Restrict number of items to search

.gitignore

@@ -0,0 +1,2 @@
+*.parquet
+.DS_Store

app.py

@@ -12,7 +12,7 @@ import numpy as np
 import scipy.spatial
 import pyminizip
 import transformers
-from transformers import AutoModel, AutoTokenizer
+from transformers import BertJapaneseTokenizer, BertModel
 from huggingface_hub import hf_hub_download, snapshot_download
 # from PIL import Image
 
@@ -94,7 +94,7 @@ class ClipTextModel(nn.Module):
             filename = hf_hub_download(repo_id=model_name_or_path, filename="output_linear.bin")
             output_linear_state_dict = torch.load(filename)
 
-        self.model = AutoModel.from_pretrained(model_name_or_path)
+        self.model = BertModel.from_pretrained(model_name_or_path)
         config = self.model.config
 
         self.max_cls_depth = 6
@@ -105,8 +105,7 @@ class ClipTextModel(nn.Module):
         # self.output_linear = nn.Linear(3 * config.hidden_size, sentence_vector_size)
         self.output_linear.load_state_dict(output_linear_state_dict)
 
-        self.tokenizer = AutoTokenizer.from_pretrained(model_name_or_path, 
-                                                       is_fast=True, do_lower_case=True)
+        self.tokenizer = BertJapaneseTokenizer.from_pretrained(model_name_or_path, do_lower_case=True)
 
         self.eval()
 

clip_zeroshot_irasuto_image_items_20210224.pq.zip

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4bb0832140a68bc4e47ef63f9600852f978b2d52fc8f2091eaf021b1291717e3
-size 72074172
+oid sha256:2f602399369a485f1586b7ca04e8ae096868ecce85527928671b08bf5e80c200
+size 54262882

clip_zeroshot_irasuto_items_20210224.pq.zip

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:61ea93dcba23ee05b0b97e880aab395f7e82375d11c82b0a743940897fa1a2e4
-size 122826331
+oid sha256:321d909ba0f92425a5107ad26a6d97dc4f7601b2b4f22ab020199f2ba2237ce7
+size 104296063