forked from princeton-nlp/DensePhrases
-
Notifications
You must be signed in to change notification settings - Fork 2
/
Makefile
532 lines (478 loc) · 18.1 KB
/
Makefile
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
############################## Single-passage Training + Normalization ###################################
model-name:
ifeq ($(MODEL_NAME),)
echo "Please set MODEL_NAME before training (e.g., MODEL_NAME=test)"; exit 2;
endif
# Dataset paths for single-passage training (QG, train, dev, semi-od)
nq-rc-data:
$(eval TRAIN_QG_DATA=nq/train_wiki3_na_filtered_qg_t5l35-sqd_filtered.json)
$(eval TRAIN_DATA=nq/train_wiki3.json)
$(eval DEV_DATA=nq/dev_wiki3.json)
$(eval SOD_DATA=open-qa/nq-open/dev_wiki3_open.json)
$(eval OPTIONS=--truecase)
sqd-rc-data:
$(eval TRAIN_QG_DATA=squad/train-v1.1_qg_ents_t5large_3500_filtered.json)
$(eval TRAIN_DATA=squad/train-v1.1.json)
$(eval DEV_DATA=squad/dev-v1.1.json)
$(eval SOD_DATA=open-qa/squad/test_preprocessed.json)
nqsqd-rc-data:
$(eval TRAIN_QG_DATA=squad-nq/train-sqdqg_nqqg_filtered.json)
$(eval TRAIN_DATA=squad-nq/train-sqd_nq.json)
$(eval DEV_DATA=nq/dev_wiki3.json)
$(eval SOD_DATA=open-qa/nq-open/dev_wiki3_open.json)
$(eval OPTIONS=--truecase)
# Choose hyperparameter
pbn-param:
$(eval PBN_OPTIONS=--pbn_size 2 --pbn_tolerance 0)
nq-param:
$(eval BS=64)
$(eval LR=1e-4)
$(eval MAX_SEQ_LEN=192)
$(eval LAMBDA_KL=2.0)
$(eval LAMBDA_NEG=4.0)
$(eval TEACHER_NAME=spanbert-base-cased-nq)
sqd-param:
$(eval BS=24)
$(eval LR=3e-5)
$(eval MAX_SEQ_LEN=384)
$(eval LAMBDA_KL=4.0)
$(eval LAMBDA_NEG=2.0)
$(eval TEACHER_NAME=spanbert-base-cased-squad)
nqsqd-param:
$(eval BS=64)
$(eval LR=1e-4)
$(eval MAX_SEQ_LEN=192)
$(eval LAMBDA_KL=4.0)
$(eval LAMBDA_NEG=4.0)
$(eval TEACHER_NAME=spanbert-base-cased-sqdnq)
# Choose index size
small-index:
$(eval NUM_CLUSTERS=256)
$(eval INDEX_TYPE=OPQ96)
medium1-index:
$(eval NUM_CLUSTERS=16384)
$(eval INDEX_TYPE=OPQ96)
medium2-index:
$(eval NUM_CLUSTERS=131072)
$(eval INDEX_TYPE=OPQ96)
large-index:
$(eval NUM_CLUSTERS=1048576)
$(eval INDEX_TYPE=OPQ96)
large-index-sq:
$(eval NUM_CLUSTERS=1048576)
$(eval INDEX_TYPE=SQ4)
# Followings are template commands. See 'run-rc-nq' for a detailed use.
# 1) Training phrase and question encoders on reading comprehension.
train-rc:
python train_rc.py \
--model_type bert \
--pretrained_name_or_path SpanBERT/spanbert-base-cased \
--data_dir $(DATA_DIR)/single-qa \
--cache_dir $(CACHE_DIR) \
--train_file $(TRAIN_DATA) \
--predict_file $(DEV_DATA) \
--do_train \
--do_eval \
--per_gpu_train_batch_size $(BS) \
--learning_rate $(LR) \
--num_train_epochs 2.0 \
--max_seq_length $(MAX_SEQ_LEN) \
--fp16 \
--lambda_kl $(LAMBDA_KL) \
--lambda_neg $(LAMBDA_NEG) \
--lambda_flt 1.0 \
--filter_threshold -2.0 \
--append_title \
--evaluate_during_training \
--teacher_dir $(SAVE_DIR)/$(TEACHER_NAME) \
--output_dir $(SAVE_DIR)/$(MODEL_NAME) \
$(OPTIONS)
# 2) Trained phrase encoders can be used to generate phrase vectors
gen-vecs:
python generate_phrase_vecs.py \
--model_type bert \
--pretrained_name_or_path SpanBERT/spanbert-base-cased \
--data_dir $(DATA_DIR)/single-qa \
--cache_dir $(CACHE_DIR) \
--predict_file $(DEV_DATA) \
--do_dump \
--max_seq_length 512 \
--doc_stride 500 \
--fp16 \
--filter_threshold -2.0 \
--append_title \
--load_dir $(SAVE_DIR)/$(MODEL_NAME) \
--output_dir $(SAVE_DIR)/$(MODEL_NAME) \
$(OPTIONS)
# 3) Build an IVFOPQ index for generated phrase vectors
index-vecs: dump-dir medium1-index
python build_phrase_index.py \
$(DUMP_DIR) all \
--replace \
--num_clusters $(NUM_CLUSTERS) \
--fine_quant $(INDEX_TYPE) \
--cuda
# 4) Compress metadata
compress-meta:
python scripts/preprocess/compress_metadata.py \
--input_dump_dir $(DUMP_DIR)/phrase \
--output_dir $(DUMP_DIR)
# 5) Evaluate the phrase index for phrase retrieval
eval-index: dump-dir model-name large-index nq-open-data
python eval_phrase_retrieval.py \
--run_mode eval \
--cuda \
--dump_dir $(DUMP_DIR) \
--index_dir start/$(NUM_CLUSTERS)_flat_$(INDEX_TYPE) \
--query_encoder_path $(SAVE_DIR)/$(MODEL_NAME) \
--test_path $(DATA_DIR)/$(TEST_DATA) \
--save_pred \
--aggregate \
$(OPTIONS)
# Sample usage (If this runs without an error, you are all set!)
draft: model-name nq-rc-data nq-param pbn-param small-index
make train-rc \
TRAIN_DATA=$(TRAIN_DATA) DEV_DATA=$(DEV_DATA) \
TEACHER_NAME=$(TEACHER_NAME) MODEL_NAME=$(MODEL_NAME) \
BS=$(BS) LR=$(LR) MAX_SEQ_LEN=$(MAX_SEQ_LEN) \
LAMBDA_KL=$(LAMBDA_KL) LAMBDA_NEG=$(LAMBDA_NEG) \
OPTIONS='$(PBN_OPTIONS) --draft'
make gen-vecs \
DEV_DATA=$(DEV_DATA) MODEL_NAME=$(MODEL_NAME)
make index-vecs \
DUMP_DIR=$(SAVE_DIR)/$(MODEL_NAME)/dump \
NUM_CLUSTERS=$(NUM_CLUSTERS) INDEX_TYPE=$(INDEX_TYPE)
make compress-meta \
DUMP_DIR=$(SAVE_DIR)/$(MODEL_NAME)/dump
make eval-index \
DUMP_DIR=$(SAVE_DIR)/$(MODEL_NAME)/dump \
NUM_CLUSTERS=$(NUM_CLUSTERS) INDEX_TYPE=$(INDEX_TYPE) \
MODEL_LANE=$(MODEL_NAME) TEST_DATA=$(SOD_DATA) \
OPTIONS=$(OPTIONS)
# Single-passage training + additional negatives for NQ
# Available datasets: NQ (nq-rc-data), SQuAD (sqd-rc-data), NQ+SQuAD (nqsqd-rc-data)
# Should change hyperparams (e.g., nq-param) accordingly
run-rc-nq: model-name nq-rc-data nq-param pbn-param small-index
make train-rc \
TRAIN_DATA=$(TRAIN_QG_DATA) DEV_DATA=$(DEV_DATA) \
TEACHER_NAME=$(TEACHER_NAME) MODEL_NAME=$(MODEL_NAME)_tmp \
BS=$(BS) LR=$(LR) MAX_SEQ_LEN=$(MAX_SEQ_LEN) \
LAMBDA_KL=$(LAMBDA_KL) LAMBDA_NEG=$(LAMBDA_NEG)
make train-rc \
TRAIN_DATA=$(TRAIN_DATA) DEV_DATA=$(DEV_DATA) \
TEACHER_NAME=$(TEACHER_NAME) MODEL_NAME=$(MODEL_NAME) \
BS=$(BS) LR=$(LR) MAX_SEQ_LEN=$(MAX_SEQ_LEN) \
LAMBDA_KL=$(LAMBDA_KL) LAMBDA_NEG=$(LAMBDA_NEG) \
OPTIONS='$(PBN_OPTIONS) --load_dir $(SAVE_DIR)/$(MODEL_NAME)_tmp'
make gen-vecs \
DEV_DATA=$(DEV_DATA) MODEL_NAME=$(MODEL_NAME)
make index-vecs \
DUMP_DIR=$(SAVE_DIR)/$(MODEL_NAME)/dump \
NUM_CLUSTERS=$(NUM_CLUSTERS) INDEX_TYPE=$(INDEX_TYPE)
make compress-meta \
DUMP_DIR=$(SAVE_DIR)/$(MODEL_NAME)/dump
make eval-index \
DUMP_DIR=$(SAVE_DIR)/$(MODEL_NAME)/dump \
NUM_CLUSTERS=$(NUM_CLUSTERS) INDEX_TYPE=$(INDEX_TYPE) \
MODEL_LANE=$(MODEL_NAME) TEST_DATA=$(SOD_DATA) \
OPTIONS=$(OPTIONS)
# Testing filter thresholds
filter-test: model-name nq-rc-data
python train_rc.py \
--model_type bert \
--pretrained_name_or_path SpanBERT/spanbert-base-cased \
--data_dir $(DATA_DIR)/single-qa \
--cache_dir $(CACHE_DIR) \
--predict_file $(DEV_DATA) \
--do_filter_test \
--append_title \
--filter_threshold_list " -4,-3,-2,-1,-0.5,0" \
--load_dir $(SAVE_DIR)/$(MODEL_NAME) \
--output_dir $(SAVE_DIR)/$(MODEL_NAME) \
--draft
############################## Large-scale Dump & Indexing ###############################
dump-dir:
ifeq ($(DUMP_DIR),)
echo "Please set DUMP_DIR before dumping/indexing (e.g., DUMP_DIR=test)"; exit 2;
endif
# Wikipedia dumps (specified as 'data_name') in diffent sizes and their recommended number of clusters for IVF
# - wiki-dev: 1/100 Wikpedia scale (sampled), num_clusters=16384 (medium1-index)
# - wiki-dev-noise: 1/10 Wikipedia scale (sampled), num_clusters=131072 (medium2-index)
# - wiki-20181220: full Wikipedia scale, num_clusters=1048576 (large-index)
# Dump phrase vectors in parallel. Dump will be saved in $(SAVE_DIR)/$(MODEL_NAME)_(data_name)/dump.
gen-vecs-parallel: model-name
nohup python scripts/parallel/dump_phrases.py \
--model_type bert \
--pretrained_name_or_path SpanBERT/spanbert-base-cased \
--cache_dir $(CACHE_DIR) \
--data_dir $(DATA_DIR)/wikidump \
--data_name wiki-dev \
--load_dir $(SAVE_DIR)/$(MODEL_NAME) \
--output_dir $(SAVE_DIR)/$(MODEL_NAME) \
--filter_threshold 1.0 \
--append_title \
--start $(START) \
--end $(END) \
> $(SAVE_DIR)/logs/$(MODEL_NAME)_$(START)-$(END).log &
# Parallel add for large-scale on-disk IVFSQ (start, end = file idx)
index-add: dump-dir large-index-sq
export MKL_SERVICE_FORCE_INTEL=1
python scripts/parallel/add_to_index.py \
--dump_dir $(DUMP_DIR) \
--num_clusters $(NUM_CLUSTERS) \
--cuda \
--start $(START) \
--end $(END)
# Merge for large-scale on-disk IVFSQ
index-merge: dump-dir large-index-sq
python build_phrase_index.py \
$(DUMP_DIR) merge \
--replace \
--num_clusters $(NUM_CLUSTERS) \
--fine_quant $(INDEX_TYPE)
############################## Open-domain Search & Query-side Fine-tuning ###################################
# Dataset paths for open-domain QA and slot filling (with options)
nq-open-data:
$(eval TRAIN_DATA=open-qa/nq-open/train_preprocessed.json)
$(eval DEV_DATA=open-qa/nq-open/dev_preprocessed.json)
$(eval TEST_DATA=open-qa/nq-open/test_preprocessed.json)
$(eval OPTIONS=--truecase)
wq-open-data:
$(eval TRAIN_DATA=open-qa/webq/WebQuestions-train-nodev_preprocessed.json)
$(eval DEV_DATA=open-qa/webq/WebQuestions-dev_preprocessed.json)
$(eval TEST_DATA=open-qa/webq/WebQuestions-test_preprocessed.json)
$(eval OPTIONS=--truecase --candidate_path $(DATA_DIR)/open-qa/webq/freebase-entities.txt)
trec-open-data:
$(eval TRAIN_DATA=open-qa/trec/CuratedTrec-train-nodev_preprocessed.json)
$(eval DEV_DATA=open-qa/trec/CuratedTrec-dev_preprocessed.json)
$(eval TEST_DATA=open-qa/trec/CuratedTrec-test_preprocessed.json)
$(eval OPTIONS=--regex)
tqa-open-data:
$(eval TRAIN_DATA=open-qa/triviaqa-unfiltered/train_preprocessed.json)
$(eval DEV_DATA=open-qa/triviaqa-unfiltered/dev_preprocessed.json)
$(eval TEST_DATA=open-qa/triviaqa-unfiltered/test_preprocessed.json)
sqd-open-data:
$(eval TRAIN_DATA=open-qa/squad/train_preprocessed.json)
$(eval DEV_DATA=open-qa/squad/dev_preprocessed.json)
$(eval TEST_DATA=open-qa/squad/test_preprocessed.json)
kilt-options:
$(eval OPTIONS=--is_kilt --title2wikiid_path $(DATA_DIR)/wikidump/title2wikiid.json)
trex-open-data: kilt-options
$(eval TRAIN_DATA=kilt/trex/trex-train-kilt_open_10000.json)
$(eval DEV_DATA=kilt/trex/trex-dev-kilt_open.json)
$(eval TEST_DATA=kilt/trex/trex-dev-kilt_open.json)
$(eval OPTIONS=$(OPTIONS) --kilt_gold_path $(DATA_DIR)/kilt/trex/trex-dev-kilt.jsonl --agg_strat opt2)
zsre-open-data: kilt-options
$(eval TRAIN_DATA=kilt/zsre/structured_zeroshot-train-kilt_open_10000.json)
$(eval DEV_DATA=kilt/zsre/structured_zeroshot-dev-kilt_open.json)
$(eval TEST_DATA=kilt/zsre/structured_zeroshot-dev-kilt_open.json)
$(eval OPTIONS=$(OPTIONS) --kilt_gold_path $(DATA_DIR)/kilt/zsre/structured_zeroshot-dev-kilt.jsonl --agg_strat opt2)
benchmark-data:
$(eval TEST_DATA=scripts/benchmark/data/nq_1000_dev_denspi.json)
train-query: dump-dir model-name trex-open-data large-index
python train_query.py \
--run_mode train_query \
--cache_dir $(CACHE_DIR) \
--train_path $(DATA_DIR)/$(TRAIN_DATA) \
--dev_path $(DATA_DIR)/$(DEV_DATA) \
--test_path $(DATA_DIR)/$(TEST_DATA) \
--per_gpu_train_batch_size 12 \
--eval_batch_size 12 \
--learning_rate 3e-5 \
--num_train_epochs 5 \
--dump_dir $(DUMP_DIR) \
--index_dir start/$(NUM_CLUSTERS)_flat_$(INDEX_TYPE) \
--query_encoder_path $(SAVE_DIR)/densephrases-multi \
--output_dir $(SAVE_DIR)/$(MODEL_NAME) \
--top_k 100 \
--cuda \
$(OPTIONS)
################################ Demo Serving ###################################
# Serve question encoder
q-serve:
nohup python run_demo.py \
--run_mode q_serve \
--cache_dir $(CACHE_DIR) \
--query_encoder_path $(SAVE_DIR)/$(MODEL_NAME) \
--cuda \
--max_query_length 32 \
--query_port $(Q_PORT) > $(SAVE_DIR)/logs/q-serve_$(Q_PORT).log &
# Serve phrase index (Q_PORT may change)
p-serve: dump-dir large-index
nohup python run_demo.py \
--run_mode p_serve \
--index_dir start/$(NUM_CLUSTERS)_flat_$(INDEX_TYPE) \
--cuda \
--truecase \
--dump_dir $(DUMP_DIR) \
--query_port $(Q_PORT) \
--index_port $(I_PORT) > $(SAVE_DIR)/logs/p-serve_$(I_PORT).log &
# Evaluation using the open QA demo (used for benchmark)
eval-demo: nq-open-data
python run_demo.py \
--run_mode eval_request \
--index_port $(I_PORT) \
--test_path $(DATA_DIR)/$(TEST_DATA) \
--eval_batch_size 64 \
--save_pred \
$(OPTIONS)
# (Optional) Serve single-passage RC demo
single-serve:
nohup python run_demo.py \
--run_mode single_serve \
--cuda \
--cache_dir $(CACHE_DIR) \
--query_encoder_path $(SAVE_DIR)/$(MODEL_NAME) \
--query_port $(Q_PORT) > $(SAVE_DIR)/logs/s-serve_$(Q_PORT).log &
############################## Data Pre/Post-processing ###################################
preprocess-openqa:
python scripts/preprocess/create_openqa.py \
$(DATA_DIR)/single-qa/squad/train-v1.1.json \
$(DATA_DIR)/open-qa/squad \
--input_type SQuAD
# Warning: many scripts below are not documented well.
# Each script may rely on external resources (e.g., original NQ datasets).
data-config:
$(eval NQORIG_DIR=$(DATA_DIR)/natural-questions)
$(eval NQOPEN_DIR=$(DATA_DIR)/nq-open)
$(eval NQREADER_DIR=$(DATA_DIR)/single-qa/nq)
$(eval SQUAD_DIR=$(DATA_DIR)/single-qa/squad)
$(eval SQUADREADER_DOC_DIR=$(DATA_DIR)/squad-reader-docs)
$(eval NQREADER_DOC_DIR=$(DATA_DIR)/nq-reader-docs)
$(eval WIKI_DIR=$(DATA_DIR)/wikidump)
nq-reader-to-wiki:
python scripts/preprocess/create_nq_reader_wiki.py \
$(DATA_DIR)/single-qa/nq/train.json,$(DATA_DIR)/single-qa/nq/dev.json \
$(DATA_DIR)/single-qa/nq \
$(DATA_DIR)/wikidump/20181220_concat/
download-wiki: data-config
python scripts/preprocess/download_wikidump.py \
--output_dir $(WIKI_DIR)
nq-reader-train: data-config
python scripts/preprocess/create_nq_reader.py \
--nq_orig_path_pattern "$(NQORIG_DIR)/train/nq-train-*.jsonl.gz" \
--nq_open_path $(NQOPEN_DIR)/train.json \
--output_path $(NQREADER_DIR)/train_79168.json
nq-reader-dev: data-config
python scripts/preprocess/create_nq_reader.py \
--nq_orig_path_pattern "$(NQORIG_DIR)/train/nq-train-*.jsonl.gz" \
--nq_open_path $(NQOPEN_DIR)/dev.json \
--output_path $(NQREADER_DIR)/dev_8757.json
nq-reader-dev-sample: data-config
python scripts/preprocess/create_nq_reader.py \
--nq_orig_path_pattern "$(NQORIG_DIR)/sample/nq-train-sample.jsonl.gz" \
--nq_open_path $(NQOPEN_DIR)/dev.json \
--output_path $(NQREADER_DIR)/dev_sample.json
nq-reader-train-docs: data-config
python scripts/preprocess/create_nq_reader_doc.py \
--nq_orig_path_pattern "$(NQORIG_DIR)/train/nq-train-*.jsonl.gz" \
--nq_open_path $(NQOPEN_DIR)/train.json \
--output_dir $(NQREADER_DOC_DIR)/train
nq-reader-dev-docs: data-config
python scripts/preprocess/create_nq_reader_doc.py \
--nq_orig_path_pattern "$(NQORIG_DIR)/train/nq-train-*.jsonl.gz" \
--nq_open_path $(NQOPEN_DIR)/dev.json \
--output_dir $(NQREADER_DOC_DIR)/dev
nq-reader-dev-docs-sample: data-config
python scripts/preprocess/create_nq_reader_doc.py \
--nq_orig_path_pattern "$(NQORIG_DIR)/sample/nq-train-sample.jsonl.gz" \
--nq_open_path $(NQOPEN_DIR)/dev.json \
--output_dir $(NQREADER_DOC_DIR)-sample
nq-_reader-train-docs-wiki: data-config
python scripts/preprocess/create_nq_reader_doc_wiki.py \
--wiki_dir $(WIKI_DIR)/20181220_concat \
--nq_reader_docs_dir $(NQREADER_DOC_DIR)/train \
--output_dir $(NQREADER_DOC_DIR)/train_wiki
nq-reader-dev-docs-wiki: data-config
python scripts/preprocess/create_nq_reader_doc_wiki.py \
--wiki_dir $(WIKI_DIR)/20181220_concat \
--nq_reader_docs_dir $(NQREADER_DOC_DIR)/dev \
--output_dir $(NQREADER_DOC_DIR)/dev_wiki
squad-reader-train-docs-wiki: data-config
python scripts/preprocess/create_nq_reader_doc_wiki.py \
--wiki_dir $(WIKI_DIR)/20181220_concat \
--nq_reader_docs_dir $(SQUAD_DIR)/train-v1.1.json \
--output_dir $(SQUADREADER_DOC_DIR)/train_wiki \
--is_squad
squad-reader-dev-docs-wiki: data-config
python scripts/preprocess/create_nq_reader_doc_wiki.py \
--wiki_dir $(WIKI_DIR)/20181220_concat \
--nq_reader_docs_dir $(SQUAD_DIR)/dev-v1.1.json \
--output_dir $(SQUADREADER_DOC_DIR)/dev_wiki \
--is_squad
build-db: data-config
python scripts/preprocess/build_db.py \
--data_path $(WIKI_DIR)/extracted \
--save_path $(WIKI_DIR)/docs_20181220_nolist.db \
--preprocess scripts/preprocess/prep_wikipedia.py \
--overwrite
build-wikisquad: data-config
python scripts/preprocess/build_wikisquad.py \
--db_path $(WIKI_DIR)/docs_20181220_nolist.db \
--out_dir $(WIKI_DIR)/20181220_nolist
concat-wikisquad: data-config
python scripts/preprocess/concat_wikisquad.py \
--input_dir $(WIKI_DIR)/20181220_nolist \
--output_dir $(WIKI_DIR)/20181220_nolist_concat
first-para-wikisquad: data-config
python scripts/preprocess/first_para_wikisquad.py \
--input_dir $(WIKI_DIR)/20181220_nolist \
--output_dir $(WIKI_DIR)/20181220_nolist_first
compare-db: data-config
python scripts/preprocess/compare_db.py \
--db1 $(DATA_DIR)/denspi/docs.db \
--db2 $(WIKI_DIR)/docs_20181220.db
sample-nq-reader-doc-wiki-train: data-config
python scripts/preprocess/sample_nq_reader_doc_wiki.py \
--sampling_ratio 0.15 \
--wiki_dir $(WIKI_DIR)/20181220_concat \
--docs_wiki_dir $(NQREADER_DOC_DIR)/train_wiki \
--output_dir $(NQREADER_DOC_DIR)/train_wiki_noise
sample-nq-reader-doc-wiki-dev: data-config
python scripts/preprocess/sample_nq_reader_doc_wiki.py \
--sampling_ratio 0.1 \
--wiki_dir $(WIKI_DIR)/20181220_concat \
--docs_wiki_dir $(NQREADER_DOC_DIR)/dev_wiki \
--output_dir $(NQREADER_DOC_DIR)/dev_wiki_noise
trex-gold:
$(eval GOLD_FILE=$(DATA_DIR)/kilt/trex/trex-dev-kilt.jsonl)
zsre-gold:
$(eval GOLD_FILE=$(DATA_DIR)/kilt/zsre/structured_zeroshot-dev-kilt.jsonl)
strip-kilt: zsre-gold
python scripts/kilt/strip_pred.py \
$(INPUT_PRED) \
$(GOLD_FILE)
#The following instructions are for students in course of professor Luo Tiejian(UCAS).
#step1: to do the prediction of samples
step1:
python generate_phrase_vecs.py \
--model_type bert \
--pretrained_name_or_path SpanBERT/spanbert-base-cased \
--data_dir ./ \
--cache_dir $(CACHE_DIR) \
--predict_file data/wiki_physics.json \
--do_dump \
--max_seq_length 512 \
--doc_stride 500 \
--fp16 \
--filter_threshold -2.0 \
--append_title \
--load_dir $(SAVE_DIR)/densephrases-multi \
--output_dir $(SAVE_DIR)/densephrases-physics
python build_phrase_index.py \
$(SAVE_DIR)/densephrases-physics/dump all \
--replace \
--num_clusters 32 \
--fine_quant OPQ96 \
--doc_sample_ratio 1.0 \
--vec_sample_ratio 1.0 \
--cuda
python scripts/preprocess/compress_metadata.py \
--input_dump_dir $(SAVE_DIR)/densephrases-physics/dump/phrase \
--output_dir $(SAVE_DIR)/densephrases-physics/dump
#step1: to test the prediction of samples
step1_test:
python step1_test_with_question.py \
--dump_dir $(SAVE_DIR)/densephrases-physics/dump \
--index_dir start/32_flat_OPQ96 \
--query_encoder_path $(SAVE_DIR)/densephrases-multi