wip

convlab/nlu/jointBERT/multiwoz/nlu.py

@@ -46,7 +46,8 @@ class BERTNLU(NLU):
 
         if not os.path.exists(output_dir):
             model_downloader(root_dir, model_file)
-        model = JointBERT(config['model'], DEVICE, dataloader.tag_dim, dataloader.intent_dim)
+        model = JointBERT(config['model'], DEVICE,
+                          dataloader.tag_dim, dataloader.intent_dim)
 
         state_dict = torch.load(os.path.join(
             output_dir, 'pytorch_model.bin'), DEVICE)
@@ -97,7 +98,8 @@ class BERTNLU(NLU):
         intents = []
         da = {}
 
-        word_seq, tag_seq, new2ori = self.dataloader.bert_tokenize(ori_word_seq, ori_tag_seq)
+        word_seq, tag_seq, new2ori = self.dataloader.bert_tokenize(
+            ori_word_seq, ori_tag_seq)
         word_seq = word_seq[:510]
         tag_seq = tag_seq[:510]
         batch_data = [[ori_word_seq, ori_tag_seq, intents, da, context_seq,

convlab/policy/ppo/emoTUS-BertNLU-RuleDST-PPOPolicy.json

@@ -23,9 +23,10 @@
     },
     "nlu_sys": {
         "BertNLU": {
-            "class_path": "convlab.nlu.jointBERT.multiwoz.BERTNLU",
+            "class_path": "convlab.nlu.jointBERT.unified_datasets.BERTNLU",
             "ini_params": {
-                "config_file": "multiwoz_all.json",
+                "mode": "all",
+                "config_file": "multiwoz21_sys_context3.json",
                 "model_file": "https://huggingface.co/ConvLab/ConvLab-2_models/resolve/main/bert_multiwoz_all.zip"
             }
         }