added fact checking

Author: drorIvry

README.md

@@ -75,3 +75,14 @@ consisTent.LabelsValidator(openai_key=OPENAI_KEY).validate(
     model_output="What do you call a rabbit that tells jokes? A funny bunny!",
 )
 ```
+
+### facts validation
+
+```python
+OPENAI_KEY = "XXXXXXXXXXXXXXX"
+
+consisTent.FactsValidator(openai_key=OPENAI_KEY).validate(
+    facts=["this car weighs 1000KG"],
+    model_output="I can lift this car",
+)
+```

consisTent/validators/semantic_validators/facts_validator.py

@@ -0,0 +1,54 @@
+from typing import List
+from langchain import LLMChain, PromptTemplate
+from langchain.llms import OpenAI
+
+from ..base_validator import Validator
+
+
+class FactsValidator(Validator):
+    def __init__(
+        self,
+        openai_key: str,
+    ):
+        self._model = OpenAI(
+            temperature=0,
+            openai_api_key=openai_key,
+            model_name="text-davinci-003",
+        )
+
+        self._template = """
+            In the next answer only address the data that was given to answer yes/no.
+            Given the following facts:
+            {facts}
+            assert if the following is factually true:
+            {response}
+            respond with yes/no
+
+            YOUR RESPONSE:
+            """
+
+        self._prompt = PromptTemplate(
+            template=self._template, input_variables=["facts", "response"]
+        )
+
+    def validate(
+        self,
+        facts: List[str],
+        model_output: str,
+    ):
+        parsed_facts = ", ".join(facts)
+
+        fact_check_chain = LLMChain(
+            prompt=self._prompt,
+            llm=self._model,
+        )
+        entails = fact_check_chain.predict(
+            facts=parsed_facts,
+            response=model_output,
+        )
+
+        entails = entails.lower().strip()
+
+        assert (
+            "yes" in entails
+        ), "llm validation check validation failed on fact check"  # noqa: E501