Merge pull request #96 from ConvLab/readme

fix bug in t5dst, t5nlg, and scgpt interface

convlab/base_models/t5/dst/dst.py

 import logging
-import os
 import torch
+from copy import deepcopy
 from transformers import AutoTokenizer, AutoModelForSeq2SeqLM, AutoConfig
 from convlab.dst.dst import DST
 from convlab.base_models.t5.dst.serialization import deserialize_dialogue_state
-from convlab.util.custom_util import model_downloader
+from convlab.util import load_ontology
 
 
 class T5DST(DST):
-    def __init__(self, speaker, context_window_size, model_name_or_path, device='cuda'):
+    def __init__(self, dataset_name, speaker, context_window_size, model_name_or_path, device='cuda'):
         assert speaker in ['user', 'system']
         assert context_window_size > 0
+        self.ontology = load_ontology(dataset_name)
         self.speaker = speaker
         self.opponent = 'system' if speaker == 'user' else 'user'
         self.context_window_size = context_window_size
@@ -24,7 +25,12 @@ class T5DST(DST):
         
         logging.info("T5DST loaded")
 
-    def update(self, context):
+    def update(self, user_action=None):
+        if self.state['history'][0][1] == 'null':
+            # skip first dummy turn
+            context = self.state['history'][1:]
+        else:
+            context = self.state['history']
         if len(context) > 0 and type(context[0]) is list and len(context[0]) > 1:
             context = [item[1] for item in context]
         context = context[-self.context_window_size:]
@@ -37,7 +43,17 @@ class T5DST(DST):
         output_seq = self.tokenizer.decode(output_seq[0], skip_special_tokens=True)
         # print(output_seq)
         state = deserialize_dialogue_state(output_seq.strip())
-        return state
+        self.state['belief_state'] = state
+        return self.state
+    
+    def init_session(self):
+        self.state = dict()
+        self.state['belief_state'] = deepcopy(self.ontology['state'])
+        self.state['booked'] = dict()
+        self.state['history'] = []
+        self.state['system_action'] = []
+        self.state['user_action'] = []
+        self.state['terminated'] = False
 
 
 if __name__ == '__main__':
@@ -59,7 +75,9 @@ if __name__ == '__main__':
         "You are welcome.  Is there anything else I can help you with today?",
         "No, I am all set.  Have a nice day.  Bye."],
     ]
-    dst = T5DST(speaker='user', context_window_size=100, model_name_or_path='output/dst/multiwoz21/user/context_100')
+    dst = T5DST('multiwoz21', speaker='user', context_window_size=100, model_name_or_path='ConvLab/t5-small-dst-multiwoz21')
+    dst.init_session()
     for context in contexts:
-        print(dst.update(context))
+        dst.state['history'] = context
+        print(dst.update())
         print()

convlab/base_models/t5/nlg/nlg.py

 import logging
-import os
 import torch
 from transformers import AutoTokenizer, AutoModelForSeq2SeqLM, AutoConfig
 from convlab.nlg.nlg import NLG
 from convlab.base_models.t5.nlu.serialization import serialize_dialogue_acts
-from convlab.util.custom_util import model_downloader
 
 
 class T5NLG(NLG):
@@ -33,7 +31,18 @@ class T5NLG(NLG):
         else:
             utts = ['']
         input_seq = '\n'.join([f"{self.opponent if (i % 2) == (len(utts) % 2) else self.speaker}: {utt}" for i, utt in enumerate(utts)])
+        if isinstance(dialogue_acts, dict):
+            # da in unified format
             dialogue_acts_seq = serialize_dialogue_acts(dialogue_acts)
+        elif isinstance(dialogue_acts[0], dict):
+            # da without da type
+            dialogue_acts_seq = serialize_dialogue_acts({'categorical': dialogue_acts})
+        elif isinstance(dialogue_acts[0], list):
+            # da is a list of list (convlab-2 format)
+            dialogue_acts_seq = serialize_dialogue_acts(
+                    {'categorical': [{'intent': da[0], 'domain': da[1], 'slot': da[2], 'value': da[3]} for da in dialogue_acts]})
+        else:
+            raise ValueError(f"invalid dialog acts format {dialogue_acts}")
         input_seq = dialogue_acts_seq + '\n' + input_seq
         # print(input_seq)
         input_seq = self.tokenizer(input_seq, return_tensors="pt").to(self.device)
@@ -47,7 +56,7 @@ class T5NLG(NLG):
 
 if __name__ == '__main__':
     das = [
-        {
+        { # da in unified format
         "categorical": [],
         "non-categorical": [],
         "binary": [
@@ -63,9 +72,7 @@ if __name__ == '__main__':
             }
         ]
         },
-        {
-        "categorical": [],
-        "non-categorical": [
+        [ # da without da type
             {
             "intent": "inform",
             "domain": "taxi",
@@ -83,25 +90,9 @@ if __name__ == '__main__':
             "end": 78
             }
         ],
-        "binary": [
-            {
-            "intent": "book",
-            "domain": "taxi",
-            "slot": ""
-            }
-        ]
-        },
-        {
-        "categorical": [],
-        "non-categorical": [],
-        "binary": [
-            {
-            "intent": "reqmore",
-            "domain": "general",
-            "slot": ""
-            }
-        ]
-        },
+        [ # da is a list of list (convlab-2 format)
+            ["reqmore", "general", "", ""]
+        ],
         {
         "categorical": [],
         "non-categorical": [],
@@ -132,7 +123,7 @@ if __name__ == '__main__':
         "You are welcome.  Is there anything else I can help you with today?"
         "No, I am all set.  Have a nice day.  Bye."],
     ]
-    nlg = T5NLG(speaker='system', context_window_size=0, model_name_or_path='output/nlg/multiwoz21/system/context_3')
+    nlg = T5NLG(speaker='system', context_window_size=0, model_name_or_path='ConvLab/t5-small-nlg-multiwoz21')
     for da, context in zip(das, contexts):
         print(da)
         print(nlg.generate(da, context))

convlab/base_models/t5/nlu/nlu.py

 import logging
-import os
 import torch
 from transformers import AutoTokenizer, AutoModelForSeq2SeqLM, AutoConfig
 from convlab.nlu.nlu import NLU
 from convlab.base_models.t5.nlu.serialization import deserialize_dialogue_acts
-from convlab.util.custom_util import model_downloader
 
 
 class T5NLU(NLU):

convlab/nlg/scgpt/scgpt.py

@@ -19,6 +19,17 @@ class SCGPT(NLG):
         self.model.load_state_dict(torch.load(model_path))
 
     def generate(self, action):
+        if isinstance(action, dict):
+            # da in unified format
+            pass
+        elif isinstance(action[0], dict):
+            # da without da type
+            action = {'categorical': action}
+        elif isinstance(action[0], list):
+            # da is a list of list (convlab-2 format)
+            action = {'categorical': [{'intent': da[0], 'domain': da[1], 'slot': da[2], 'value': da[3]} for da in action]}
+        else:
+            raise ValueError(f"invalid dialog acts format {action}")
         action_str = act2str(action)
         output = self._inference_batch([action_str])[0]
         return output