Add new BERT calibration dataset

calibration/SQuAD-v1.1/README.md

@@ -1 +1,8 @@
-The calibration file has 100 randomly selected samples from dev-1.1.json, which contains 10570 samples in total.
+The calibration files each contain 100 randomly selected samples from dev-1.1.json, which contains 10570 samples in total.
+
+The integers in bert_calibration_features.txt map to indices in the dev-v1.1.json dataset after being converted to squad features by calling `convert_examples_to_features` in `language/bert/create_squad_data.py`.
+
+The values in bert_calibration_qas_ids.txt map to qas ids in the dev-v1.1.json file.
+
+Please only use at most 1 calibration file from this folder for calibration.
+

calibration/SQuAD-v1.1/bert_calibration_qas_ids.txt

@@ -0,0 +1,100 @@
+57268e2bf1498d1400e8e3b4
+57297781af94a219006aa4a3
+5725c95f38643c19005accf7
+56d9c92bdc89441400fdb80f
+5726bc505951b619008f7c7c
+57283c464b864d19001647c8
+5728e3c33acd2414000e0135
+56e1262fcd28a01900c67657
+57265285708984140094c25e
+572a1046af94a219006aa790
+572fd264b2c2fd14005684ab
+5726eb8bf1498d1400e8efe4
+572a06af3f37b31900478669
+5725f2c838643c19005acef1
+56e17a7ccd28a01900c679a5
+573027d6a23a5019007fce9f
+573085ea8ab72b1400f9c54e
+573408ef4776f41900661758
+56beb86b3aeaaa14008c92c0
+56e08a1f7aa994140058e5db
+57373d0cc3c5551400e51e87
+571142b3a58dae1900cd6d5a
+56e74bf937bdd419002c3e35
+57264f18f1498d1400e8dbae
+5726a299dd62a815002e8b9e
+572a1f086aef0514001552c2
+5726bc505951b619008f7c7a
+5730b541396df919000962c6
+5726af765951b619008f7a52
+572885c44b864d1900164a7a
+56f8bbb09e9bad19000a03dd
+57286192ff5b5019007da1e0
+572824f13acd2414000df591
+572920d73f37b31900478058
+56dfa24a38dc421700152145
+5711541350c2381900b54a70
+56d7282f0d65d21400198409
+5737898f1c45671900574497
+572fc49d04bcaa1900d76ccf
+56e19724cd28a01900c679f8
+573011de04bcaa1900d770f9
+5727cd7dff5b5019007d958f
+571153422419e31400955580
+5727580bf1498d1400e8f69d
+56d98f0ddc89441400fdb559
+57287f6a3acd2414000dfa51
+5711648850c2381900b54ac6
+5729f799af94a219006aa709
+572917743f37b3190047800d
+5725d0e3271a42140099d237
+56bebc383aeaaa14008c9321
+56e08a1f7aa994140058e5dd
+572ff7ab04bcaa1900d76f51
+57094b4f9928a814004714f7
+572659ea5951b619008f7054
+57293f8a6aef051400154bdf
+57274a1edd62a815002e9a9a
+56dfad0a7aa994140058dfc7
+572fffb1b2c2fd14005686fc
+5726a9ff708984140094cd4f
+572fc80b04bcaa1900d76d13
+56d728e00d65d2140019841d
+572883153acd2414000dfa72
+5726701add62a815002e84c9
+57108d69b654c5140001f985
+571155ae2419e31400955591
+57264cfa708984140094c1c4
+5730b1e28ab72b1400f9c6af
+572a03086aef0514001551a4
+572835854b864d1900164730
+5730040f947a6a140053cf4d
+570d2556fed7b91900d45c48
+57269e8a5951b619008f77f6
+572a13841d0469140077973f
+5726bdc6f1498d1400e8e9c3
+572976183f37b31900478432
+5729de24af94a219006aa68c
+571c9074dd7acb1400e4c103
+56f8c7029e9bad19000a04a2
+57340d124776f419006617c0
+57309d31396df91900096213
+572fce13a23a5019007fca14
+56e1c2eee3433e1400423138
+572648d1708984140094c160
+5725f6e138643c19005acf1c
+572a13841d0469140077973d
+57268882f1498d1400e8e306
+57269aa65951b619008f77ae
+572fad30a23a5019007fc871
+56f851b1a6d7ea1400e17561
+5730035e04bcaa1900d77005
+572ff293947a6a140053ce52
+56f881e3a6d7ea1400e176f7
+56e1cbe2cd28a01900c67bb0
+56e17e6ee3433e1400422f80
+573085ea8ab72b1400f9c54f
+5729f9953f37b31900478621
+572fcb6da23a5019007fc9f4
+57283a392ca10214002da119
+56e763e800c9c71400d77089