_id_85>", "<extra_id_86>", "<extra_id_87>", "<extra_id_88>", "<extra_id_89>", "<extra_id_90>", "<extra_id_91>", "<extra_id_92>", "<extra_id_93>", "<extra_id_94>", "<extra_id_95>", "<extra_id_96>", "<extra_id_97>", "<extra_id_98>", "<extra_id_99>"], "sp_model_kwargs": {}, "model_max_length": 512, "name_or_path": "google/t5-base-lm-adapt", "special_tokens_map_file": "/home/patrick/.cache/huggingface/transformers/76bf19bfedb85afbe644966ca9ab7b0404d753a41bf601115bced39f825ffa9c.c94798918c92ded6aeef2d2f0e666d2cc4145eca1aa6e1336fde07f2e13e2f46", "tokenizer_class": "T5Tokenizer"}

trainer_state.json

@@ -0,0 +1,1056 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 2.0,
+  "global_step": 20000,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.1,
+      "learning_rate": 5e-05,
+      "loss": 1.5652,
+      "step": 1000
+    },
+    {
+      "epoch": 0.1,
+      "eval_exact_match": 38.3413,
+      "eval_exact_match_for_squad-like": 38.3413,
+      "eval_exact_match_for_task000_NewsQA_dev": 23.8367,
+      "eval_exact_match_for_task000_SQuAD_Shifts_amazon_dev": 50.0455,
+      "eval_exact_match_for_task000_SQuAD_Shifts_new_wiki_dev": 52.0534,
+      "eval_exact_match_for_task000_SQuAD_Shifts_nyt_dev": 53.7009,
+      "eval_exact_match_for_task000_SQuAD_Shifts_reddit_dev": 50.0357,
+      "eval_exact_match_for_task000_SQuAD_dev": 55.8009,
+      "eval_exact_match_for_task000_SearchQA_dev": 2.0554,
+      "eval_f1": 52.1049,
+      "eval_f1_for_squad-like": 52.1049,
+      "eval_f1_for_task000_NewsQA_dev": 40.9979,
+      "eval_f1_for_task000_SQuAD_Shifts_amazon_dev": 68.6298,
+      "eval_f1_for_task000_SQuAD_Shifts_new_wiki_dev": 69.9454,
+      "eval_f1_for_task000_SQuAD_Shifts_nyt_dev": 69.6682,
+      "eval_f1_for_task000_SQuAD_Shifts_reddit_dev": 66.9921,
+      "eval_f1_for_task000_SQuAD_dev": 71.031,
+      "eval_f1_for_task000_SearchQA_dev": 6.183,
+      "eval_gen_len": 6.4379,
+      "eval_global_step": 1000,
+      "eval_loss": 1.6005868911743164,
+      "eval_rouge1": 51.5552,
+      "eval_rouge1_for_squad-like": 51.5552,
+      "eval_rouge1_for_task000_NewsQA_dev": 40.9869,
+      "eval_rouge1_for_task000_SQuAD_Shifts_amazon_dev": 68.2955,
+      "eval_rouge1_for_task000_SQuAD_Shifts_new_wiki_dev": 69.008,
+      "eval_rouge1_for_task000_SQuAD_Shifts_nyt_dev": 68.733,
+      "eval_rouge1_for_task000_SQuAD_Shifts_reddit_dev": 65.7541,
+      "eval_rouge1_for_task000_SQuAD_dev": 70.1352,
+      "eval_rouge1_for_task000_SearchQA_dev": 6.3956,
+      "eval_rougeL": 51.4856,
+      "eval_rougeL_for_squad-like": 51.4856,
+      "eval_rougeL_for_task000_NewsQA_dev": 40.8766,
+      "eval_rougeL_for_task000_SQuAD_Shifts_amazon_dev": 68.2039,
+      "eval_rougeL_for_task000_SQuAD_Shifts_new_wiki_dev": 68.8974,
+      "eval_rougeL_for_task000_SQuAD_Shifts_nyt_dev": 68.6842,
+      "eval_rougeL_for_task000_SQuAD_Shifts_reddit_dev": 65.6558,
+      "eval_rougeL_for_task000_SQuAD_dev": 70.0603,
+      "eval_rougeL_for_task000_SearchQA_dev": 6.3754,
+      "eval_runtime": 5547.761,
+      "eval_samples_per_second": 12.508,
+      "eval_steps_per_second": 6.254,
+      "step": 1000
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 5e-05,
+      "loss": 1.0877,
+      "step": 2000
+    },
+    {
+      "epoch": 0.2,
+      "eval_exact_match": 42.2972,
+      "eval_exact_match_for_squad-like": 42.2972,
+      "eval_exact_match_for_task000_NewsQA_dev": 28.8936,
+      "eval_exact_match_for_task000_SQuAD_Shifts_amazon_dev": 55.0531,
+      "eval_exact_match_for_task000_SQuAD_Shifts_new_wiki_dev": 56.4374,
+      "eval_exact_match_for_task000_SQuAD_Shifts_nyt_dev": 58.6687,
+      "eval_exact_match_for_task000_SQuAD_Shifts_reddit_dev": 54.8506,
+      "eval_exact_match_for_task000_SQuAD_dev": 61.3876,
+      "eval_exact_match_for_task000_SearchQA_dev": 2.821,
+      "eval_f1": 54.8049,
+      "eval_f1_for_squad-like": 54.8049,
+      "eval_f1_for_task000_NewsQA_dev": 46.0401,
+      "eval_f1_for_task000_SQuAD_Shifts_amazon_dev": 71.6103,
+      "eval_f1_for_task000_SQuAD_Shifts_new_wiki_dev": 72.681,
+      "eval_f1_for_task000_SQuAD_Shifts_nyt_dev": 72.9829,
+      "eval_f1_for_task000_SQuAD_Shifts_reddit_dev": 69.6718,
+      "eval_f1_for_task000_SQuAD_dev": 74.8343,
+      "eval_f1_for_task000_SearchQA_dev": 7.0869,
+      "eval_gen_len": 5.9867,
+      "eval_global_step": 2000,
+      "eval_loss": 1.5502516031265259,
+      "eval_rouge1": 54.5481,
+      "eval_rouge1_for_squad-like": 54.5481,
+      "eval_rouge1_for_task000_NewsQA_dev": 46.2121,
+      "eval_rouge1_for_task000_SQuAD_Shifts_amazon_dev": 71.3277,
+      "eval_rouge1_for_task000_SQuAD_Shifts_new_wiki_dev": 72.3257,
+      "eval_rouge1_for_task000_SQuAD_Shifts_nyt_dev": 72.7441,
+      "eval_rouge1_for_task000_SQuAD_Shifts_reddit_dev": 68.4653,
+      "eval_rouge1_for_task000_SQuAD_dev": 74.5387,
+      "eval_rouge1_for_task000_SearchQA_dev": 7.3461,
+      "eval_rougeL": 54.4937,
+      "eval_rougeL_for_squad-like": 54.4937,
+      "eval_rougeL_for_task000_NewsQA_dev": 46.1235,
+      "eval_rougeL_for_task000_SQuAD_Shifts_amazon_dev": 71.2521,
+      "eval_rougeL_for_task000_SQuAD_Shifts_new_wiki_dev": 72.2345,
+      "eval_rougeL_for_task000_SQuAD_Shifts_nyt_dev": 72.7112,
+      "eval_rougeL_for_task000_SQuAD_Shifts_reddit_dev": 68.3891,
+      "eval_rougeL_for_task000_SQuAD_dev": 74.4799,
+      "eval_rougeL_for_task000_SearchQA_dev": 7.3323,
+      "eval_runtime": 5244.5628,
+      "eval_samples_per_second": 13.231,
+      "eval_steps_per_second": 6.615,
+      "step": 2000
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 5e-05,
+      "loss": 1.0892,
+      "step": 3000
+    },
+    {
+      "epoch": 0.3,
+      "eval_exact_match": 42.9471,
+      "eval_exact_match_for_squad-like": 42.9471,
+      "eval_exact_match_for_task000_NewsQA_dev": 28.4188,
+      "eval_exact_match_for_task000_SQuAD_Shifts_amazon_dev": 56.3784,
+      "eval_exact_match_for_task000_SQuAD_Shifts_new_wiki_dev": 58.2514,
+      "eval_exact_match_for_task000_SQuAD_Shifts_nyt_dev": 59.7417,
+      "eval_exact_match_for_task000_SQuAD_Shifts_reddit_dev": 55.177,
+      "eval_exact_match_for_task000_SQuAD_dev": 62.387,
+      "eval_exact_match_for_task000_SearchQA_dev": 2.5324,
+      "eval_f1": 56.2906,
+      "eval_f1_for_squad-like": 56.2906,
+      "eval_f1_for_task000_NewsQA_dev": 46.7128,
+      "eval_f1_for_task000_SQuAD_Shifts_amazon_dev": 73.6538,
+      "eval_f1_for_task000_SQuAD_Shifts_new_wiki_dev": 75.7309,
+      "eval_f1_for_task000_SQuAD_Shifts_nyt_dev": 75.2649,
+      "eval_f1_for_task000_SQuAD_Shifts_reddit_dev": 71.4267,
+      "eval_f1_for_task000_SQuAD_dev": 76.8954,
+      "eval_f1_for_task000_SearchQA_dev": 6.7347,
+      "eval_gen_len": 6.1098,
+      "eval_global_step": 3000,
+      "eval_loss": 1.4990659952163696,
+      "eval_rouge1": 56.0886,
+      "eval_rouge1_for_squad-like": 56.0886,
+      "eval_rouge1_for_task000_NewsQA_dev": 46.9106,
+      "eval_rouge1_for_task000_SQuAD_Shifts_amazon_dev": 73.4005,
+      "eval_rouge1_for_task000_SQuAD_Shifts_new_wiki_dev": 75.5803,
+      "eval_rouge1_for_task000_SQuAD_Shifts_nyt_dev": 75.0322,
+      "eval_rouge1_for_task000_SQuAD_Shifts_reddit_dev": 70.2838,
+      "eval_rouge1_for_task000_SQuAD_dev": 76.6652,
+      "eval_rouge1_for_task000_SearchQA_dev": 7.018,
+      "eval_rougeL": 56.0392,
+      "eval_rougeL_for_squad-like": 56.0392,
+      "eval_rougeL_for_task000_NewsQA_dev": 46.8585,
+      "eval_rougeL_for_task000_SQuAD_Shifts_amazon_dev": 73.3224,
+      "eval_rougeL_for_task000_SQuAD_Shifts_new_wiki_dev": 75.5113,
+      "eval_rougeL_for_task000_SQuAD_Shifts_nyt_dev": 75.0019,
+      "eval_rougeL_for_task000_SQuAD_Shifts_reddit_dev": 70.2237,
+      "eval_rougeL_for_task000_SQuAD_dev": 76.5983,
+      "eval_rougeL_for_task000_SearchQA_dev": 7.0011,
+      "eval_runtime": 5239.5633,
+      "eval_samples_per_second": 13.243,
+      "eval_steps_per_second": 6.622,
+      "step": 3000
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 5e-05,
+      "loss": 0.9586,
+      "step": 4000
+    },
+    {
+      "epoch": 0.4,
+      "eval_exact_match": 43.2815,
+      "eval_exact_match_for_squad-like": 43.2815,
+      "eval_exact_match_for_task000_NewsQA_dev": 29.416,
+      "eval_exact_match_for_task000_SQuAD_Shifts_amazon_dev": 56.6313,
+      "eval_exact_match_for_task000_SQuAD_Shifts_new_wiki_dev": 59.0199,
+      "eval_exact_match_for_task000_SQuAD_Shifts_nyt_dev": 59.8013,
+      "eval_exact_match_for_task000_SQuAD_Shifts_reddit_dev": 56.2583,
+      "eval_exact_match_for_task000_SQuAD_dev": 62.406,
+      "eval_exact_match_for_task000_SearchQA_dev": 2.4735,
+      "eval_f1": 56.5384,
+      "eval_f1_for_squad-like": 56.5384,
+      "eval_f1_for_task000_NewsQA_dev": 47.1146,
+      "eval_f1_for_task000_SQuAD_Shifts_amazon_dev": 74.0447,
+      "eval_f1_for_task000_SQuAD_Shifts_new_wiki_dev": 76.4587,
+      "eval_f1_for_task000_SQuAD_Shifts_nyt_dev": 75.331,
+      "eval_f1_for_task000_SQuAD_Shifts_reddit_dev": 71.9027,
+      "eval_f1_for_task000_SQuAD_dev": 77.2248,
+      "eval_f1_for_task000_SearchQA_dev": 6.562,
+      "eval_gen_len": 6.5536,
+      "eval_global_step": 4000,
+      "eval_loss": 1.4763996601104736,
+      "eval_rouge1": 56.4181,
+      "eval_rouge1_for_squad-like": 56.4181,
+      "eval_rouge1_for_task000_NewsQA_dev": 47.2027,
+      "eval_rouge1_for_task000_SQuAD_Shifts_amazon_dev": 73.8315,
+      "eval_rouge1_for_task000_SQuAD_Shifts_new_wiki_dev": 76.3451,
+      "eval_rouge1_for_task000_SQuAD_Shifts_nyt_dev": 75.187,
+      "eval_rouge1_for_task000_SQuAD_Shifts_reddit_dev": 70.9723,
+      "eval_rouge1_for_task000_SQuAD_dev": 77.2012,
+      "eval_rouge1_for_task000_SearchQA_dev": 6.8629,
+      "eval_rougeL": 56.3585,
+      "eval_rougeL_for_squad-like": 56.3585,
+      "eval_rougeL_for_task000_NewsQA_dev": 47.124,
+      "eval_rougeL_for_task000_SQuAD_Shifts_amazon_dev": 73.7295,
+      "eval_rougeL_for_task000_SQuAD_Shifts_new_wiki_dev": 76.2619,
+      "eval_rougeL_for_task000_SQuAD_Shifts_nyt_dev": 75.156,
+      "eval_rougeL_for_task000_SQuAD_Shifts_reddit_dev": 70.9096,
+      "eval_rougeL_for_task000_SQuAD_dev": 77.129,
+      "eval_rougeL_for_task000_SearchQA_dev": 6.8364,
+      "eval_runtime": 5411.8899,
+      "eval_samples_per_second": 12.822,
+      "eval_steps_per_second": 6.411,
+      "step": 4000
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 5e-05,
+      "loss": 0.8703,
+      "step": 5000
+    },
+    {
+      "epoch": 0.5,
+      "eval_exact_match": 44.8091,
+      "eval_exact_match_for_squad-like": 44.8091,
+      "eval_exact_match_for_task000_NewsQA_dev": 29.1548,
+      "eval_exact_match_for_task000_SQuAD_Shifts_amazon_dev": 58.8569,
+      "eval_exact_match_for_task000_SQuAD_Shifts_new_wiki_dev": 60.7458,
+      "eval_exact_match_for_task000_SQuAD_Shifts_nyt_dev": 61.6294,
+      "eval_exact_match_for_task000_SQuAD_Shifts_reddit_dev": 58.7881,
+      "eval_exact_match_for_task000_SQuAD_dev": 65.661,
+      "eval_exact_match_for_task000_SearchQA_dev": 2.1201,
+      "eval_f1": 57.5067,
+      "eval_f1_for_squad-like": 57.5067,
+      "eval_f1_for_task000_NewsQA_dev": 47.528,
+      "eval_f1_for_task000_SQuAD_Shifts_amazon_dev": 75.5801,
+      "eval_f1_for_task000_SQuAD_Shifts_new_wiki_dev": 77.4224,
+      "eval_f1_for_task000_SQuAD_Shifts_nyt_dev": 76.6725,
+      "eval_f1_for_task000_SQuAD_Shifts_reddit_dev": 73.7737,
+      "eval_f1_for_task000_SQuAD_dev": 79.5359,
+      "eval_f1_for_task000_SearchQA_dev": 5.7665,
+      "eval_gen_len": 6.5617,
+      "eval_global_step": 5000,
+      "eval_loss": 1.4937905073165894,
+      "eval_rouge1": 57.5127,
+      "eval_rouge1_for_squad-like": 57.5127,
+      "eval_rouge1_for_task000_NewsQA_dev": 47.7949,
+      "eval_rouge1_for_task000_SQuAD_Shifts_amazon_dev": 75.4332,
+      "eval_rouge1_for_task000_SQuAD_Shifts_new_wiki_dev": 77.5705,
+      "eval_rouge1_for_task000_SQuAD_Shifts_nyt_dev": 76.5839,
+      "eval_rouge1_for_task000_SQuAD_Shifts_reddit_dev": 73.2737,
+      "eval_rouge1_for_task000_SQuAD_dev": 79.6618,
+      "eval_rouge1_for_task000_SearchQA_dev": 6.0046,
+      "eval_rougeL": 57.4686,
+      "eval_rougeL_for_squad-like": 57.4686,
+      "eval_rougeL_for_task000_NewsQA_dev": 47.6893,
+      "eval_rougeL_for_task000_SQuAD_Shifts_amazon_dev": 75.3688,
+      "eval_rougeL_for_task000_SQuAD_Shifts_new_wiki_dev": 77.5261,
+      "eval_rougeL_for_task000_SQuAD_Shifts_nyt_dev": 76.5531,
+      "eval_rougeL_for_task000_SQuAD_Shifts_reddit_dev": 73.2278,
+      "eval_rougeL_for_task000_SQuAD_dev": 79.6122,
+      "eval_rougeL_for_task000_SearchQA_dev": 5.9839,
+      "eval_runtime": 5610.9818,
+      "eval_samples_per_second": 12.367,
+      "eval_steps_per_second": 6.183,
+      "step": 5000
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 5e-05,
+      "loss": 0.9186,
+      "step": 6000
+    },
+    {
+      "epoch": 0.6,
+      "eval_exact_match": 45.2284,
+      "eval_exact_match_for_squad-like": 45.2284,
+      "eval_exact_match_for_task000_NewsQA_dev": 31.529,
+      "eval_exact_match_for_task000_SQuAD_Shifts_amazon_dev": 58.9277,
+      "eval_exact_match_for_task000_SQuAD_Shifts_new_wiki_dev": 60.4308,
+      "eval_exact_match_for_task000_SQuAD_Shifts_nyt_dev": 61.5897,
+      "eval_exact_match_for_task000_SQuAD_Shifts_reddit_dev": 58.8595,
+      "eval_exact_match_for_task000_SQuAD_dev": 66.0893,
+      "eval_exact_match_for_task000_SearchQA_dev": 3.0683,
+      "eval_f1": 57.6298,
+      "eval_f1_for_squad-like": 57.6298,
+      "eval_f1_for_task000_NewsQA_dev": 49.1496,
+      "eval_f1_for_task000_SQuAD_Shifts_amazon_dev": 75.4059,
+      "eval_f1_for_task000_SQuAD_Shifts_new_wiki_dev": 76.631,
+      "eval_f1_for_task000_SQuAD_Shifts_nyt_dev": 76.5732,
+      "eval_f1_for_task000_SQuAD_Shifts_reddit_dev": 73.1298,
+      "eval_f1_for_task000_SQuAD_dev": 79.143,
+      "eval_f1_for_task000_SearchQA_dev": 7.0128,
+      "eval_gen_len": 5.6208,
+      "eval_global_step": 6000,
+      "eval_loss": 1.4303617477416992,
+      "eval_rouge1": 57.5885,
+      "eval_rouge1_for_squad-like": 57.5885,
+      "eval_rouge1_for_task000_NewsQA_dev": 49.3293,
+      "eval_rouge1_for_task000_SQuAD_Shifts_amazon_dev": 75.2476,
+      "eval_rouge1_for_task000_SQuAD_Shifts_new_wiki_dev": 76.6477,
+      "eval_rouge1_for_task000_SQuAD_Shifts_nyt_dev": 76.397,
+      "eval_rouge1_for_task000_SQuAD_Shifts_reddit_dev": 72.5855,
+      "eval_rouge1_for_task000_SQuAD_dev": 79.167,
+      "eval_rouge1_for_task000_SearchQA_dev": 7.2874,
+      "eval_rougeL": 57.5543,
+      "eval_rougeL_for_squad-like": 57.5543,
+      "eval_rougeL_for_task000_NewsQA_dev": 49.2858,
+      "eval_rougeL_for_task000_SQuAD_Shifts_amazon_dev": 75.2047,
+      "eval_rougeL_for_task000_SQuAD_Shifts_new_wiki_dev": 76.6027,
+      "eval_rougeL_for_task000_SQuAD_Shifts_nyt_dev": 76.369,
+      "eval_rougeL_for_task000_SQuAD_Shifts_reddit_dev": 72.54,
+      "eval_rougeL_for_task000_SQuAD_dev": 79.1235,
+      "eval_rougeL_for_task000_SearchQA_dev": 7.2743,
+      "eval_runtime": 4976.2912,
+      "eval_samples_per_second": 13.944,
+      "eval_steps_per_second": 6.972,
+      "step": 6000
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 5e-05,
+      "loss": 0.8626,
+      "step": 7000
+    },
+    {
+      "epoch": 0.7,
+      "eval_exact_match": 46.145,
+      "eval_exact_match_for_squad-like": 46.145,
+      "eval_exact_match_for_task000_NewsQA_dev": 30.1282,
+      "eval_exact_match_for_task000_SQuAD_Shifts_amazon_dev": 59.6055,
+      "eval_exact_match_for_task000_SQuAD_Shifts_new_wiki_dev": 63.5676,
+      "eval_exact_match_for_task000_SQuAD_Shifts_nyt_dev": 63.5469,
+      "eval_exact_match_for_task000_SQuAD_Shifts_reddit_dev": 60.2775,
+      "eval_exact_match_for_task000_SQuAD_dev": 67.7453,
+      "eval_exact_match_for_task000_SearchQA_dev": 2.2968,
+      "eval_f1": 58.3057,
+      "eval_f1_for_squad-like": 58.3057,
+      "eval_f1_for_task000_NewsQA_dev": 48.4752,
+      "eval_f1_for_task000_SQuAD_Shifts_amazon_dev": 76.1677,
+      "eval_f1_for_task000_SQuAD_Shifts_new_wiki_dev": 79.2423,
+      "eval_f1_for_task000_SQuAD_Shifts_nyt_dev": 78.2045,
+      "eval_f1_for_task000_SQuAD_Shifts_reddit_dev": 74.6885,
+      "eval_f1_for_task000_SQuAD_dev": 80.3613,
+      "eval_f1_for_task000_SearchQA_dev": 5.657,
+      "eval_gen_len": 5.8196,
+      "eval_global_step": 7000,
+      "eval_loss": 1.3808618783950806,
+      "eval_rouge1": 58.2675,
+      "eval_rouge1_for_squad-like": 58.2675,
+      "eval_rouge1_for_task000_NewsQA_dev": 48.7542,
+      "eval_rouge1_for_task000_SQuAD_Shifts_amazon_dev": 75.9378,
+      "eval_rouge1_for_task000_SQuAD_Shifts_new_wiki_dev": 79.3134,
+      "eval_rouge1_for_task000_SQuAD_Shifts_nyt_dev": 78.1724,
+      "eval_rouge1_for_task000_SQuAD_Shifts_reddit_dev": 74.0395,
+      "eval_rouge1_for_task000_SQuAD_dev": 80.3759,
+      "eval_rouge1_for_task000_SearchQA_dev": 5.9169,
+      "eval_rougeL": 58.2236,
+      "eval_rougeL_for_squad-like": 58.2236,
+      "eval_rougeL_for_task000_NewsQA_dev": 48.6986,
+      "eval_rougeL_for_task000_SQuAD_Shifts_amazon_dev": 75.8622,
+      "eval_rougeL_for_task000_SQuAD_Shifts_new_wiki_dev": 79.2595,
+      "eval_rougeL_for_task000_SQuAD_Shifts_nyt_dev": 78.1477,
+      "eval_rougeL_for_task000_SQuAD_Shifts_reddit_dev": 73.978,
+      "eval_rougeL_for_task000_SQuAD_dev": 80.3316,
+      "eval_rougeL_for_task000_SearchQA_dev": 5.8983,
+      "eval_runtime": 5035.1499,
+      "eval_samples_per_second": 13.781,
+      "eval_steps_per_second": 6.891,
+      "step": 7000
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 5e-05,
+      "loss": 0.8322,
+      "step": 8000
+    },
+    {
+      "epoch": 0.8,
+      "eval_exact_match": 46.3338,
+      "eval_exact_match_for_squad-like": 46.3338,
+      "eval_exact_match_for_task000_NewsQA_dev": 31.5527,
+      "eval_exact_match_for_task000_SQuAD_Shifts_amazon_dev": 60.6576,
+      "eval_exact_match_for_task000_SQuAD_Shifts_new_wiki_dev": 63.4669,
+      "eval_exact_match_for_task000_SQuAD_Shifts_nyt_dev": 63.8351,
+      "eval_exact_match_for_task000_SQuAD_Shifts_reddit_dev": 60.5325,
+      "eval_exact_match_for_task000_SQuAD_dev": 67.0505,
+      "eval_exact_match_for_task000_SearchQA_dev": 2.2615,
+      "eval_f1": 58.5957,
+      "eval_f1_for_squad-like": 58.5957,
+      "eval_f1_for_task000_NewsQA_dev": 49.4319,
+      "eval_f1_for_task000_SQuAD_Shifts_amazon_dev": 76.9441,
+      "eval_f1_for_task000_SQuAD_Shifts_new_wiki_dev": 79.665,
+      "eval_f1_for_task000_SQuAD_Shifts_nyt_dev": 78.3856,
+      "eval_f1_for_task000_SQuAD_Shifts_reddit_dev": 75.1449,
+      "eval_f1_for_task000_SQuAD_dev": 80.3231,
+      "eval_f1_for_task000_SearchQA_dev": 5.608,
+      "eval_gen_len": 5.9989,
+      "eval_global_step": 8000,
+      "eval_loss": 1.4950507879257202,
+      "eval_rouge1": 58.6055,
+      "eval_rouge1_for_squad-like": 58.6055,
+      "eval_rouge1_for_task000_NewsQA_dev": 49.7455,
+      "eval_rouge1_for_task000_SQuAD_Shifts_amazon_dev": 76.7235,
+      "eval_rouge1_for_task000_SQuAD_Shifts_new_wiki_dev": 79.7546,
+      "eval_rouge1_for_task000_SQuAD_Shifts_nyt_dev": 78.4596,
+      "eval_rouge1_for_task000_SQuAD_Shifts_reddit_dev": 74.565,
+      "eval_rouge1_for_task000_SQuAD_dev": 80.4772,
+      "eval_rouge1_for_task000_SearchQA_dev": 5.8525,
+      "eval_rougeL": 58.5662,
+      "eval_rougeL_for_squad-like": 58.5662,
+      "eval_rougeL_for_task000_NewsQA_dev": 49.6823,
+      "eval_rougeL_for_task000_SQuAD_Shifts_amazon_dev": 76.6541,
+      "eval_rougeL_for_task000_SQuAD_Shifts_new_wiki_dev": 79.7031,
+      "eval_rougeL_for_task000_SQuAD_Shifts_nyt_dev": 78.4393,
+      "eval_rougeL_for_task000_SQuAD_Shifts_reddit_dev": 74.5146,
+      "eval_rougeL_for_task000_SQuAD_dev": 80.4338,
+      "eval_rougeL_for_task000_SearchQA_dev": 5.84,
+      "eval_runtime": 5037.486,
+      "eval_samples_per_second": 13.775,
+      "eval_steps_per_second": 6.887,
+      "step": 8000
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5e-05,
+      "loss": 0.8616,
+      "step": 9000
+    },
+    {
+      "epoch": 0.9,
+      "eval_exact_match": 46.805,
+      "eval_exact_match_for_squad-like": 46.805,
+      "eval_exact_match_for_task000_NewsQA_dev": 31.6002,
+      "eval_exact_match_for_task000_SQuAD_Shifts_amazon_dev": 60.9611,
+      "eval_exact_match_for_task000_SQuAD_Shifts_new_wiki_dev": 64.0086,
+      "eval_exact_match_for_task000_SQuAD_Shifts_nyt_dev": 63.686,
+      "eval_exact_match_for_task000_SQuAD_Shifts_reddit_dev": 61.0935,
+      "eval_exact_match_for_task000_SQuAD_dev": 68.4686,
+      "eval_exact_match_for_task000_SearchQA_dev": 2.6325,
+      "eval_f1": 58.9037,
+      "eval_f1_for_squad-like": 58.9037,
+      "eval_f1_for_task000_NewsQA_dev": 49.0476,
+      "eval_f1_for_task000_SQuAD_Shifts_amazon_dev": 77.1558,
+      "eval_f1_for_task000_SQuAD_Shifts_new_wiki_dev": 79.7546,
+      "eval_f1_for_task000_SQuAD_Shifts_nyt_dev": 78.4691,
+      "eval_f1_for_task000_SQuAD_Shifts_reddit_dev": 75.349,
+      "eval_f1_for_task000_SQuAD_dev": 81.0326,
+      "eval_f1_for_task000_SearchQA_dev": 6.1907,
+      "eval_gen_len": 6.0577,
+      "eval_global_step": 9000,
+      "eval_loss": 1.3887202739715576,
+      "eval_rouge1": 59.0007,
+      "eval_rouge1_for_squad-like": 59.0007,
+      "eval_rouge1_for_task000_NewsQA_dev": 49.3458,
+      "eval_rouge1_for_task000_SQuAD_Shifts_amazon_dev": 77.0702,
+      "eval_rouge1_for_task000_SQuAD_Shifts_new_wiki_dev": 79.9826,
+      "eval_rouge1_for_task000_SQuAD_Shifts_nyt_dev": 78.6904,
+      "eval_rouge1_for_task000_SQuAD_Shifts_reddit_dev": 74.9073,
+      "eval_rouge1_for_task000_SQuAD_dev": 81.2597,
+      "eval_rouge1_for_task000_SearchQA_dev": 6.4396,
+      "eval_rougeL": 58.9619,
+      "eval_rougeL_for_squad-like": 58.9619,
+      "eval_rougeL_for_task000_NewsQA_dev": 49.2793,
+      "eval_rougeL_for_task000_SQuAD_Shifts_amazon_dev": 77.0101,
+      "eval_rougeL_for_task000_SQuAD_Shifts_new_wiki_dev": 79.9214,
+      "eval_rougeL_for_task000_SQuAD_Shifts_nyt_dev": 78.6681,
+      "eval_rougeL_for_task000_SQuAD_Shifts_reddit_dev": 74.8591,
+      "eval_rougeL_for_task000_SQuAD_dev": 81.2181,
+      "eval_rougeL_for_task000_SearchQA_dev": 6.4279,
+      "eval_runtime": 5346.6704,
+      "eval_samples_per_second": 12.978,
+      "eval_steps_per_second": 6.489,
+      "step": 9000
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 5e-05,
+      "loss": 0.76,
+      "step": 10000
+    },
+    {
+      "epoch": 1.0,
+      "eval_exact_match": 47.0558,
+      "eval_exact_match_for_squad-like": 47.0558,
+      "eval_exact_match_for_task000_NewsQA_dev": 32.075,
+      "eval_exact_match_for_task000_SQuAD_Shifts_amazon_dev": 60.7688,
+      "eval_exact_match_for_task000_SQuAD_Shifts_new_wiki_dev": 64.6762,
+      "eval_exact_match_for_task000_SQuAD_Shifts_nyt_dev": 64.0139,
+      "eval_exact_match_for_task000_SQuAD_Shifts_reddit_dev": 61.3894,
+      "eval_exact_match_for_task000_SQuAD_dev": 68.7066,
+      "eval_exact_match_for_task000_SearchQA_dev": 2.8269,
+      "eval_f1": 59.6922,
+      "eval_f1_for_squad-like": 59.6922,
+      "eval_f1_for_task000_NewsQA_dev": 50.5612,
+      "eval_f1_for_task000_SQuAD_Shifts_amazon_dev": 77.7643,
+      "eval_f1_for_task000_SQuAD_Shifts_new_wiki_dev": 80.537,
+      "eval_f1_for_task000_SQuAD_Shifts_nyt_dev": 79.1097,
+      "eval_f1_for_task000_SQuAD_Shifts_reddit_dev": 76.3824,
+      "eval_f1_for_task000_SQuAD_dev": 81.8903,
+      "eval_f1_for_task000_SearchQA_dev": 6.8104,
+      "eval_gen_len": 6.2411,
+      "eval_global_step": 10000,
+      "eval_loss": 1.4306766986846924,
+      "eval_rouge1": 59.8786,
+      "eval_rouge1_for_squad-like": 59.8786,
+      "eval_rouge1_for_task000_NewsQA_dev": 51.1274,
+      "eval_rouge1_for_task000_SQuAD_Shifts_amazon_dev": 77.7056,
+      "eval_rouge1_for_task000_SQuAD_Shifts_new_wiki_dev": 80.8227,
+      "eval_rouge1_for_task000_SQuAD_Shifts_nyt_dev": 79.5036,
+      "eval_rouge1_for_task000_SQuAD_Shifts_reddit_dev": 76.0271,
+      "eval_rouge1_for_task000_SQuAD_dev": 82.2167,
+      "eval_rouge1_for_task000_SearchQA_dev": 7.1018,
+      "eval_rougeL": 59.8423,
+      "eval_rougeL_for_squad-like": 59.8423,
+      "eval_rougeL_for_task000_NewsQA_dev": 51.0694,
+      "eval_rougeL_for_task000_SQuAD_Shifts_amazon_dev": 77.66,
+      "eval_rougeL_for_task000_SQuAD_Shifts_new_wiki_dev": 80.7668,
+      "eval_rougeL_for_task000_SQuAD_Shifts_nyt_dev": 79.4863,
+      "eval_rougeL_for_task000_SQuAD_Shifts_reddit_dev": 75.9724,
+      "eval_rougeL_for_task000_SQuAD_dev": 82.178,
+      "eval_rougeL_for_task000_SearchQA_dev": 7.0864,
+      "eval_runtime": 5392.0755,
+      "eval_samples_per_second": 12.869,
+      "eval_steps_per_second": 6.434,
+      "step": 10000
+    },
+    {
+      "epoch": 1.1,
+      "learning_rate": 5e-05,
+      "loss": 0.6424,
+      "step": 11000
+    },
+    {
+      "epoch": 1.1,
+      "eval_exact_match": 47.7403,
+      "eval_exact_match_for_squad-like": 47.7403,
+      "eval_exact_match_for_task000_NewsQA_dev": 32.9772,
+      "eval_exact_match_for_task000_SQuAD_Shifts_amazon_dev": 61.1128,
+      "eval_exact_match_for_task000_SQuAD_Shifts_new_wiki_dev": 64.9912,
+      "eval_exact_match_for_task000_SQuAD_Shifts_nyt_dev": 65.7327,
+      "eval_exact_match_for_task000_SQuAD_Shifts_reddit_dev": 61.8892,
+      "eval_exact_match_for_task000_SQuAD_dev": 70.1342,
+      "eval_exact_match_for_task000_SearchQA_dev": 2.8622,
+      "eval_f1": 59.9977,
+      "eval_f1_for_squad-like": 59.9977,
+      "eval_f1_for_task000_NewsQA_dev": 51.0113,
+      "eval_f1_for_task000_SQuAD_Shifts_amazon_dev": 77.6964,
+      "eval_f1_for_task000_SQuAD_Shifts_new_wiki_dev": 80.5474,
+      "eval_f1_for_task000_SQuAD_Shifts_nyt_dev": 80.0965,
+      "eval_f1_for_task000_SQuAD_Shifts_reddit_dev": 76.6229,
+      "eval_f1_for_task000_SQuAD_dev": 82.6467,
+      "eval_f1_for_task000_SearchQA_dev": 6.79,
+      "eval_gen_len": 6.0952,
+      "eval_global_step": 11000,
+      "eval_loss": 1.458894968032837,
+      "eval_rouge1": 60.1081,
+      "eval_rouge1_for_squad-like": 60.1081,
+      "eval_rouge1_for_task000_NewsQA_dev": 51.3906,
+      "eval_rouge1_for_task000_SQuAD_Shifts_amazon_dev": 77.5815,
+      "eval_rouge1_for_task000_SQuAD_Shifts_new_wiki_dev": 80.8219,
+      "eval_rouge1_for_task000_SQuAD_Shifts_nyt_dev": 80.3102,
+      "eval_rouge1_for_task000_SQuAD_Shifts_reddit_dev": 76.1436,
+      "eval_rouge1_for_task000_SQuAD_dev": 82.8847,
+      "eval_rouge1_for_task000_SearchQA_dev": 7.0882,
+      "eval_rougeL": 60.0649,
+      "eval_rougeL_for_squad-like": 60.0649,
+      "eval_rougeL_for_task000_NewsQA_dev": 51.3201,
+      "eval_rougeL_for_task000_SQuAD_Shifts_amazon_dev": 77.5161,
+      "eval_rougeL_for_task000_SQuAD_Shifts_new_wiki_dev": 80.7665,
+      "eval_rougeL_for_task000_SQuAD_Shifts_nyt_dev": 80.2852,
+      "eval_rougeL_for_task000_SQuAD_Shifts_reddit_dev": 76.0825,
+      "eval_rougeL_for_task000_SQuAD_dev": 82.8429,
+      "eval_rougeL_for_task000_SearchQA_dev": 7.0691,
+      "eval_runtime": 5216.6936,
+      "eval_samples_per_second": 13.302,
+      "eval_steps_per_second": 6.651,
+      "step": 11000
+    },
+    {
+      "epoch": 1.2,
+      "learning_rate": 5e-05,
+      "loss": 0.6121,
+      "step": 12000
+    },
+    {
+      "epoch": 1.2,
+      "eval_exact_match": 47.0053,
+      "eval_exact_match_for_squad-like": 47.0053,
+      "eval_exact_match_for_task000_NewsQA_dev": 32.2175,
+      "eval_exact_match_for_task000_SQuAD_Shifts_amazon_dev": 60.3541,
+      "eval_exact_match_for_task000_SQuAD_Shifts_new_wiki_dev": 63.9204,
+      "eval_exact_match_for_task000_SQuAD_Shifts_nyt_dev": 64.2921,
+      "eval_exact_match_for_task000_SQuAD_Shifts_reddit_dev": 61.1547,
+      "eval_exact_match_for_task000_SQuAD_dev": 69.3633,
+      "eval_exact_match_for_task000_SearchQA_dev": 2.7444,
+      "eval_f1": 58.8892,
+      "eval_f1_for_squad-like": 58.8892,
+      "eval_f1_for_task000_NewsQA_dev": 50.0588,
+      "eval_f1_for_task000_SQuAD_Shifts_amazon_dev": 76.4924,
+      "eval_f1_for_task000_SQuAD_Shifts_new_wiki_dev": 79.1391,
+      "eval_f1_for_task000_SQuAD_Shifts_nyt_dev": 78.5755,
+      "eval_f1_for_task000_SQuAD_Shifts_reddit_dev": 75.0157,
+      "eval_f1_for_task000_SQuAD_dev": 81.7388,
+      "eval_f1_for_task000_SearchQA_dev": 6.2467,
+      "eval_gen_len": 5.4838,
+      "eval_global_step": 12000,
+      "eval_loss": 1.5454285144805908,
+      "eval_rouge1": 59.1209,
+      "eval_rouge1_for_squad-like": 59.1209,
+      "eval_rouge1_for_task000_NewsQA_dev": 50.5568,
+      "eval_rouge1_for_task000_SQuAD_Shifts_amazon_dev": 76.4035,
+      "eval_rouge1_for_task000_SQuAD_Shifts_new_wiki_dev": 79.652,
+      "eval_rouge1_for_task000_SQuAD_Shifts_nyt_dev": 78.959,
+      "eval_rouge1_for_task000_SQuAD_Shifts_reddit_dev": 74.82,
+      "eval_rouge1_for_task000_SQuAD_dev": 82.0668,
+      "eval_rouge1_for_task000_SearchQA_dev": 6.5645,
+      "eval_rougeL": 59.0936,
+      "eval_rougeL_for_squad-like": 59.0936,
+      "eval_rougeL_for_task000_NewsQA_dev": 50.5017,
+      "eval_rougeL_for_task000_SQuAD_Shifts_amazon_dev": 76.3605,
+      "eval_rougeL_for_task000_SQuAD_Shifts_new_wiki_dev": 79.6138,
+      "eval_rougeL_for_task000_SQuAD_Shifts_nyt_dev": 78.939,
+      "eval_rougeL_for_task000_SQuAD_Shifts_reddit_dev": 74.7889,
+      "eval_rougeL_for_task000_SQuAD_dev": 82.0374,
+      "eval_rougeL_for_task000_SearchQA_dev": 6.5579,
+      "eval_runtime": 4739.7954,
+      "eval_samples_per_second": 14.64,
+      "eval_steps_per_second": 7.32,
+      "step": 12000
+    },
+    {
+      "epoch": 1.3,
+      "learning_rate": 5e-05,
+      "loss": 0.6267,
+      "step": 13000
+    },
+    {
+      "epoch": 1.3,
+      "eval_exact_match": 46.8425,
+      "eval_exact_match_for_squad-like": 46.8425,
+      "eval_exact_match_for_task000_NewsQA_dev": 31.2441,
+      "eval_exact_match_for_task000_SQuAD_Shifts_amazon_dev": 59.8381,
+      "eval_exact_match_for_task000_SQuAD_Shifts_new_wiki_dev": 64.8652,
+      "eval_exact_match_for_task000_SQuAD_Shifts_nyt_dev": 64.7491,
+      "eval_exact_match_for_task000_SQuAD_Shifts_reddit_dev": 60.3081,
+      "eval_exact_match_for_task000_SQuAD_dev": 69.0968,
+      "eval_exact_match_for_task000_SearchQA_dev": 2.5618,
+      "eval_f1": 59.0311,
+      "eval_f1_for_squad-like": 59.0311,
+      "eval_f1_for_task000_NewsQA_dev": 49.2573,
+      "eval_f1_for_task000_SQuAD_Shifts_amazon_dev": 76.2242,
+      "eval_f1_for_task000_SQuAD_Shifts_new_wiki_dev": 80.2601,
+      "eval_f1_for_task000_SQuAD_Shifts_nyt_dev": 79.0037,
+      "eval_f1_for_task000_SQuAD_Shifts_reddit_dev": 74.8635,
+      "eval_f1_for_task000_SQuAD_dev": 81.6974,
+      "eval_f1_for_task000_SearchQA_dev": 6.5171,
+      "eval_gen_len": 6.1312,
+      "eval_global_step": 13000,
+      "eval_loss": 1.3520243167877197,
+      "eval_rouge1": 59.3382,
+      "eval_rouge1_for_squad-like": 59.3382,
+      "eval_rouge1_for_task000_NewsQA_dev": 49.8146,
+      "eval_rouge1_for_task000_SQuAD_Shifts_amazon_dev": 76.3477,
+      "eval_rouge1_for_task000_SQuAD_Shifts_new_wiki_dev": 80.7192,
+      "eval_rouge1_for_task000_SQuAD_Shifts_nyt_dev": 79.3855,
+      "eval_rouge1_for_task000_SQuAD_Shifts_reddit_dev": 74.8334,
+      "eval_rouge1_for_task000_SQuAD_dev": 82.1319,
+      "eval_rouge1_for_task000_SearchQA_dev": 6.8695,
+      "eval_rougeL": 59.2915,
+      "eval_rougeL_for_squad-like": 59.2915,
+      "eval_rougeL_for_task000_NewsQA_dev": 49.7207,
+      "eval_rougeL_for_task000_SQuAD_Shifts_amazon_dev": 76.2786,
+      "eval_rougeL_for_task000_SQuAD_Shifts_new_wiki_dev": 80.6445,
+      "eval_rougeL_for_task000_SQuAD_Shifts_nyt_dev": 79.3596,
+      "eval_rougeL_for_task000_SQuAD_Shifts_reddit_dev": 74.7793,
+      "eval_rougeL_for_task000_SQuAD_dev": 82.1035,
+      "eval_rougeL_for_task000_SearchQA_dev": 6.8414,
+      "eval_runtime": 5365.8691,
+      "eval_samples_per_second": 12.932,
+      "eval_steps_per_second": 6.466,
+      "step": 13000
+    },
+    {
+      "epoch": 1.4,
+      "learning_rate": 5e-05,
+      "loss": 0.6173,
+      "step": 14000
+    },
+    {
+      "epoch": 1.4,
+      "eval_exact_match": 47.0529,
+      "eval_exact_match_for_squad-like": 47.0529,
+      "eval_exact_match_for_task000_NewsQA_dev": 32.265,
+      "eval_exact_match_for_task000_SQuAD_Shifts_amazon_dev": 60.0506,
+      "eval_exact_match_for_task000_SQuAD_Shifts_new_wiki_dev": 65.4069,
+      "eval_exact_match_for_task000_SQuAD_Shifts_nyt_dev": 64.7988,
+      "eval_exact_match_for_task000_SQuAD_Shifts_reddit_dev": 60.3897,
+      "eval_exact_match_for_task000_SQuAD_dev": 68.5828,
+      "eval_exact_match_for_task000_SearchQA_dev": 3.033,
+      "eval_f1": 59.4463,
+      "eval_f1_for_squad-like": 59.4463,
+      "eval_f1_for_task000_NewsQA_dev": 49.3437,
+      "eval_f1_for_task000_SQuAD_Shifts_amazon_dev": 76.8803,
+      "eval_f1_for_task000_SQuAD_Shifts_new_wiki_dev": 81.055,
+      "eval_f1_for_task000_SQuAD_Shifts_nyt_dev": 78.972,
+      "eval_f1_for_task000_SQuAD_Shifts_reddit_dev": 75.2939,
+      "eval_f1_for_task000_SQuAD_dev": 81.7056,
+      "eval_f1_for_task000_SearchQA_dev": 7.2043,
+      "eval_gen_len": 6.2021,
+      "eval_global_step": 14000,
+      "eval_loss": 1.421161413192749,
+      "eval_rouge1": 59.7198,
+      "eval_rouge1_for_squad-like": 59.7198,
+      "eval_rouge1_for_task000_NewsQA_dev": 49.8919,
+      "eval_rouge1_for_task000_SQuAD_Shifts_amazon_dev": 76.9311,
+      "eval_rouge1_for_task000_SQuAD_Shifts_new_wiki_dev": 81.4792,
+      "eval_rouge1_for_task000_SQuAD_Shifts_nyt_dev": 79.4279,
+      "eval_rouge1_for_task000_SQuAD_Shifts_reddit_dev": 75.1727,
+      "eval_rouge1_for_task000_SQuAD_dev": 82.0675,
+      "eval_rouge1_for_task000_SearchQA_dev": 7.5337,
+      "eval_rougeL": 59.6836,
+      "eval_rougeL_for_squad-like": 59.6836,
+      "eval_rougeL_for_task000_NewsQA_dev": 49.8304,
+      "eval_rougeL_for_task000_SQuAD_Shifts_amazon_dev": 76.8717,
+      "eval_rougeL_for_task000_SQuAD_Shifts_new_wiki_dev": 81.4312,
+      "eval_rougeL_for_task000_SQuAD_Shifts_nyt_dev": 79.4022,
+      "eval_rougeL_for_task000_SQuAD_Shifts_reddit_dev": 75.1244,
+      "eval_rougeL_for_task000_SQuAD_dev": 82.035,
+      "eval_rougeL_for_task000_SearchQA_dev": 7.5215,
+      "eval_runtime": 5210.8523,
+      "eval_samples_per_second": 13.316,
+      "eval_steps_per_second": 6.658,
+      "step": 14000
+    },
+    {
+      "epoch": 1.5,
+      "learning_rate": 5e-05,
+      "loss": 0.6122,
+      "step": 15000
+    },
+    {
+      "epoch": 1.5,
+      "eval_exact_match": 47.7187,
+      "eval_exact_match_for_squad-like": 47.7187,
+      "eval_exact_match_for_task000_NewsQA_dev": 31.9801,
+      "eval_exact_match_for_task000_SQuAD_Shifts_amazon_dev": 60.6677,
+      "eval_exact_match_for_task000_SQuAD_Shifts_new_wiki_dev": 65.7722,
+      "eval_exact_match_for_task000_SQuAD_Shifts_nyt_dev": 66.1003,
+      "eval_exact_match_for_task000_SQuAD_Shifts_reddit_dev": 61.4302,
+      "eval_exact_match_for_task000_SQuAD_dev": 70.4197,
+      "eval_exact_match_for_task000_SearchQA_dev": 2.7856,
+      "eval_f1": 60.0424,
+      "eval_f1_for_squad-like": 60.0424,
+      "eval_f1_for_task000_NewsQA_dev": 49.8937,
+      "eval_f1_for_task000_SQuAD_Shifts_amazon_dev": 77.6636,
+      "eval_f1_for_task000_SQuAD_Shifts_new_wiki_dev": 81.1636,
+      "eval_f1_for_task000_SQuAD_Shifts_nyt_dev": 79.9381,
+      "eval_f1_for_task000_SQuAD_Shifts_reddit_dev": 76.2199,
+      "eval_f1_for_task000_SQuAD_dev": 82.7855,
+      "eval_f1_for_task000_SearchQA_dev": 7.2213,
+      "eval_gen_len": 6.7893,
+      "eval_global_step": 15000,
+      "eval_loss": 1.4648000001907349,
+      "eval_rouge1": 60.3199,
+      "eval_rouge1_for_squad-like": 60.3199,
+      "eval_rouge1_for_task000_NewsQA_dev": 50.5586,
+      "eval_rouge1_for_task000_SQuAD_Shifts_amazon_dev": 77.6507,
+      "eval_rouge1_for_task000_SQuAD_Shifts_new_wiki_dev": 81.6021,
+      "eval_rouge1_for_task000_SQuAD_Shifts_nyt_dev": 80.4139,
+      "eval_rouge1_for_task000_SQuAD_Shifts_reddit_dev": 75.9828,
+      "eval_rouge1_for_task000_SQuAD_dev": 83.1733,
+      "eval_rouge1_for_task000_SearchQA_dev": 7.608,
+      "eval_rougeL": 60.27,
+      "eval_rougeL_for_squad-like": 60.27,
+      "eval_rougeL_for_task000_NewsQA_dev": 50.4674,
+      "eval_rougeL_for_task000_SQuAD_Shifts_amazon_dev": 77.5673,
+      "eval_rougeL_for_task000_SQuAD_Shifts_new_wiki_dev": 81.5322,
+      "eval_rougeL_for_task000_SQuAD_Shifts_nyt_dev": 80.3954,
+      "eval_rougeL_for_task000_SQuAD_Shifts_reddit_dev": 75.9265,
+      "eval_rougeL_for_task000_SQuAD_dev": 83.1281,
+      "eval_rougeL_for_task000_SearchQA_dev": 7.5794,
+      "eval_runtime": 5599.7297,
+      "eval_samples_per_second": 12.392,
+      "eval_steps_per_second": 6.196,
+      "step": 15000
+    },
+    {
+      "epoch": 1.6,
+      "learning_rate": 5e-05,
+      "loss": 0.5975,
+      "step": 16000
+    },
+    {
+      "epoch": 1.6,
+      "eval_exact_match": 47.7028,
+      "eval_exact_match_for_squad-like": 47.7028,
+      "eval_exact_match_for_task000_NewsQA_dev": 32.906,
+      "eval_exact_match_for_task000_SQuAD_Shifts_amazon_dev": 60.6373,
+      "eval_exact_match_for_task000_SQuAD_Shifts_new_wiki_dev": 65.4825,
+      "eval_exact_match_for_task000_SQuAD_Shifts_nyt_dev": 65.0373,
+      "eval_exact_match_for_task000_SQuAD_Shifts_reddit_dev": 61.3996,
+      "eval_exact_match_for_task000_SQuAD_dev": 70.1056,
+      "eval_exact_match_for_task000_SearchQA_dev": 3.4865,
+      "eval_f1": 60.0719,
+      "eval_f1_for_squad-like": 60.0719,
+      "eval_f1_for_task000_NewsQA_dev": 50.6484,
+      "eval_f1_for_task000_SQuAD_Shifts_amazon_dev": 77.4577,
+      "eval_f1_for_task000_SQuAD_Shifts_new_wiki_dev": 80.7913,
+      "eval_f1_for_task000_SQuAD_Shifts_nyt_dev": 79.553,
+      "eval_f1_for_task000_SQuAD_Shifts_reddit_dev": 75.9467,
+      "eval_f1_for_task000_SQuAD_dev": 82.5792,
+      "eval_f1_for_task000_SearchQA_dev": 7.9625,
+      "eval_gen_len": 6.5056,
+      "eval_global_step": 16000,
+      "eval_loss": 1.4460376501083374,
+      "eval_rouge1": 60.3765,
+      "eval_rouge1_for_squad-like": 60.3765,
+      "eval_rouge1_for_task000_NewsQA_dev": 51.3099,
+      "eval_rouge1_for_task000_SQuAD_Shifts_amazon_dev": 77.4873,
+      "eval_rouge1_for_task000_SQuAD_Shifts_new_wiki_dev": 81.2315,
+      "eval_rouge1_for_task000_SQuAD_Shifts_nyt_dev": 80.091,
+      "eval_rouge1_for_task000_SQuAD_Shifts_reddit_dev": 75.8312,
+      "eval_rouge1_for_task000_SQuAD_dev": 83.0172,
+      "eval_rouge1_for_task000_SearchQA_dev": 8.2968,
+      "eval_rougeL": 60.3322,
+      "eval_rougeL_for_squad-like": 60.3322,
+      "eval_rougeL_for_task000_NewsQA_dev": 51.2196,
+      "eval_rougeL_for_task000_SQuAD_Shifts_amazon_dev": 77.4096,
+      "eval_rougeL_for_task000_SQuAD_Shifts_new_wiki_dev": 81.1818,
+      "eval_rougeL_for_task000_SQuAD_Shifts_nyt_dev": 80.0692,
+      "eval_rougeL_for_task000_SQuAD_Shifts_reddit_dev": 75.7759,
+      "eval_rougeL_for_task000_SQuAD_dev": 82.9814,
+      "eval_rougeL_for_task000_SearchQA_dev": 8.2737,
+      "eval_runtime": 5456.9286,
+      "eval_samples_per_second": 12.716,
+      "eval_steps_per_second": 6.358,
+      "step": 16000
+    },
+    {
+      "epoch": 1.7,
+      "learning_rate": 5e-05,
+      "loss": 0.5826,
+      "step": 17000
+    },
+    {
+      "epoch": 1.7,
+      "eval_exact_match": 47.6063,
+      "eval_exact_match_for_squad-like": 47.6063,
+      "eval_exact_match_for_task000_NewsQA_dev": 32.6923,
+      "eval_exact_match_for_task000_SQuAD_Shifts_amazon_dev": 60.2428,
+      "eval_exact_match_for_task000_SQuAD_Shifts_new_wiki_dev": 65.8478,
+      "eval_exact_match_for_task000_SQuAD_Shifts_nyt_dev": 65.6433,
+      "eval_exact_match_for_task000_SQuAD_Shifts_reddit_dev": 61.3384,
+      "eval_exact_match_for_task000_SQuAD_dev": 69.9534,
+      "eval_exact_match_for_task000_SearchQA_dev": 2.9741,
+      "eval_f1": 59.786,
+      "eval_f1_for_squad-like": 59.786,
+      "eval_f1_for_task000_NewsQA_dev": 50.6166,
+      "eval_f1_for_task000_SQuAD_Shifts_amazon_dev": 77.0853,
+      "eval_f1_for_task000_SQuAD_Shifts_new_wiki_dev": 81.0205,
+      "eval_f1_for_task000_SQuAD_Shifts_nyt_dev": 79.6158,
+      "eval_f1_for_task000_SQuAD_Shifts_reddit_dev": 75.6525,
+      "eval_f1_for_task000_SQuAD_dev": 82.5741,
+      "eval_f1_for_task000_SearchQA_dev": 7.0474,
+      "eval_gen_len": 6.2252,
+      "eval_global_step": 17000,
+      "eval_loss": 1.4596614837646484,
+      "eval_rouge1": 60.1349,
+      "eval_rouge1_for_squad-like": 60.1349,
+      "eval_rouge1_for_task000_NewsQA_dev": 51.2595,
+      "eval_rouge1_for_task000_SQuAD_Shifts_amazon_dev": 77.2979,
+      "eval_rouge1_for_task000_SQuAD_Shifts_new_wiki_dev": 81.5869,
+      "eval_rouge1_for_task000_SQuAD_Shifts_nyt_dev": 80.098,
+      "eval_rouge1_for_task000_SQuAD_Shifts_reddit_dev": 75.5838,
+      "eval_rouge1_for_task000_SQuAD_dev": 83.0602,
+      "eval_rouge1_for_task000_SearchQA_dev": 7.378,
+      "eval_rougeL": 60.0935,
+      "eval_rougeL_for_squad-like": 60.0935,
+      "eval_rougeL_for_task000_NewsQA_dev": 51.189,
+      "eval_rougeL_for_task000_SQuAD_Shifts_amazon_dev": 77.2209,
+      "eval_rougeL_for_task000_SQuAD_Shifts_new_wiki_dev": 81.5258,
+      "eval_rougeL_for_task000_SQuAD_Shifts_nyt_dev": 80.0791,
+      "eval_rougeL_for_task000_SQuAD_Shifts_reddit_dev": 75.5283,
+      "eval_rougeL_for_task000_SQuAD_dev": 83.0318,
+      "eval_rougeL_for_task000_SearchQA_dev": 7.3608,
+      "eval_runtime": 5176.1682,
+      "eval_samples_per_second": 13.406,
+      "eval_steps_per_second": 6.703,
+      "step": 17000
+    },
+    {
+      "epoch": 1.8,
+      "learning_rate": 5e-05,
+      "loss": 0.5955,
+      "step": 18000
+    },
+    {
+      "epoch": 1.8,
+      "eval_exact_match": 47.0024,
+      "eval_exact_match_for_squad-like": 47.0024,
+      "eval_exact_match_for_task000_NewsQA_dev": 32.3837,
+      "eval_exact_match_for_task000_SQuAD_Shifts_amazon_dev": 60.0303,
+      "eval_exact_match_for_task000_SQuAD_Shifts_new_wiki_dev": 64.8526,
+      "eval_exact_match_for_task000_SQuAD_Shifts_nyt_dev": 64.7591,
+      "eval_exact_match_for_task000_SQuAD_Shifts_reddit_dev": 60.5631,
+      "eval_exact_match_for_task000_SQuAD_dev": 69.3728,
+      "eval_exact_match_for_task000_SearchQA_dev": 2.5029,
+      "eval_f1": 59.6025,
+      "eval_f1_for_squad-like": 59.6025,
+      "eval_f1_for_task000_NewsQA_dev": 49.9466,
+      "eval_f1_for_task000_SQuAD_Shifts_amazon_dev": 77.0998,
+      "eval_f1_for_task000_SQuAD_Shifts_new_wiki_dev": 80.7246,
+      "eval_f1_for_task000_SQuAD_Shifts_nyt_dev": 79.6329,
+      "eval_f1_for_task000_SQuAD_Shifts_reddit_dev": 75.662,
+      "eval_f1_for_task000_SQuAD_dev": 82.5021,
+      "eval_f1_for_task000_SearchQA_dev": 6.6223,
+      "eval_gen_len": 7.2863,
+      "eval_global_step": 18000,
+      "eval_loss": 1.5001791715621948,
+      "eval_rouge1": 59.8829,
+      "eval_rouge1_for_squad-like": 59.8829,
+      "eval_rouge1_for_task000_NewsQA_dev": 50.5762,
+      "eval_rouge1_for_task000_SQuAD_Shifts_amazon_dev": 77.2014,
+      "eval_rouge1_for_task000_SQuAD_Shifts_new_wiki_dev": 81.1701,
+      "eval_rouge1_for_task000_SQuAD_Shifts_nyt_dev": 80.1226,
+      "eval_rouge1_for_task000_SQuAD_Shifts_reddit_dev": 75.4758,
+      "eval_rouge1_for_task000_SQuAD_dev": 82.8425,
+      "eval_rouge1_for_task000_SearchQA_dev": 6.9512,
+      "eval_rougeL": 59.8296,
+      "eval_rougeL_for_squad-like": 59.8296,
+      "eval_rougeL_for_task000_NewsQA_dev": 50.4466,
+      "eval_rougeL_for_task000_SQuAD_Shifts_amazon_dev": 77.1325,
+      "eval_rougeL_for_task000_SQuAD_Shifts_new_wiki_dev": 81.0927,
+      "eval_rougeL_for_task000_SQuAD_Shifts_nyt_dev": 80.0963,
+      "eval_rougeL_for_task000_SQuAD_Shifts_reddit_dev": 75.4105,
+      "eval_rougeL_for_task000_SQuAD_dev": 82.7958,
+      "eval_rougeL_for_task000_SearchQA_dev": 6.924,
+      "eval_runtime": 5979.7657,
+      "eval_samples_per_second": 11.604,
+      "eval_steps_per_second": 5.802,
+      "step": 18000
+    },
+    {
+      "epoch": 1.9,
+      "learning_rate": 5e-05,
+      "loss": 0.6105,
+      "step": 19000
+    },
+    {
+      "epoch": 1.9,
+      "eval_exact_match": 48.3542,
+      "eval_exact_match_for_squad-like": 48.3542,
+      "eval_exact_match_for_task000_NewsQA_dev": 35.1377,
+      "eval_exact_match_for_task000_SQuAD_Shifts_amazon_dev": 61.4972,
+      "eval_exact_match_for_task000_SQuAD_Shifts_new_wiki_dev": 66.3895,
+      "eval_exact_match_for_task000_SQuAD_Shifts_nyt_dev": 66.4382,
+      "eval_exact_match_for_task000_SQuAD_Shifts_reddit_dev": 61.9198,
+      "eval_exact_match_for_task000_SQuAD_dev": 71.1335,
+      "eval_exact_match_for_task000_SearchQA_dev": 2.9034,
+      "eval_f1": 60.527,
+      "eval_f1_for_squad-like": 60.527,
+      "eval_f1_for_task000_NewsQA_dev": 53.3584,
+      "eval_f1_for_task000_SQuAD_Shifts_amazon_dev": 78.0211,
+      "eval_f1_for_task000_SQuAD_Shifts_new_wiki_dev": 81.8373,
+      "eval_f1_for_task000_SQuAD_Shifts_nyt_dev": 80.6907,
+      "eval_f1_for_task000_SQuAD_Shifts_reddit_dev": 76.7447,
+      "eval_f1_for_task000_SQuAD_dev": 83.3642,
+      "eval_f1_for_task000_SearchQA_dev": 6.712,
+      "eval_gen_len": 6.1527,
+      "eval_global_step": 19000,
+      "eval_loss": 1.4464625120162964,
+      "eval_rouge1": 60.7571,
+      "eval_rouge1_for_squad-like": 60.7571,
+      "eval_rouge1_for_task000_NewsQA_dev": 53.7911,
+      "eval_rouge1_for_task000_SQuAD_Shifts_amazon_dev": 77.9828,
+      "eval_rouge1_for_task000_SQuAD_Shifts_new_wiki_dev": 82.3105,
+      "eval_rouge1_for_task000_SQuAD_Shifts_nyt_dev": 81.1026,
+      "eval_rouge1_for_task000_SQuAD_Shifts_reddit_dev": 76.486,
+      "eval_rouge1_for_task000_SQuAD_dev": 83.7045,
+      "eval_rouge1_for_task000_SearchQA_dev": 7.0408,
+      "eval_rougeL": 60.7148,
+      "eval_rougeL_for_squad-like": 60.7148,
+      "eval_rougeL_for_task000_NewsQA_dev": 53.7027,
+      "eval_rougeL_for_task000_SQuAD_Shifts_amazon_dev": 77.9168,
+      "eval_rougeL_for_task000_SQuAD_Shifts_new_wiki_dev": 82.2503,
+      "eval_rougeL_for_task000_SQuAD_Shifts_nyt_dev": 81.0755,
+      "eval_rougeL_for_task000_SQuAD_Shifts_reddit_dev": 76.4296,
+      "eval_rougeL_for_task000_SQuAD_dev": 83.673,
+      "eval_rougeL_for_task000_SearchQA_dev": 7.0247,
+      "eval_runtime": 5228.9365,
+      "eval_samples_per_second": 13.27,
+      "eval_steps_per_second": 6.635,
+      "step": 19000
+    },
+    {
+      "epoch": 2.0,
+      "learning_rate": 5e-05,
+      "loss": 0.6235,
+      "step": 20000
+    },
+    {
+      "epoch": 2.0,
+      "eval_exact_match": 48.1741,
+      "eval_exact_match_for_squad-like": 48.1741,
+      "eval_exact_match_for_task000_NewsQA_dev": 33.3571,
+      "eval_exact_match_for_task000_SQuAD_Shifts_amazon_dev": 60.9813,
+      "eval_exact_match_for_task000_SQuAD_Shifts_new_wiki_dev": 66.6163,
+      "eval_exact_match_for_task000_SQuAD_Shifts_nyt_dev": 66.5673,
+      "eval_exact_match_for_task000_SQuAD_Shifts_reddit_dev": 61.7668,
+      "eval_exact_match_for_task000_SQuAD_dev": 70.8099,
+      "eval_exact_match_for_task000_SearchQA_dev": 3.0153,
+      "eval_f1": 60.3279,
+      "eval_f1_for_squad-like": 60.3279,
+      "eval_f1_for_task000_NewsQA_dev": 51.5268,
+      "eval_f1_for_task000_SQuAD_Shifts_amazon_dev": 77.7923,
+      "eval_f1_for_task000_SQuAD_Shifts_new_wiki_dev": 81.3576,
+      "eval_f1_for_task000_SQuAD_Shifts_nyt_dev": 80.6284,
+      "eval_f1_for_task000_SQuAD_Shifts_reddit_dev": 76.1448,
+      "eval_f1_for_task000_SQuAD_dev": 83.223,
+      "eval_f1_for_task000_SearchQA_dev": 7.181,
+      "eval_gen_len": 6.5938,
+      "eval_global_step": 20000,
+      "eval_loss": 1.4583032131195068,
+      "eval_rouge1": 60.5924,
+      "eval_rouge1_for_squad-like": 60.5924,
+      "eval_rouge1_for_task000_NewsQA_dev": 52.0147,
+      "eval_rouge1_for_task000_SQuAD_Shifts_amazon_dev": 77.7925,
+      "eval_rouge1_for_task000_SQuAD_Shifts_new_wiki_dev": 81.8888,
+      "eval_rouge1_for_task000_SQuAD_Shifts_nyt_dev": 81.0775,
+      "eval_rouge1_for_task000_SQuAD_Shifts_reddit_dev": 75.956,
+      "eval_rouge1_for_task000_SQuAD_dev": 83.5919,
+      "eval_rouge1_for_task000_SearchQA_dev": 7.507,
+      "eval_rougeL": 60.5491,
+      "eval_rougeL_for_squad-like": 60.5491,
+      "eval_rougeL_for_task000_NewsQA_dev": 51.9291,
+      "eval_rougeL_for_task000_SQuAD_Shifts_amazon_dev": 77.7222,
+      "eval_rougeL_for_task000_SQuAD_Shifts_new_wiki_dev": 81.8211,
+      "eval_rougeL_for_task000_SQuAD_Shifts_nyt_dev": 81.0484,
+      "eval_rougeL_for_task000_SQuAD_Shifts_reddit_dev": 75.9099,
+      "eval_rougeL_for_task000_SQuAD_dev": 83.5543,
+      "eval_rougeL_for_task000_SearchQA_dev": 7.4907,
+      "eval_runtime": 5575.4762,
+      "eval_samples_per_second": 12.446,
+      "eval_steps_per_second": 6.223,
+      "step": 20000
+    }
+  ],
+  "max_steps": 20000,
+  "num_train_epochs": 2,
+  "total_flos": 3.03606909632512e+16,
+  "trial_name": null,
+  "trial_params": null
+}

training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0acfe607594fbedce1f298fa9386bef262f4bfc9f39641ae385b31958eb516d1
+size 4335

zero_pp_rank_0_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c6372675d32c160ccc77caff6914e383bb811193b92e9bebd9175d2340b9347e
+size 1485211683

zero_pp_rank_1_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:297883e4f28079d20226855061989d4a806c99b8762420dcd495a4fb973ae26c
+size 1485211683

zero_to_fp32.py

@@ -0,0 +1,468 @@
+#!/usr/bin/env python
+
+# This script extracts fp32 consolidated weights from a zero 2 and 3 DeepSpeed checkpoints. It gets
+# copied into the top level checkpoint dir, so the user can easily do the conversion at any point in
+# the future. Once extracted, the weights don't require DeepSpeed and can be used in any
+# application.
+#
+# example: python zero_to_fp32.py . pytorch_model.bin
+
+import argparse
+import torch
+import glob
+import math
+import os
+from collections import OrderedDict
+
+# while this script doesn't use deepspeed to recover data, since the checkpoints are pickled with
+# DeepSpeed data structures it has to be available in the current python environment.
+import deepspeed
+from deepspeed.utils import logger
+from deepspeed.checkpoint.constants import (DS_VERSION,
+                                            OPTIMIZER_STATE_DICT,
+                                            PARAM_SHAPES,
+                                            SINGLE_PARTITION_OF_FP32_GROUPS,
+                                            FP32_FLAT_GROUPS,
+                                            ZERO_STAGE,
+                                            PARTITION_COUNT,
+                                            PARAM_SHAPES,
+                                            BUFFER_NAMES)
+
+debug = 0
+
+# load to cpu
+device = torch.device('cpu')
+
+
+def get_model_state_file(checkpoint_dir, zero_stage):
+    if not os.path.isdir(checkpoint_dir):
+        raise FileNotFoundError(f"Directory '{checkpoint_dir}' doesn't exist")
+
+    # there should be only one file
+    if zero_stage == 2:
+        file = os.path.join(checkpoint_dir, "mp_rank_00_model_states.pt")
+    elif zero_stage == 3:
+        file = os.path.join(checkpoint_dir, "zero_pp_rank_0_mp_rank_00_model_states.pt")
+
+    if not os.path.exists(file):
+        raise FileNotFoundError(f"can't find model states file at '{file}'")
+
+    return file
+
+
+def get_optim_files(checkpoint_dir):
+    # XXX: need to test that this simple glob rule works for multi-node setup too
+    optim_files = sorted(glob.glob(os.path.join(checkpoint_dir, "*_optim_states.pt")))
+
+    if len(optim_files) == 0:
+        raise FileNotFoundError(
+            f"can't find '*_optim_states.pt' files in directory '{checkpoint_dir}'")
+
+    return optim_files
+
+
+def parse_model_state(file):
+    state_dict = torch.load(file, map_location=device)
+
+    if BUFFER_NAMES not in state_dict:
+        raise ValueError(f"{file} is not a model state checkpoint")
+    buffer_names = state_dict[BUFFER_NAMES]
+    if debug:
+        print("Found buffers:", buffer_names)
+
+    # recover just the buffers while restoring them to fp32 if they were saved in fp16
+    buffers = {
+        k: v.float()
+        for k,
+        v in state_dict["module"].items() if k in buffer_names
+    }
+    param_shapes = state_dict[PARAM_SHAPES]
+
+    ds_version = state_dict.get(DS_VERSION, None)
+
+    return buffers, param_shapes, ds_version
+
+
+def parse_optim_states(files, ds_checkpoint_dir):
+
+    total_files = len(files)
+    state_dicts = []
+    for f in files:
+        state_dicts.append(torch.load(f, map_location=device))
+
+    if not ZERO_STAGE in state_dicts[0][OPTIMIZER_STATE_DICT]:
+        raise ValueError(f"{files[0]} is not a zero checkpoint")
+    zero_stage = state_dicts[0][OPTIMIZER_STATE_DICT][ZERO_STAGE]
+    world_size = state_dicts[0][OPTIMIZER_STATE_DICT][PARTITION_COUNT]
+
+    # For ZeRO-2 each param group can have different partition_count as data parallelism for expert
+    # parameters can be different from data parallelism for non-expert parameters. So we can just
+    # use the max of the partition_count to get the dp world_size.
+
+    if type(world_size) is list:
+        world_size = max(world_size)
+
+    if world_size != total_files:
+        raise ValueError(
+            f"Expected {world_size} of '*_optim_states.pt' under '{ds_checkpoint_dir}' but found {total_files} files. "
+            "Possibly due to an overwrite of an old checkpoint, or a checkpoint didn't get saved by one or more processes."
+        )
+
+    # the groups are named differently in each stage
+    if zero_stage == 2:
+        fp32_groups_key = SINGLE_PARTITION_OF_FP32_GROUPS
+    elif zero_stage == 3:
+        fp32_groups_key = FP32_FLAT_GROUPS
+    else:
+        raise ValueError(f"unknown zero stage {zero_stage}")
+
+    if zero_stage == 2:
+        fp32_flat_groups = [
+            state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key]
+            for i in range(len(state_dicts))
+        ]
+    elif zero_stage == 3:
+        # if there is more than one param group, there will be multiple flattened tensors - one
+        # flattened tensor per group - for simplicity merge them into a single tensor
+        #
+        # XXX: could make the script more memory efficient for when there are multiple groups - it
+        # will require matching the sub-lists of param_shapes for each param group flattened tensor
+
+        fp32_flat_groups = [
+            torch.cat(state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key],
+                      0) for i in range(len(state_dicts))
+        ]
+
+    return zero_stage, world_size, fp32_flat_groups
+
+
+def _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir):
+    """
+    Returns fp32 state_dict reconstructed from ds checkpoint
+
+    Args:
+        - ``ds_checkpoint_dir``: path to the deepspeed checkpoint folder (where the optimizer files are)
+
+    """
+    print(f"Processing zero checkpoint '{ds_checkpoint_dir}'")
+
+    optim_files = get_optim_files(ds_checkpoint_dir)
+    zero_stage, world_size, fp32_flat_groups = parse_optim_states(optim_files, ds_checkpoint_dir)
+    print(
+        f"Detected checkpoint of type zero stage {zero_stage}, world_size: {world_size}")
+
+    model_file = get_model_state_file(ds_checkpoint_dir, zero_stage)
+    buffers, param_shapes, ds_version = parse_model_state(model_file)
+    print(f'Parsing checkpoint created by deepspeed=={ds_version}')
+
+    if zero_stage == 2:
+        return _get_fp32_state_dict_from_zero2_checkpoint(world_size,
+                                                          param_shapes,
+                                                          fp32_flat_groups,
+                                                          buffers)
+    elif zero_stage == 3:
+        return _get_fp32_state_dict_from_zero3_checkpoint(world_size,
+                                                          param_shapes,
+                                                          fp32_flat_groups,
+                                                          buffers)
+
+
+def _get_fp32_state_dict_from_zero2_checkpoint(world_size,
+                                               param_shapes,
+                                               fp32_flat_groups,
+                                               buffers):
+
+    # Reconstruction protocol:
+    #
+    # XXX: document this
+
+    if debug:
+        for i in range(world_size):
+            for j in range(len(fp32_flat_groups[0])):
+                print(
+                    f"{FP32_FLAT_GROUPS}[{i}][{j}].shape={fp32_flat_groups[i][j].shape}")
+
+    # XXX: memory usage doubles here (zero2)
+    num_param_groups = len(fp32_flat_groups[0])
+    merged_single_partition_of_fp32_groups = []
+    for i in range(num_param_groups):
+        merged_partitions = [sd[i] for sd in fp32_flat_groups]
+        full_single_fp32_vector = torch.cat(merged_partitions, 0)
+        merged_single_partition_of_fp32_groups.append(full_single_fp32_vector)
+    avail_numel = sum([
+        full_single_fp32_vector.numel()
+        for full_single_fp32_vector in merged_single_partition_of_fp32_groups
+    ])
+
+    if debug:
+        wanted_params = sum([len(shapes) for shapes in param_shapes])
+        wanted_numel = sum(
+            [sum(shape.numel() for shape in shapes.values()) for shapes in param_shapes])
+        # not asserting if there is a mismatch due to possible padding
+        print(f"Have {avail_numel} numels to process.")
+        print(f"Need {wanted_numel} numels in {wanted_params} params.")
+
+    state_dict = OrderedDict()
+
+    # buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    # params
+    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+    # out-of-core computing solution
+    total_numel = 0
+    total_params = 0
+    for shapes, full_single_fp32_vector in zip(param_shapes, merged_single_partition_of_fp32_groups):
+        offset = 0
+        avail_numel = full_single_fp32_vector.numel()
+        for name, shape in shapes.items():
+
+            unpartitioned_numel = shape.numel()
+            total_numel += unpartitioned_numel
+            total_params += 1
+
+            if debug:
+                print(
+                    f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} "
+                )
+            state_dict[name] = full_single_fp32_vector.narrow(
+                0,
+                offset,
+                unpartitioned_numel).view(shape)
+            offset += unpartitioned_numel
+
+        # Z2 started to align to 2*world_size to improve nccl performance. Therefore both offset and
+        # avail_numel can differ by anywhere between 0..2*world_size. Due to two unrelated complex
+        # paddings performed in the code it's almost impossible to predict the exact numbers w/o the
+        # live optimizer object, so we are checking that the numbers are within the right range
+        align_to = 2 * world_size
+
+        def zero2_align(x):
+            return align_to * math.ceil(x / align_to)
+
+        if debug:
+            print(f"original offset={offset}, avail_numel={avail_numel}")
+
+        offset = zero2_align(offset)
+        avail_numel = zero2_align(avail_numel)
+
+        if debug:
+            print(f"aligned  offset={offset}, avail_numel={avail_numel}")
+
+        # Sanity check
+        if offset != avail_numel:
+            raise ValueError(
+                f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+    print(
+        f"Reconstructed fp32 state dict with {total_params} params {total_numel} elements"
+    )
+
+    return state_dict
+
+
+def zero3_partitioned_param_info(unpartitioned_numel, world_size):
+    remainder = unpartitioned_numel % world_size
+    padding_numel = (world_size - remainder) if remainder else 0
+    partitioned_numel = math.ceil(unpartitioned_numel / world_size)
+    return partitioned_numel, padding_numel
+
+
+def _get_fp32_state_dict_from_zero3_checkpoint(world_size,
+                                               param_shapes,
+                                               fp32_flat_groups,
+                                               buffers):
+
+    # Reconstruction protocol: For zero3 we need to zip the partitions together at boundary of each
+    # param, re-consolidating each param, while dealing with padding if any
+
+    avail_numel = fp32_flat_groups[0].numel() * world_size
+    # merge list of dicts, preserving order
+    param_shapes = {k: v for d in param_shapes for k, v in d.items()}
+
+    if debug:
+        for i in range(world_size):
+            print(f"{FP32_FLAT_GROUPS}[{i}].shape={fp32_flat_groups[i].shape}")
+
+        wanted_params = len(param_shapes)
+        wanted_numel = sum(shape.numel() for shape in param_shapes.values())
+        # not asserting if there is a mismatch due to possible padding
+        print(f"Have {avail_numel} numels to process.")
+        print(f"Need {wanted_numel} numels in {wanted_params} params.")
+
+    state_dict = OrderedDict()
+
+    # buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    # params
+    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+    # out-of-core computing solution
+    offset = 0
+    total_numel = 0
+    total_params = 0
+    for name, shape in param_shapes.items():
+
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+        total_params += 1
+
+        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
+
+        if debug:
+            print(
+                f"{total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+            )
+
+        # XXX: memory usage doubles here
+        state_dict[name] = torch.cat(
+            tuple(fp32_flat_groups[i].narrow(0,
+                                             offset,
+                                             partitioned_numel)
+                  for i in range(world_size)),
+            0).narrow(0,
+                      0,
+                      unpartitioned_numel).view(shape)
+        offset += partitioned_numel
+
+    offset *= world_size
+
+    # Sanity check
+    if offset != avail_numel:
+        raise ValueError(
+            f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+    print(
+        f"Reconstructed fp32 state dict with {total_params} params {total_numel} elements"
+    )
+
+    return state_dict
+
+
+def get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag=None):
+    """
+    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated state_dict that can be loaded with
+    ``load_state_dict()`` and used for training without DeepSpeed or shared with others, for example
+    via a model hub.
+
+    Args:
+        - ``checkpoint_dir``: path to the desired checkpoint folder
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in 'latest' file. e.g., ``global_step14``
+
+    Returns:
+        - pytorch ``state_dict``
+
+    Note: this approach may not work if your application doesn't have sufficient free CPU memory and
+    you may need to use the offline approach using the ``zero_to_fp32.py`` script that is saved with
+    the checkpoint.
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
+        # do the training and checkpoint saving
+        state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir) # already on cpu
+        model = model.cpu() # move to cpu
+        model.load_state_dict(state_dict)
+        # submit to model hub or save the model to share with others
+
+    In this example the ``model`` will no longer be usable in the deepspeed context of the same
+    application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    If you want it all done for you, use ``load_state_dict_from_zero_checkpoint`` instead.
+
+    """
+    if tag is None:
+        latest_path = os.path.join(checkpoint_dir, 'latest')
+        if os.path.isfile(latest_path):
+            with open(latest_path, 'r') as fd:
+                tag = fd.read().strip()
+        else:
+            raise ValueError(f"Unable to find 'latest' file at {latest_path}")
+
+    ds_checkpoint_dir = os.path.join(checkpoint_dir, tag)
+
+    if not os.path.isdir(ds_checkpoint_dir):
+        raise FileNotFoundError(f"Directory '{ds_checkpoint_dir}' doesn't exist")
+
+    return _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir)
+
+
+def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir, output_file, tag=None):
+    """
+    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict`` file that can be
+    loaded with ``torch.load(file)`` + ``load_state_dict()`` and used for training without DeepSpeed.
+
+    Args:
+        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+        - ``output_file``: path to the pytorch fp32 state_dict output file (e.g. path/pytorch_model.bin)
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+    """
+
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
+    print(f"Saving fp32 state dict to {output_file}")
+    torch.save(state_dict, output_file)
+
+
+def load_state_dict_from_zero_checkpoint(model, checkpoint_dir, tag=None):
+    """
+    1. Put the provided model to cpu
+    2. Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict``
+    3. Load it into the provided model
+
+    Args:
+        - ``model``: the model object to update
+        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+
+    Returns:
+        - ``model`: modified model
+
+    Make sure you have plenty of CPU memory available before you call this function. If you don't
+    have enough use the ``zero_to_fp32.py`` utility to do the conversion. You will find it
+    conveniently placed for you in the checkpoint folder.
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import load_state_dict_from_zero_checkpoint
+        model = load_state_dict_from_zero_checkpoint(trainer.model, checkpoint_dir)
+        # submit to model hub or save the model to share with others
+
+    Note, that once this was run, the ``model`` will no longer be usable in the deepspeed context
+    of the same application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    """
+    logger.info(f"Extracting fp32 weights")
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
+
+    logger.info(f"Overwriting model with fp32 weights")
+    model = model.cpu()
+    model.load_state_dict(state_dict, strict=False)
+
+    return model
+
+
+if __name__ == "__main__":
+
+    parser = argparse.ArgumentParser()
+    parser.add_argument(
+        "checkpoint_dir",
+        type=str,
+        help="path to the desired checkpoint folder, e.g., path/checkpoint-12")
+    parser.add_argument(
+        "output_file",
+        type=str,
+        help=
+        "path to the pytorch fp32 state_dict output file (e.g. path/checkpoint-12/pytorch_model.bin)"
+    )
+    parser.add_argument("-d", "--debug", action='store_true', help="enable debug")
+    args = parser.parse_args()
+
+    debug = args.debug
+
+    convert_zero_checkpoint_to_fp32_state_dict(args.checkpoint_dir, args.output_file)