parallex 0.1.3__tar.gz → 0.1.4__tar.gz

{parallex-0.1.3 → parallex-0.1.4}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: parallex
-Version: 0.1.3
+Version: 0.1.4
 Summary: PDF to markdown using Azure OpenAI batch processing
 Home-page: https://github.com/Summed-AI/parallex
 Author: Jeff Hostetler

{parallex-0.1.3 → parallex-0.1.4}/parallex/ai/batch_processor.py

@@ -35,8 +35,6 @@ async def wait_for_batch_completion(client: OpenAIClient, batch: UploadBatch) ->
         await asyncio.sleep(delay)
         batch_response = await client.retrieve_batch(batch.id)
         status = batch_response.status
-        batch.output_file_id = batch_response.output_file_id
-        batch.error_file_id = batch_response.error_file_id
         delay = 30
         if status == "completed":
             return batch_response.output_file_id

{parallex-0.1.3 → parallex-0.1.4}/parallex/ai/open_ai_client.py

@@ -4,13 +4,15 @@ from openai import AsyncAzureOpenAI
 from openai._legacy_response import HttpxBinaryResponseContent
 from openai.types import FileObject, Batch, FileDeleted
 
+from parallex.file_management.remote_file_handler import RemoteFileHandler
 from parallex.utils.logger import logger
 
 
 # Exceptions for missing keys, etc
 class OpenAIClient:
-    def __init__(self, model: str):
+    def __init__(self, model: str, remote_file_handler: RemoteFileHandler):
         self.model = model
+        self.file_handler = remote_file_handler
 
         self._client = AsyncAzureOpenAI(
             azure_endpoint=os.getenv("AZURE_OPENAI_ENDPOINT"),
@@ -19,19 +21,29 @@ class OpenAIClient:
         )
 
     async def upload(self, file_path: str) -> FileObject:
-        return await self._client.files.create(
+        file = await self._client.files.create(
             file=open(file_path, "rb"), purpose="batch"
         )
+        self.file_handler.add_file(file.id)
+        return file
 
     async def create_batch(self, upload_file_id: str) -> Batch:
-        return await self._client.batches.create(
+        batch = await self._client.batches.create(
             input_file_id=upload_file_id,
             endpoint="/chat/completions",
             completion_window="24h",
         )
+        self.file_handler.add_file(batch.input_file_id)
+        self.file_handler.add_file(batch.output_file_id)
+        self.file_handler.add_file(batch.error_file_id)
+        return batch
 
     async def retrieve_batch(self, batch_id: str) -> Batch:
-        return await self._client.batches.retrieve(batch_id)
+        batch = await self._client.batches.retrieve(batch_id)
+        self.file_handler.add_file(batch.input_file_id)
+        self.file_handler.add_file(batch.output_file_id)
+        self.file_handler.add_file(batch.error_file_id)
+        return batch
 
     async def retrieve_file(self, file_id: str) -> HttpxBinaryResponseContent:
         return await self._client.files.content(file_id)

{parallex-0.1.3 → parallex-0.1.4}/parallex/ai/uploader.py

@@ -16,7 +16,7 @@ async def upload_images_for_processing(
     image_files: list[ImageFile],
     temp_directory: str,
     prompt_text: str,
-):
+) -> list[BatchFile]:
     """Base64 encodes image, converts to expected jsonl format and uploads"""
     trace_id = image_files[0].trace_id
     current_index = 0

parallex-0.1.4/parallex/file_management/remote_file_handler.py

@@ -0,0 +1,7 @@
+class RemoteFileHandler:
+    def __init__(self):
+        self.created_files = set()
+
+    def add_file(self, file_name: str) -> None:
+        if file_name is not None:
+            self.created_files.add(file_name)

{parallex-0.1.3 → parallex-0.1.4}/parallex/parallex.py

@@ -9,7 +9,8 @@ from parallex.ai.output_processor import process_output
 from parallex.ai.uploader import upload_images_for_processing
 from parallex.file_management.converter import convert_pdf_to_images
 from parallex.file_management.file_finder import add_file_to_temp_directory
-from parallex.models.image_file import ImageFile
+from parallex.file_management.remote_file_handler import RemoteFileHandler
+from parallex.models.batch_file import BatchFile
 from parallex.models.parallex_callable_output import ParallexCallableOutput
 from parallex.models.upload_batch import UploadBatch
 from parallex.utils.constants import DEFAULT_PROMPT
@@ -26,9 +27,32 @@ async def parallex(
     log_level: Optional[str] = "ERROR",
 ) -> ParallexCallableOutput:
     setup_logger(log_level)
-    with tempfile.TemporaryDirectory() as temp_directory:
-        open_ai_client = OpenAIClient(model=model)
+    remote_file_handler = RemoteFileHandler()
+    open_ai_client = OpenAIClient(model=model, remote_file_handler=remote_file_handler)
+    try:
+        return await _execute(
+            open_ai_client=open_ai_client,
+            pdf_source_url=pdf_source_url,
+            post_process_callable=post_process_callable,
+            concurrency=concurrency,
+            prompt_text=prompt_text,
+        )
+    except Exception as e:
+        logger.error(f"Error occurred: {e}")
+    finally:
+        for file in remote_file_handler.created_files:
+            logger.info(f"deleting - {file}")
+            await open_ai_client.delete_file(file)
 
+
+async def _execute(
+    open_ai_client: OpenAIClient,
+    pdf_source_url: str,
+    post_process_callable: Optional[Callable[..., None]] = None,
+    concurrency: Optional[int] = 20,
+    prompt_text: Optional[str] = DEFAULT_PROMPT,
+) -> ParallexCallableOutput:
+    with tempfile.TemporaryDirectory() as temp_directory:
         raw_file = await add_file_to_temp_directory(
             pdf_source_url=pdf_source_url, temp_directory=temp_directory
         )
@@ -47,7 +71,7 @@ async def parallex(
         start_batch_tasks = []
         for file in batch_files:
             batch_task = asyncio.create_task(
-                _create_images_and_batch_jobs(
+                _create_batch_jobs(
                     batch_file=file,
                     client=open_ai_client,
                     trace_id=trace_id,
@@ -55,11 +79,11 @@ async def parallex(
                 )
             )
             start_batch_tasks.append(batch_task)
-        batches = await asyncio.gather(*start_batch_tasks)
+        batch_jobs = await asyncio.gather(*start_batch_tasks)
 
         pages_tasks = []
         process_semaphore = asyncio.Semaphore(concurrency)
-        for batch in batches:
+        for batch in batch_jobs:
             page_task = asyncio.create_task(
                 _wait_and_create_pages(
                     batch=batch, client=open_ai_client, semaphore=process_semaphore
@@ -69,7 +93,7 @@ async def parallex(
         page_groups = await asyncio.gather(*pages_tasks)
 
         pages = [page for batch_pages in page_groups for page in batch_pages]
-        logger.debug(f"pages done. total pages- {len(pages)} - {trace_id}")
+        logger.info(f"pages done. total pages- {len(pages)} - {trace_id}")
         sorted_pages = sorted(pages, key=lambda x: x.page_number)
 
         # TODO add combined version of MD to output / save to file system
@@ -88,30 +112,23 @@ async def _wait_and_create_pages(
     batch: UploadBatch, client: OpenAIClient, semaphore: asyncio.Semaphore
 ):
     async with semaphore:
-        logger.debug(f"waiting for batch to complete - {batch.id} - {batch.trace_id}")
+        logger.info(f"waiting for batch to complete - {batch.id} - {batch.trace_id}")
         output_file_id = await wait_for_batch_completion(client=client, batch=batch)
-        logger.debug(f"batch completed - {batch.id} - {batch.trace_id}")
+        logger.info(f"batch completed - {batch.id} - {batch.trace_id}")
         page_responses = await process_output(
             client=client, output_file_id=output_file_id
         )
-        await _remove_global_batch_files(client=client, batch=batch)
         return page_responses
 
 
-async def _remove_global_batch_files(client: OpenAIClient, batch: UploadBatch):
-    file_ids = [batch.input_file_id, batch.output_file_id, batch.error_file_id]
-    for file_id in file_ids:
-        await client.delete_file(file_id)
-
-
-async def _create_images_and_batch_jobs(
-    batch_file: ImageFile,
+async def _create_batch_jobs(
+    batch_file: BatchFile,
     client: OpenAIClient,
     trace_id: UUID,
     semaphore: asyncio.Semaphore,
 ):
     async with semaphore:
-        batch = await create_batch(
+        upload_batch = await create_batch(
             client=client, file_id=batch_file.id, trace_id=trace_id
         )
-        return batch
+        return upload_batch

{parallex-0.1.3 → parallex-0.1.4}/parallex/utils/logger.py

@@ -1,8 +1,8 @@
 import logging
 
-from aiologger import Logger
+from aiologger.loggers.json import JsonLogger
 
-logger = Logger.with_default_handlers(name="parallex")
+logger = JsonLogger.with_default_handlers(name="parallex")
 
 
 def setup_logger(level: str = "ERROR"):
@@ -15,6 +15,4 @@ def setup_logger(level: str = "ERROR"):
         "NOTSET": logging.NOTSET,
     }.get(level, logging.INFO)
 
-    logging.basicConfig(
-        level=level, format="%(asctime)s - %(name)s - %(levelname)s - %(message)s"
-    )
+    logger.setLevel = level

{parallex-0.1.3 → parallex-0.1.4}/pyproject.toml

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "parallex"
-version = "0.1.3"
+version = "0.1.4"
 description = "PDF to markdown using Azure OpenAI batch processing"
 authors = ["Jeff Hostetler <jeff@summed.ai>", "Kevin Bao <kevin@summed.ai>"]
 repository = "https://github.com/Summed-AI/parallex"