feat: add proofread provider

Signed-off-by: Jana Peper <[email protected]>

lib/proofread.py

@@ -0,0 +1,40 @@
+# SPDX-FileCopyrightText: 2024 Nextcloud GmbH and Nextcloud contributors
+# SPDX-License-Identifier: AGPL-3.0-or-later
+"""A chain to proofread a text
+"""
+
+from typing import Any
+from langchain.prompts import PromptTemplate
+from langchain.schema.prompt_template import BasePromptTemplate
+from langchain_core.runnables import Runnable
+
+
+class ProofreadProcessor:
+    """
+    A proofreading chain
+    """
+    system_prompt: str = "You're an AI assistant tasked with proofreading the text given to you by the user."
+    user_prompt: BasePromptTemplate = PromptTemplate(
+        input_variables=["text"],
+        template="""
+Detect all grammar and spelling mistakes of the following text in its original language. Output only the list of mistakes in bullet points.
+
+"
+{text}
+"
+
+Give me the list of all mistakes in the above text in its original language. Do not output the language. Output only the list in bullet points, nothing else, no introductory or explanatory text.
+        """
+    )
+
+    runnable: Runnable
+
+    def __init__(self, runnable: Runnable):
+        self.runnable = runnable
+
+    def __call__(
+            self,
+            inputs: dict[str, Any],
+    ) -> dict[str, Any]:
+        output = self.runnable.invoke({"user_prompt": self.user_prompt.format_prompt(text=inputs['input']), "system_prompt": self.system_prompt})
+        return {'output': output}

lib/task_processors.py

@@ -16,6 +16,7 @@
 from chat import ChatProcessor
 from free_prompt import FreePromptProcessor
 from headline import HeadlineProcessor
+from proofread import ProofreadProcessor
 from chatwithtools import ChatWithToolsProcessor
 from topics import TopicsProcessor
 from summarize import SummarizeProcessor
@@ -132,5 +133,7 @@ def generate_task_processors_for_model(file_name, task_processors):
     # chains[model_name + ":core:text2text:reformulation"] = lambda: ReformulateChain(llm_chain=llm_chain(), chunk_size=chunk_size)
     task_processors[model_name + ":core:text2text"] = lambda: FreePromptProcessor(generate_llm_chain(file_name))
     task_processors[model_name + ":core:text2text:chat"] = lambda: ChatProcessor(generate_chat_chain(file_name))
+    task_processors[model_name + ":core:text2text:proofread"] = lambda: ProofreadProcessor(generate_llm_chain(file_name))
     task_processors[model_name + ":core:text2text:chatwithtools"] = lambda: ChatWithToolsProcessor(generate_chat_chain(file_name))
+
     # chains[model_name + ":core:contextwrite"] = lambda: ContextWriteChain(llm_chain=llm_chain())