Add Text Generation Inference with JSON output (#235)

keybert/llm/__init__.py

@@ -1,6 +1,12 @@
 from keybert._utils import NotInstalled
 from keybert.llm._base import BaseLLM
 
+# TextGenerationInference
+try:
+    from keybert.llm._textgenerationinference import TextGenerationInference
+except ModuleNotFoundError:
+    msg = "`pip install huggingface-hub pydantic ` \n\n"
+    TextGenerationInference = NotInstalled("TextGenerationInference", "huggingface-hub", custom_msg=msg)
 
 # TextGeneration
 try:
@@ -43,6 +49,7 @@
     "Cohere",
     "OpenAI",
     "TextGeneration",
+    "TextGenerationInference",
     "LangChain",
     "LiteLLM"
 ]

keybert/llm/_textgenerationinference.py

@@ -0,0 +1,125 @@
+from tqdm import tqdm
+from pydantic import BaseModel
+from huggingface_hub import InferenceClient
+from typing import Mapping, List, Any
+from keybert.llm._base import BaseLLM
+from keybert.llm._utils import process_candidate_keywords
+import json
+
+DEFAULT_PROMPT = """
+    I have the following document:
+    [DOCUMENT]
+
+    With the following candidate keywords:
+    [CANDIDATES]
+
+    Based on the information above, improve the candidate keywords to best describe the topic of the document.
+
+    Output in JSON format:
+"""
+
+
+class Keywords(BaseModel):
+    keywords: List[str]
+
+
+class TextGenerationInference(BaseLLM):
+    """ Tex
+
+    Arguments:
+        client: InferenceClient from huggingface_hub.
+        prompt: The prompt to be used in the model. If no prompt is given,
+                `self.default_prompt_` is used instead.
+                NOTE: Use `"[KEYWORDS]"` and `"[DOCUMENTS]"` in the prompt
+                to decide where the keywords and documents need to be
+                inserted.
+        client_kwargs: Kwargs that you can pass to the client.text_generation
+                         when it is called.
+        json_schema: Pydantic BaseModel to be used as guidance for keywords.
+                By default uses:
+                class Keywords(BaseModel):
+                    keywords: List[str]
+
+    Usage:
+
+    ```python
+    from pydantic import BaseModel
+    from huggingface_hub import InferenceClient
+    from keybert.llm import TextGenerationInference
+    from keybert import KeyLLM
+
+    # Json Schema
+    class Keywords(BaseModel):
+        keywords: List[str]
+
+    # Create your LLM
+    generator = InferenceClient('url')
+    llm = TextGenerationInference(generator, Keywords)
+
+    # Load it in KeyLLM
+    kw_model = KeyLLM(llm)
+
+    # Extract keywords
+    document = "The website mentions that it only takes a couple of days to deliver but I still have not received mine."
+    keywords = kw_model.extract_keywords(document)
+    ```
+
+    You can use a custom prompt and decide where the document should
+    be inserted with the `[DOCUMENT]` tag:
+
+    ```python
+    from keybert.llm import TextGenerationInference
+
+    prompt = "I have the following documents '[DOCUMENT]'. Please give me the keywords that are present in this document and separate them with commas:"
+
+    # Create your representation model
+    from huggingface_hub import InferenceClient
+    generator = InferenceClient('url')
+    llm = TextGenerationInference(generator)
+    ```
+    """
+
+    def __init__(self,
+                 client: InferenceClient,
+                 prompt: str = None,
+                 json_schema: BaseModel = Keywords
+                 ):
+        self.client = client
+        self.prompt = prompt if prompt is not None else DEFAULT_PROMPT
+        self.default_prompt_ = DEFAULT_PROMPT
+        self.json_schema = json_schema
+
+    def extract_keywords(
+            self,
+            documents: List[str], candidate_keywords: List[List[str]] = None,
+            inference_kwargs: Mapping[str, Any] = {}
+    ):
+        """ Extract topics
+
+        Arguments:
+            documents: The documents to extract keywords from
+            candidate_keywords: A list of candidate keywords that the LLM will fine-tune
+                        For example, it will create a nicer representation of
+                        the candidate keywords, remove redundant keywords, or
+                        shorten them depending on the input prompt.
+
+        Returns:
+            all_keywords: All keywords for each document
+        """
+        all_keywords = []
+        candidate_keywords = process_candidate_keywords(documents, candidate_keywords)
+
+        for document, candidates in tqdm(zip(documents, candidate_keywords), disable=not self.verbose):
+            prompt = self.prompt.replace("[DOCUMENT]", document)
+            if candidates is not None:
+                prompt = prompt.replace("[CANDIDATES]", ", ".join(candidates))
+
+            # Extract result from generator and use that as label
+            response = self.client.text_generation(
+                prompt=prompt,
+                grammar={"type": "json", "value": self.json_schema.schema()},
+                **inference_kwargs
+            )
+            all_keywords = json.loads(response)["keywords"]
+
+        return all_keywords

pyproject.toml

@@ -40,7 +40,7 @@ dependencies = [
     "numpy>=1.18.5",
     "rich>=10.4.0",
     "scikit-learn>=0.22.2",
-    "sentence-transformers>=0.3.8",
+    "sentence-transformers>=0.3.8"
 ]
 
 [project.optional-dependencies]
@@ -70,6 +70,10 @@ test = [
     "pytest-cov>=2.6.1",
     "pytest>=5.4.3",
 ]
+tgi = [
+    "huggingface-hub>=0.23.3",
+    "pydantic>=2.7.4"
+]
 use = [
     "tensorflow",
     "tensorflow_hub",