Merge pull request #46 from texttron/fast_tokenizer

to use Fast tokenizer

src/tevatron/arguments.py

@@ -99,12 +99,15 @@ def __post_init__(self):
             self.dataset_split = 'train'
             self.dataset_language = 'default'
         if self.train_dir is not None:
-            files = os.listdir(self.train_dir)
-            self.train_path = [
-                os.path.join(self.train_dir, f)
-                for f in files
-                if f.endswith('jsonl') or f.endswith('json')
-            ]
+            if os.path.isdir(self.train_dir):
+                files = os.listdir(self.train_dir)
+                self.train_path = [
+                    os.path.join(self.train_dir, f)
+                    for f in files
+                    if f.endswith('jsonl') or f.endswith('json')
+                ]
+            else:
+                self.train_path = [self.train_dir]
         else:
             self.train_path = None
 

src/tevatron/data.py

@@ -30,7 +30,7 @@ def __init__(
         self.total_len = len(self.train_data)
 
     def create_one_example(self, text_encoding: List[int], is_query=False):
-        item = self.tok.encode_plus(
+        item = self.tok.prepare_for_model(
             text_encoding,
             truncation='only_first',
             max_length=self.data_args.q_max_len if is_query else self.data_args.p_max_len,
@@ -95,7 +95,7 @@ def __len__(self):
 
     def __getitem__(self, item) -> Tuple[str, BatchEncoding]:
         text_id, text = (self.encode_data[item][f] for f in self.input_keys)
-        encoded_text = self.tok.encode_plus(
+        encoded_text = self.tok.prepare_for_model(
             text,
             max_length=self.max_len,
             truncation='only_first',

src/tevatron/datasets/preprocessor.py

@@ -54,4 +54,4 @@ def __call__(self, example):
                                      add_special_tokens=False,
                                      max_length=self.text_max_length,
                                      truncation=True)
-        return {'text_id': docid, 'text': text}
+        return {'text_id': docid, 'text': text}

src/tevatron/driver/encode.py

@@ -52,8 +52,7 @@ def main():
     )
     tokenizer = AutoTokenizer.from_pretrained(
         model_args.tokenizer_name if model_args.tokenizer_name else model_args.model_name_or_path,
-        cache_dir=model_args.cache_dir,
-        use_fast=False,
+        cache_dir=model_args.cache_dir
     )
 
     model = DenseModel.load(

src/tevatron/driver/train.py

@@ -2,6 +2,7 @@
 import os
 import sys
 
+import torch
 from transformers import AutoConfig, AutoTokenizer
 from transformers import (
     HfArgumentParser,
@@ -66,8 +67,7 @@ def main():
     )
     tokenizer = AutoTokenizer.from_pretrained(
         model_args.tokenizer_name if model_args.tokenizer_name else model_args.model_name_or_path,
-        cache_dir=model_args.cache_dir,
-        use_fast=False,
+        cache_dir=model_args.cache_dir
     )
     model = DenseModel.build(
         model_args,
@@ -78,7 +78,13 @@ def main():
 
     train_dataset = HFTrainDataset(tokenizer=tokenizer, data_args=data_args,
                                    cache_dir=data_args.data_cache_dir or model_args.cache_dir)
+    if training_args.local_rank > 0:
+        print("Waiting for main process to perform the mapping")
+        torch.distributed.barrier()
     train_dataset = TrainDataset(data_args, train_dataset.process(), tokenizer)
+    if training_args.local_rank == 0:
+        print("Loading results from main process")
+        torch.distributed.barrier()
 
     trainer_cls = GCTrainer if training_args.grad_cache else Trainer
     trainer = trainer_cls(

src/tevatron/faiss_retriever/__main__.py

@@ -19,7 +19,7 @@
 
 def search_queries(retriever, q_reps, p_lookup, args):
     if args.batch_size > 0:
-        all_scores, all_indices = retriever.batch_search(q_reps, args.depth, args.batch_size)
+        all_scores, all_indices = retriever.batch_search(q_reps, args.depth, args.batch_size, args.quiet)
     else:
         all_scores, all_indices = retriever.search(q_reps, args.depth)
 
@@ -56,6 +56,7 @@ def main():
     parser.add_argument('--depth', type=int, default=1000)
     parser.add_argument('--save_ranking_to', required=True)
     parser.add_argument('--save_text', action='store_true')
+    parser.add_argument('--quiet', action='store_true')
 
     args = parser.parse_args()
 

src/tevatron/faiss_retriever/retriever.py

@@ -2,6 +2,7 @@
 import faiss
 
 import logging
+from tqdm import tqdm
 
 logger = logging.getLogger(__name__)
 
@@ -17,11 +18,11 @@ def add(self, p_reps: np.ndarray):
     def search(self, q_reps: np.ndarray, k: int):
         return self.index.search(q_reps, k)
 
-    def batch_search(self, q_reps: np.ndarray, k: int, batch_size: int):
+    def batch_search(self, q_reps: np.ndarray, k: int, batch_size: int, quiet: bool=False):
         num_query = q_reps.shape[0]
         all_scores = []
         all_indices = []
-        for start_idx in range(0, num_query, batch_size):
+        for start_idx in tqdm(range(0, num_query, batch_size), disable=quiet):
             nn_scores, nn_indices = self.search(q_reps[start_idx: start_idx + batch_size], k)
             all_scores.append(nn_scores)
             all_indices.append(nn_indices)