adapt view and configure env vars

bothub/api/v2/zeroshot/usecases/format_prompt.py

@@ -0,0 +1,57 @@
+
+class FormatPrompt:
+    const_prompt_data = {
+        "pt_br": {
+            "prompt_context": "Você é muito especialista em ",
+            "prompt_has_classes": ". Você possui as classes:\n\n",
+            "prompt_class_prefix": "Classe: ",
+            "prompt_context_classes": "Contexto da classe ",
+            "is_class_context_prefix": True,
+            "prompt_none_name": "Nenhuma",
+            "prompt_none_definition": "Classe: Nenhuma\nContexto da classe Nenhuma: A classe nenhuma é usada quando a frase não se relaciona com as outras classes definidas.\n\n",
+            "prompt_phrase_prefix": "Frase: ",
+            "prompt_analyse_text": "Pare, pense bem e analise qual é a melhor resposta de classe para a frase, responda só se você tiver muita certeza.\n\nClasse: "
+        }, 
+        "en": {
+            "prompt_context": "You are very expert in ",
+            "prompt_has_classes": ". You have the following classes:\n\n",
+            "prompt_class_suffix": "Class: ",
+            "is_class_context_prefix": False,
+            "prompt_context_classes": " class context",
+            "prompt_none_name": "None",
+            "prompt_none_definition": "Class: None\n\nNone class context: Use when a sentence doesn't align with any of the classes above.\n",
+            "prompt_phrase_prefix": "Sentence: ",
+            "prompt_analyse_text": "Stop, think carefully and analyze what the best class answer to the sentence is, only answer if you are very sure.\n\nClass:"
+        },
+        "es": {
+            "prompt_context": "Eres muy experto en ",
+            "prompt_has_classes": ". Usted posee las clases:\n\n",
+            "prompt_class_prefix": "Clase: ",
+            "is_class_context_prefix": True,
+            "prompt_context_classes": "Contexto de la clase ",
+            "prompt_none_name": "Ninguna",
+            "prompt_none_definition": "Clase: Ninguna\n\nContexto de la clase Ninguna: Aplicable si el tema no corresponde con las clases establecidas.",
+            "prompt_phrase_prefix": "Frase: ",
+            "prompt_analyse_text": "Detente, piensa detenidamente y analiza cuál es la mejor respuesta de clase a la frase, responde sólo si estás muy seguro.\n\nClase:"
+        }
+    }
+
+    def generate_prompt(self, language: str, zeroshot_data: dict):
+
+        translated_text = self.const_prompt_data[language]
+        prompt = translated_text.get("prompt_context") + zeroshot_data.get("context") + translated_text.get("prompt_has_classes")
+        for option in zeroshot_data.get("options"):
+            prompt += translated_text.get("prompt_class_prefix") + option.get("class", "").capitalize() + "\n"
+            if translated_text.get("is_class_context_prefix"):
+                prompt += translated_text.get("prompt_context_classes") + option.get("class", "").capitalize() + ": "
+            else:
+                prompt += option.get("class", "").capitalize() + translated_text.get("prompt_context_classes") + ": "
+            prompt += option.get("context") + "\n"
+        prompt += translated_text.get("prompt_none_definition") + translated_text.get("prompt_phrase_prefix")
+        prompt += zeroshot_data.get("text") + "\n" + translated_text.get("prompt_analyse_text")
+
+        return prompt
+
+    def get_none_class(self, language: str):
+        data = self.const_prompt_data.get(language)
+        return data.get("prompt_none_name", "Nenhuma")

bothub/api/v2/zeroshot/views.py

@@ -16,6 +16,8 @@
     ZeroshotLogs
 )
 
+from .usecases.format_prompt import FormatPrompt
+
 from bothub.api.v2.zeroshot.permissions import ZeroshotTokenPermission
 
 logger = logging.getLogger(__name__)
@@ -83,20 +85,24 @@ class ZeroShotFastPredictAPIView(APIView):
     def post(self, request):
 
         data = request.data
-
-        classes = {}
-
-        for categorie in data.get("categories"):
-            option = categorie.get("option")
-            classes[option] = [option]
-            for synonym in categorie.get("synonyms"):
-                classes[option].append(synonym)
+        formatter = FormatPrompt()
+        prompt = formatter.generate_prompt(data.get("language"), data)
 
         body = {
             "input": {
                 "text": data.get("text"),
                 "language": data.get("language"),
-                "classes": classes
+                "prompt": prompt,
+                "sampling_params": {
+                    "max_tokens": settings.ZEROSHOT_MAX_TOKENS,
+                    "n": settings.ZEROSHOT_N,
+                    "top_p": settings.ZEROSHOT_TOP_P,
+                    "tok_k": settings.ZEROSHOT_TOK_K,
+                    "temperature": settings.ZEROSHOT_TEMPERATURE,
+                    "do_sample": settings.ZEROSHOT_DO_SAMPLE,
+                    "stop": settings.ZEROSHOT_STOP
+                }
+
             }
         }
 
@@ -114,17 +120,32 @@ def post(self, request):
                 url=url,
                 json=body
             )
+
+            response = {}
+            other = False
+            classification = None
+
             if response_nlp.status_code == 200:
                 classification_data = response_nlp.json().get("output")
-                ZeroshotLogs.objects.create(
-                    text=data.get("text"),
-                    classification=classification_data.get("classification"),
-                    other=classification_data.get("other"),
-                    categories=classes,
-                    nlp_log=json.dumps(response_nlp.json()),
-                    language=data.get("language"),
-                )
-            return Response(status=response_nlp.status_code, data=response_nlp.json() if response_nlp.status_code == 200 else {"error": response_nlp.text})
+                classification = classification_data.get("text")[0].strip()
+                other = formatter.get_none_class(language=data.get("language")) in classification
+                response = {
+                    "output": {
+                        "classification": classification,
+                        "other": other
+                    }
+                }
+
+            ZeroshotLogs.objects.create(
+                text=data.get("text"),
+                classification=classification,
+                other=other,
+                options=data.get("options"),
+                nlp_log=str(response_nlp.json()),
+                language=data.get("language")
+            )
+
+            return Response(status=response_nlp.status_code, data=response if response_nlp.status_code == 200 else {"error": response_nlp.text})
         except Exception as error:
             logger.error(f"[ - ] Zeroshot fast predict: {error}")
             return Response(status=response_nlp.status_code if response_nlp else 500, data={"error": error})

bothub/settings.py

@@ -123,7 +123,14 @@
     ZEROSHOT_BASE_NLP_URL=(str, ""),
     FLOWS_TOKEN_ZEROSHOT=(str, ""),
     ZEROSHOT_SUFFIX=(str, ""),
-    ZEROSHOT_TOKEN=(str, "")
+    ZEROSHOT_TOKEN=(str, ""),
+    ZEROSHOT_MAX_TOKENS = (int, 20),
+    ZEROSHOT_N = (int, 1),
+    ZEROSHOT_TOP_P = (float, 0.95),
+    ZEROSHOT_TOK_K = (int, 10),
+    ZEROSHOT_TEMPERATURE = (float, 0.1),
+    ZEROSHOT_DO_SAMPLE = (bool, False),
+    ZEROSHOT_STOP = (str, "\n"),
 )
 
 # Build paths inside the project like this: os.path.join(BASE_DIR, ...)
@@ -710,3 +717,10 @@
 FLOWS_TOKEN_ZEROSHOT = env.str("FLOWS_TOKEN_ZEROSHOT")
 ZEROSHOT_SUFFIX = env.str("ZEROSHOT_SUFFIX")
 ZEROSHOT_TOKEN = env.str("ZEROSHOT_TOKEN")
+ZEROSHOT_MAX_TOKENS = env.int("ZEROSHOT_MAX_TOKENS")
+ZEROSHOT_N = env.int("ZEROSHOT_N")
+ZEROSHOT_TOP_P = env.float("ZEROSHOT_TOP_P")
+ZEROSHOT_TOK_K = env.int("ZEROSHOT_TOK_K")
+ZEROSHOT_TEMPERATURE = env.float("ZEROSHOT_TEMPERATURE")
+ZEROSHOT_DO_SAMPLE = env.bool("ZEROSHOT_DO_SAMPLE")
+ZEROSHOT_STOP = env.str("ZEROSHOT_STOP")