Merge pull request #2 from Minitour/feature/support-generator

Added generator support

evaluations/raw.py

@@ -1,6 +1,6 @@
 import os
 
-from verbalizer.nlp import LlamaModelParaphrase, ChatGptModelParaphrase
+from verbalizer.nlp import ChatGptModelParaphrase
 
 examples = [
     """
@@ -556,9 +556,9 @@
 ]
 
 if __name__ == '__main__':
-    llama_model = LlamaModelParaphrase('http://localhost:11434/v1', temperature=0.1)
+    # llama_model = LlamaModelParaphrase('http://localhost:11434/v1', temperature=0.1)
     openai_model = ChatGptModelParaphrase(api_key=os.getenv('OPENAI_API_KEY'), model='gpt-4o', temperature=0.7)
-    models = [openai_model, llama_model]
+    models = [openai_model]
 
     for model in models:
         print(f'Running on {model.name}:')

playground.py

@@ -1,7 +1,7 @@
 import logging
 import os
 
-from verbalizer.nlp import ChatGptModelParaphrase, LlamaModelParaphrase
+from verbalizer.nlp import ChatGptModelParaphrase
 from verbalizer.process import Processor
 from verbalizer.sampler import Sampler
 from verbalizer.verbalizer import Verbalizer
@@ -83,9 +83,8 @@
 }
 
 if __name__ == '__main__':
-    llama_model = LlamaModelParaphrase('http://localhost:11434/v1', temperature=0.1)
     openai_model = ChatGptModelParaphrase(api_key=os.getenv('OPENAI_API_KEY'), model='gpt-4o', temperature=0.7)
-    models = [openai_model, llama_model]
+    models = [openai_model]
 
     sampler = Sampler(sample_n=100, seed=42)
 

pyproject.toml

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "ontology-verbalizer"
-version = "1.1.0"
+version = "1.1.1"
 description = "A Python package for ontology verbalization"
 authors = ["Antonio Zaitoun <tony.z.1711@gmail.com>"]
 license = "MIT"
@@ -12,7 +12,6 @@ repository = "https://github.com/Minitour/ontology-verbalizer"
 [tool.poetry.dependencies]
 python = "^3.12"
 rdflib = "~7.0.0"
-openai = "~1.12.0"
 pandas = "~2.2.0"
 tqdm = "~4.66.2"
 

tests/test_verbalization.py

@@ -1,3 +1,4 @@
+import types
 import unittest
 
 from rdflib import Graph
@@ -64,3 +65,16 @@ def test_verbalization_with_sampler(self):
 
         # although we sampled 10, only 7 were applicable.
         self.assertEqual(7, len(results))
+
+    def test_verbalization_with_generator(self):
+        ontology = Processor.from_file('./data/foaf.owl')
+
+        # create vocabulary
+        vocab = Vocabulary(ontology, ignore=ignore_iri, rephrased=rename_iri)
+
+        # create verbalizer
+        verbalizer = Verbalizer(vocab)
+
+        results = Processor.verbalize_with(verbalizer, namespace='foaf', as_generator=True)
+        self.assertTrue(isinstance(results, types.GeneratorType))
+        self.assertEqual(12, len(list(results)))

verbalizer/nlp.py

@@ -2,10 +2,40 @@
 from abc import ABC, abstractmethod
 from typing import Optional
 
-from openai import OpenAI
+try:
+    from openai import OpenAI
 
-logging.getLogger("openai").setLevel(logging.ERROR)
-logging.getLogger("httpx").setLevel(logging.ERROR)
+    logging.getLogger("openai").setLevel(logging.ERROR)
+    logging.getLogger("httpx").setLevel(logging.ERROR)
+except ModuleNotFoundError as err:
+    OpenAI = None
+
+
+class ParaphraseLanguageModel(ABC):
+
+    @abstractmethod
+    def pseudo_to_text(self, pseudo_text: str, extra: str = None) -> str:
+        """
+        Given a pseudo text or controlled natural language, return a rephrased version of that same text.
+        :param pseudo_text: The CNL set of statements,
+        :param extra: Additional context to include as part of the prompt.
+        :return: Paraphrased text.
+        """
+        return pseudo_text
+
+    @property
+    def cost(self) -> float:
+        """
+        The usage cost so far of the model.
+        """
+        return 0.0
+
+    @property
+    def name(self) -> str:
+        """
+        The name of the model used.
+        """
+        return 'Unknown'
 
 
 def get_messages(pseudo_text: str, extra_context: Optional[str] = None):
@@ -50,33 +80,6 @@ def get_messages(pseudo_text: str, extra_context: Optional[str] = None):
     ]
 
 
-class ParaphraseLanguageModel(ABC):
-
-    @abstractmethod
-    def pseudo_to_text(self, pseudo_text: str, extra: str = None) -> str:
-        """
-        Given a pseudo text or controlled natural language, return a rephrased version of that same text.
-        :param pseudo_text: The CNL set of statements,
-        :param extra: Additional context to include as part of the prompt.
-        :return: Paraphrased text.
-        """
-        return pseudo_text
-
-    @property
-    def cost(self) -> float:
-        """
-        The usage cost so far of the model.
-        """
-        return 0.0
-
-    @property
-    def name(self) -> str:
-        """
-        The name of the model used.
-        """
-        return 'Unknown'
-
-
 class ChatGptModelParaphrase(ParaphraseLanguageModel):
     """
     OpenAI wrapper implementation.
@@ -138,6 +141,9 @@ class ChatGptModelParaphrase(ParaphraseLanguageModel):
     }
 
     def __init__(self, api_key: str, model: str = 'gpt-3.5-turbo-0613', temperature=0.5):
+        if not OpenAI:
+            raise ModuleNotFoundError("OpenAI is not installed. Please install it with `pip install openai`")
+
         self.model = model
         self.temperature = temperature
         self.client = OpenAI(api_key=api_key)
@@ -156,7 +162,7 @@ def pseudo_to_text(self, pseudo_text: str, extra: str = None) -> str:
 
     @property
     def cost(self) -> float:
-        model_pricing = self.models.get(self.model)
+        model_pricing = self.models.get(self.model) or {'input': 0.0, 'output': 0.0}
 
         in_tokens = self._in_token_usage / 1000
         out_tokens = self._out_token_usage / 1000
@@ -166,33 +172,3 @@ def cost(self) -> float:
     @property
     def name(self) -> str:
         return self.model
-
-
-class LlamaModelParaphrase(ParaphraseLanguageModel):
-    """
-    Llama model wrapper implementation.
-    """
-
-    def __init__(self, base_url, model='llama3', temperature=0.5):
-        self.temperature = temperature
-        self.model = model
-        self.client = OpenAI(
-            base_url=base_url,
-            api_key="sk-no-key-required"
-        )
-
-    def pseudo_to_text(self, pseudo_text: str, extra: str = None) -> str:
-        response = self.client.chat.completions.create(
-            model=self.model,
-            messages=get_messages(pseudo_text, extra),
-            temperature=self.temperature
-        )
-        return response.choices[0].message.content.strip()
-
-    @property
-    def cost(self) -> float:
-        return 0.0
-
-    @property
-    def name(self) -> str:
-        return self.model

verbalizer/process.py

@@ -27,14 +27,39 @@ def verbalize_with(cls,
                        namespace: str,
                        output_dir: Optional[str] = None,
                        chunk_size: int = 1000,
-                       sampler: Optional[Sampler] = None):
+                       sampler: Optional[Sampler] = None,
+                       as_generator: bool = False):
+        gen = cls.verbalize_with_stream(
+            verbalizer,
+            namespace=namespace,
+            output_dir=output_dir,
+            chunk_size=chunk_size,
+            sampler=sampler,
+            as_generator=as_generator
+        )
+        if as_generator:
+            return gen
+
+        return next(gen)
+
+    @classmethod
+    def verbalize_with_stream(
+            cls,
+            verbalizer: Verbalizer,
+            *,
+            namespace: str,
+            output_dir: Optional[str] = None,
+            chunk_size: int = 1000,
+            sampler: Optional[Sampler] = None,
+            as_generator: bool = False):
         """
         Start the verbalization process.
         :param verbalizer: The verbalizer to use.
         :param namespace: Name of the directory to create under the output directory.
         :param output_dir: Name of the output directory.
         :param chunk_size: Number of entries (rows) per file. default = 1000
         :param sampler: A sampling configuration, use to sample large ontologies.
+        :param as_generator: If True, returns a generator instead of a list.
         """
 
         # current timestamp
@@ -67,7 +92,7 @@ def verbalize_with(cls,
             if stats.statements == 0:
                 continue
 
-            chunk_dataset.append({
+            element = {
                 'ontology': namespace,
                 'root': entry,
                 'fragment': fragment,
@@ -79,7 +104,12 @@ def verbalize_with(cls,
                 'unique_relationships': len(stats.relationship_counter),
                 'total_relationships': sum(stats.relationship_counter.values()),
                 **stats.relationship_counter
-            })
+            }
+
+            chunk_dataset.append(element)
+
+            if as_generator:
+                yield element
 
             if len(chunk_dataset) != chunk_size:
                 continue
@@ -104,7 +134,8 @@ def verbalize_with(cls,
         if llm:
             logger.info(f'LLM usage cost: ${llm.cost}')
 
-        return full_dataset
+        if not as_generator:
+            yield full_dataset
 
     @staticmethod
     def _get_classes(graph):