vlmparse 0.1.6__tar.gz → 0.1.7__tar.gz

{vlmparse-0.1.6 → vlmparse-0.1.7}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: vlmparse
-Version: 0.1.6
+Version: 0.1.7
 Requires-Python: >=3.11.0
 Description-Content-Type: text/markdown
 License-File: LICENSE

{vlmparse-0.1.6 → vlmparse-0.1.7}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "vlmparse"
-version = "0.1.6"
+version = "0.1.7"
 authors = []
 description = ""
 readme = "README.md"

{vlmparse-0.1.6 → vlmparse-0.1.7}/tests/test_batch_parser.py

@@ -34,6 +34,8 @@ class TestBatchParser:
         mock_server = MagicMock()
         mock_client = MagicMock()
 
+        mock_server.start.return_value = ("http://localhost:8056", None)
+
         mock_config.get_server.return_value = mock_server
         mock_config.get_client.return_value = mock_client
         mock_docker_registry.get.return_value = mock_config
@@ -87,7 +89,11 @@ class TestBatchParser:
         """Test that parse method updates client configuration and calls batch."""
         # Setup mocks
         mock_config = MagicMock()
+        mock_server = MagicMock()
         mock_client = MagicMock()
+
+        mock_server.start.return_value = ("http://localhost:8056", None)
+        mock_config.get_server.return_value = mock_server
         mock_config.get_client.return_value = mock_client
         mock_docker_registry.get.return_value = mock_config
 
@@ -136,7 +142,11 @@ class TestBatchParser:
 
         # Setup mocks
         mock_config = MagicMock()
+        mock_server = MagicMock()
         mock_client = MagicMock()
+
+        mock_server.start.return_value = ("http://localhost:8056", None)
+        mock_config.get_server.return_value = mock_server
         mock_config.get_client.return_value = mock_client
         mock_docker_registry.get.return_value = mock_config
 
@@ -161,7 +171,11 @@ class TestBatchParser:
     ):
         """Test that retrylast raises ValueError if no previous runs found."""
         mock_config = MagicMock()
+        mock_server = MagicMock()
         mock_client = MagicMock()
+
+        mock_server.start.return_value = ("http://localhost:8056", None)
+        mock_config.get_server.return_value = mock_server
         mock_config.get_client.return_value = mock_client
         mock_docker_registry.get.return_value = mock_config
 

{vlmparse-0.1.6 → vlmparse-0.1.7}/tests/test_cli.py

@@ -46,7 +46,7 @@ def mock_docker_server():
         mock_container.name = "test_container_name"
 
         # Configure server.start() to return base_url and container
-        mock_server.start.return_value = ("http://localhost:8056", mock_container)
+        mock_server.start.return_value = ("http://localhost:8056", None)
         mock_config.get_server.return_value = mock_server
         mock_registry.get.return_value = mock_config
 
@@ -79,7 +79,7 @@ class TestServeCommand:
         cli.serve(model="lightonocr")
 
         # Verify registry was called with correct model
-        mock_registry.get.assert_called_once_with("lightonocr")
+        mock_registry.get.assert_called_once_with("lightonocr", default=True)
 
         # Verify port was set to default
         assert mock_config.docker_port == 8056
@@ -128,7 +128,7 @@ class TestServeCommand:
             # Should not raise an exception, just log warning
             cli.serve(model="unknown_model")
 
-            mock_registry.get.assert_called_once_with("unknown_model")
+            mock_registry.get.assert_called_once_with("unknown_model", default=True)
 
 
 class TestConvertCommand:
@@ -216,6 +216,7 @@ class TestConvertCommand:
                 mock_converter = MagicMock()
                 mock_converter.batch.return_value = None
 
+                mock_server.start.return_value = ("http://localhost:8056", None)
                 mock_docker_config.get_server.return_value = mock_server
                 mock_docker_config.get_client.return_value = mock_converter
                 mock_docker_registry.get.return_value = mock_docker_config
@@ -241,6 +242,7 @@ class TestConvertCommand:
                 mock_converter = MagicMock()
                 mock_converter.batch.return_value = None
 
+                mock_server.start.return_value = ("http://localhost:8056", None)
                 mock_docker_config.get_server.return_value = mock_server
                 mock_docker_config.get_client.return_value = mock_converter
                 mock_docker_registry.get.return_value = mock_docker_config
@@ -354,6 +356,7 @@ class TestCLIIntegration:
 
                 mock_converter.batch.return_value = None
 
+                mock_server.start.return_value = ("http://localhost:8056", None)
                 mock_docker_config.get_server.return_value = mock_server
                 mock_docker_config.get_client.return_value = mock_converter
                 mock_docker_registry.get.return_value = mock_docker_config
@@ -377,7 +380,7 @@ class TestCLIIntegration:
             mock_container = MagicMock()
             mock_container.id = "test_id"
             mock_container.name = "test_name"
-            mock_server.start.return_value = ("http://localhost:8056", mock_container)
+            mock_server.start.return_value = ("http://localhost:8056", None)
             mock_docker_config.get_server.return_value = mock_server
             mock_docker_registry.get.return_value = mock_docker_config
 
@@ -464,7 +467,7 @@ class TestCLIConvertInDepth:
 
             mock_server.start.return_value = (
                 "http://localhost:8000/v1",
-                mock_container,
+                None,
             )
             mock_docker_config.get_server.return_value = mock_server
 
@@ -727,36 +730,3 @@ class TestCLIConvertInDepth:
         assert len(page_contents) == 2
         assert "Page 1" in page_contents[0]
         assert "Page 2" in page_contents[1]
-
-    def test_convert_with_nanonet_postprompt(self, cli, file_path, mock_openai_api):
-        """Test that Nanonet model uses its postprompt correctly."""
-        from vlmparse.registries import converter_config_registry
-
-        messages_sent = []
-
-        async def capture_messages(*args, **kwargs):
-            messages_sent.append(kwargs.get("messages", []))
-            mock_response = MagicMock()
-            mock_response.choices = [MagicMock()]
-            mock_response.choices[0].message.content = "Extracted text"
-            return mock_response
-
-        mock_openai_api.chat.completions.create = AsyncMock(
-            side_effect=capture_messages
-        )
-
-        cli.convert(
-            inputs=[str(file_path)],
-            model="nanonets/Nanonets-OCR2-3B",
-            uri="http://mocked-api/v1",
-        )
-
-        # Verify messages were sent (2 pages)
-        assert len(messages_sent) == 2
-
-        # Verify config has postprompt set
-        config = converter_config_registry.get(
-            "nanonets/Nanonets-OCR2-3B", uri="http://mocked-api/v1"
-        )
-        assert config.postprompt is not None
-        assert len(config.postprompt) > 0

{vlmparse-0.1.6 → vlmparse-0.1.7}/vlmparse/cli.py

@@ -11,8 +11,8 @@ class DParseCLI:
         model: str,
         port: int | None = None,
         gpus: str | None = None,
-        vllm_kwargs: dict | None = None,
-        forget_predefined_vllm_kwargs: bool = False,
+        vllm_args: list[str] | None = None,
+        forget_predefined_vllm_args: bool = False,
     ):
         """Deploy a VLLM server in a Docker container.
 
@@ -20,43 +20,26 @@ class DParseCLI:
             model: Model name
             port: VLLM server port (default: 8056)
             gpus: Comma-separated GPU device IDs (e.g., "0" or "0,1,2"). If not specified, all GPUs will be used.
-            vllm_kwargs: Additional keyword arguments to pass to the VLLM server.
-            forget_predefined_vllm_kwargs: If True, the predefined VLLM kwargs from the docker config will be replaced by vllm_kwargs otherwise the predefined kwargs will be updated with vllm_kwargs with a risk of collision of argument names.
+            vllm_args: Additional keyword arguments to pass to the VLLM server.
+            forget_predefined_vllm_args: If True, the predefined VLLM kwargs from the docker config will be replaced by vllm_args otherwise the predefined kwargs will be updated with vllm_args with a risk of collision of argument names.
         """
-        if port is None:
-            port = 8056
 
-        from vlmparse.registries import docker_config_registry
+        from vlmparse.converter_with_server import start_server
 
-        docker_config = docker_config_registry.get(model)
-        if docker_config is None:
-            logger.warning(
-                f"No Docker configuration found for model: {model}, using default configuration"
-            )
-            return
-
-        docker_config.docker_port = port
-
-        # Only override GPU configuration if explicitly specified
-        # This preserves CPU-only settings from the config
-        if gpus is not None:
-            docker_config.gpu_device_ids = [g.strip() for g in str(gpus).split(",")]
-        server = docker_config.get_server(auto_stop=False)
-
-        if server is None:
-            logger.error(f"Model server not found for model: {model}")
-            return
-
-        # Deploy server and leave it running (cleanup=False)
-        logger.info(
-            f"Deploying VLLM server for {docker_config.model_name} on port {port}..."
+        base_url, container, _, _ = start_server(
+            model=model,
+            gpus=gpus,
+            port=port,
+            with_vllm_server=True,
+            vllm_args=vllm_args,
+            forget_predefined_vllm_args=forget_predefined_vllm_args,
+            auto_stop=False,
         )
 
-        base_url, container = server.start()
-
         logger.info(f"✓ VLLM server ready at {base_url}")
-        logger.info(f"✓ Container ID: {container.id}")
-        logger.info(f"✓ Container name: {container.name}")
+        if container is not None:
+            logger.info(f"✓ Container ID: {container.id}")
+            logger.info(f"✓ Container name: {container.name}")
 
     def convert(
         self,
@@ -69,7 +52,6 @@ class DParseCLI:
         with_vllm_server: bool = False,
         concurrency: int = 10,
         dpi: int | None = None,
-        vllm_kwargs: dict | None = None,
         debug: bool = False,
     ):
         """Parse PDF documents and save results.
@@ -84,7 +66,6 @@ class DParseCLI:
             mode: Output mode - "document" (save as JSON zip), "md" (save as markdown file), "md_page" (save as folder of markdown pages)
             with_vllm_server: If True, a local VLLM server will be deployed if the model is not found in the registry. Note that if the model is in the registry and the uri is None, the server will be anyway deployed.
             dpi: DPI to use for the conversion. If not specified, the default DPI will be used.
-            vllm_kwargs: Additional keyword arguments to pass to the VLLM server.
             debug: If True, run in debug mode (single-threaded, no concurrency)
         """
         from vlmparse.converter_with_server import ConverterWithServer
@@ -95,7 +76,6 @@ class DParseCLI:
             gpus=gpus,
             with_vllm_server=with_vllm_server,
             concurrency=concurrency,
-            vllm_kwargs=vllm_kwargs,
         ) as converter_with_server:
             return converter_with_server.parse(
                 inputs=inputs, out_folder=out_folder, mode=mode, dpi=dpi, debug=debug

{vlmparse-0.1.6 → vlmparse-0.1.7}/vlmparse/clients/granite_docling.py

@@ -34,6 +34,7 @@ class GraniteDoclingDockerServerConfig(VLLMDockerServerConfig):
 class GraniteDoclingConverterConfig(OpenAIConverterConfig):
     """Granite Docling converter configuration."""
 
+    model_name: str = "ibm-granite/granite-docling-258M"
     preprompt: str | None = None
     postprompt: str | None = "Convert this page to docling."
     completion_kwargs: dict | None = {

{vlmparse-0.1.6 → vlmparse-0.1.7}/vlmparse/clients/openai_converter.py

@@ -1,5 +1,5 @@
 import os
-from typing import Literal
+from typing import Literal, Optional
 
 from loguru import logger
 from pydantic import Field
@@ -101,7 +101,7 @@ class OpenAIConverterClient(BaseConverter):
 
     async def _get_chat_completion(
         self, messages: list[dict], completion_kwargs: dict | None = None
-    ) -> tuple[str, "CompletionUsage"]:  # noqa: F821
+    ) -> tuple[str, Optional["CompletionUsage"]]:  # noqa: F821
         """Helper to handle chat completion with optional streaming."""
         if completion_kwargs is None:
             completion_kwargs = self.config.completion_kwargs
@@ -117,7 +117,8 @@ class OpenAIConverterClient(BaseConverter):
             async for chunk in response_stream:
                 if chunk.choices and chunk.choices[0].delta.content:
                     response_parts.append(chunk.choices[0].delta.content)
-            return "".join(response_parts)
+
+            return "".join(response_parts), None
         else:
             response_obj = await self.model.chat.completions.create(
                 model=self.config.llm_params.model_name,
@@ -175,9 +176,10 @@ class OpenAIConverterClient(BaseConverter):
 
         text = html_to_md_keep_tables(text)
         page.text = text
-        page.prompt_tokens = usage.prompt_tokens
-        page.completion_tokens = usage.completion_tokens
-        if hasattr(usage, "reasoning_tokens"):
-            page.reasoning_tokens = usage.reasoning_tokens
+        if usage is not None:
+            page.prompt_tokens = usage.prompt_tokens
+            page.completion_tokens = usage.completion_tokens
+            if hasattr(usage, "reasoning_tokens"):
+                page.reasoning_tokens = usage.reasoning_tokens
 
         return page

{vlmparse-0.1.6 → vlmparse-0.1.7}/vlmparse/constants.py

@@ -1,2 +1,5 @@
+import os
+
 IMAGE_EXTENSIONS = [".jpg", ".jpeg", ".png", ".tiff", ".tif", ".bmp", ".gif", ".webp"]
 PDF_EXTENSION = ".pdf"
+DEFAULT_SERVER_PORT = os.getenv("VLMPARSE_DEFAULT_PORT", 8056)

{vlmparse-0.1.6 → vlmparse-0.1.7}/vlmparse/converter_with_server.py

@@ -5,10 +5,61 @@ from typing import Literal
 
 from loguru import logger
 
+from vlmparse.constants import DEFAULT_SERVER_PORT
 from vlmparse.servers.utils import get_model_from_uri
 from vlmparse.utils import get_file_paths
 
 
+def start_server(
+    model: str,
+    gpus: str,
+    port: None | int = None,
+    with_vllm_server: bool = True,
+    vllm_args: list[str] = {},
+    forget_predefined_vllm_args: bool = False,
+    auto_stop: bool = False,
+):
+    from vlmparse.registries import docker_config_registry
+
+    base_url = ""
+    container = None
+    docker_config = docker_config_registry.get(model, default=with_vllm_server)
+
+    if port is None:
+        port = DEFAULT_SERVER_PORT
+
+    if docker_config is None:
+        logger.warning(
+            f"No Docker configuration found for model: {model}, using default configuration"
+        )
+        return "", container, None, docker_config
+
+    gpu_device_ids = None
+    if gpus is not None:
+        gpu_device_ids = [g.strip() for g in str(gpus).split(",")]
+
+    if docker_config is not None:
+        if port is not None:
+            docker_config.docker_port = port
+        docker_config.gpu_device_ids = gpu_device_ids
+        docker_config.update_command_args(
+            vllm_args,
+            forget_predefined_vllm_args=forget_predefined_vllm_args,
+        )
+
+        logger.info(
+            f"Deploying VLLM server for {docker_config.model_name} on port {port}..."
+        )
+        server = docker_config.get_server(auto_stop=auto_stop)
+        if server is None:
+            logger.error(f"Model server not found for model: {model}")
+            return "", container, None, docker_config
+
+        base_url, container = server.start()
+
+    return base_url, container, server, docker_config
+
+
 class ConverterWithServer:
     def __init__(
         self,
@@ -18,8 +69,8 @@ class ConverterWithServer:
         port: int | None = None,
         with_vllm_server: bool = False,
         concurrency: int = 10,
-        vllm_kwargs: dict | None = None,
-        forget_predefined_vllm_kwargs: bool = False,
+        vllm_args: dict | None = None,
+        forget_predefined_vllm_args: bool = False,
     ):
         self.model = model
         self.uri = uri
@@ -27,8 +78,8 @@ class ConverterWithServer:
         self.gpus = gpus
         self.with_vllm_server = with_vllm_server
         self.concurrency = concurrency
-        self.vllm_kwargs = vllm_kwargs
-        self.forget_predefined_vllm_kwargs = forget_predefined_vllm_kwargs
+        self.vllm_args = vllm_args
+        self.forget_predefined_vllm_args = forget_predefined_vllm_args
         self.server = None
         self.client = None
 
@@ -36,32 +87,20 @@ class ConverterWithServer:
             self.model = get_model_from_uri(self.uri)
 
     def start_server_and_client(self):
-        from vlmparse.registries import (
-            converter_config_registry,
-            docker_config_registry,
-        )
-
-        gpu_device_ids = None
-        if self.gpus is not None:
-            gpu_device_ids = [g.strip() for g in self.gpus.split(",")]
+        from vlmparse.registries import converter_config_registry
 
         if self.uri is None:
-            docker_config = docker_config_registry.get(
-                self.model, default=self.with_vllm_server
+            _, _, self.server, docker_config = start_server(
+                model=self.model,
+                gpus=self.gpus,
+                port=self.port,
+                with_vllm_server=self.with_vllm_server,
+                vllm_args=self.vllm_args,
+                forget_predefined_vllm_args=self.forget_predefined_vllm_args,
+                auto_stop=True,
             )
 
             if docker_config is not None:
-                if self.port is not None:
-                    docker_config.docker_port = self.port
-                docker_config.gpu_device_ids = gpu_device_ids
-                docker_config.update_command_args(
-                    self.vllm_kwargs,
-                    forget_predefined_vllm_kwargs=self.forget_predefined_vllm_kwargs,
-                )
-                self.server = docker_config.get_server(auto_stop=True)
-
-                self.server.start()
-
                 self.client = docker_config.get_client()
             else:
                 self.client = converter_config_registry.get(self.model).get_client()

{vlmparse-0.1.6 → vlmparse-0.1.7}/vlmparse/servers/docker_server.py

@@ -49,15 +49,14 @@ class DockerServerConfig(BaseModel):
 
     def update_command_args(
         self,
-        vllm_kwargs: dict | None = None,
-        forget_predefined_vllm_kwargs: bool = False,
+        vllm_args: dict | None = None,
+        forget_predefined_vllm_args: bool = False,
     ) -> list[str]:
-        if vllm_kwargs is not None:
-            new_kwargs = [f"--{k}={v}" for k, v in vllm_kwargs.items()]
-            if forget_predefined_vllm_kwargs:
-                self.command_args = new_kwargs
+        if vllm_args is not None:
+            if forget_predefined_vllm_args:
+                self.command_args = vllm_args
             else:
-                self.command_args.extend(new_kwargs)
+                self.command_args.extend(vllm_args)
 
         return self.command_args
 

{vlmparse-0.1.6 → vlmparse-0.1.7}/vlmparse/servers/utils.py

@@ -2,6 +2,7 @@ import getpass
 import time
 from contextlib import contextmanager
 from pathlib import Path
+from urllib.parse import parse_qsl, urlparse
 
 import docker
 from loguru import logger
@@ -222,25 +223,52 @@ def docker_server(
             logger.info("Container stopped")
 
 
+def normalize_uri(uri: str) -> tuple:
+    u = urlparse(uri)
+
+    # --- Normalize scheme ---
+    scheme = (u.scheme or "http").lower()
+
+    # --- Normalize host ---
+    host = (u.hostname or "").lower()
+    if host in ("localhost", "0.0.0.0"):
+        host = "localhost"
+
+    # --- Normalize port (apply defaults) ---
+    if u.port:
+        port = u.port
+    else:
+        port = 443 if scheme == "https" else 80
+
+    # --- Normalize path ---
+    # Treat empty path as "/" and remove trailing slash (except root)
+    path = u.path or "/"
+    if path != "/" and path.endswith("/"):
+        path = path.rstrip("/")
+
+    # Collapse duplicate slashes
+    while "//" in path:
+        path = path.replace("//", "/")
+
+    # --- Normalize query parameters (sorted) ---
+    query_pairs = parse_qsl(u.query, keep_blank_values=True)
+    query = "&".join(f"{k}={v}" for k, v in sorted(query_pairs))
+
+    return (scheme, host, port, path, query)
+
+
 def get_model_from_uri(uri: str) -> str:
     model = None
     client = docker.from_env()
     containers = client.containers.list()
+
+    uri = normalize_uri(uri)
+
     for container in containers:
         c_uri = container.labels.get("vlmparse_uri")
         c_model = container.labels.get("vlmparse_model_name")
-        if c_uri is not None:
-            c_uri = c_uri.replace("localhost", "0.0.0.0")
-
-        # Check if user URI matches container URI (ignoring /v1 suffix if missing)
-        if c_uri and (
-            c_uri == uri or c_uri.startswith(uri.rstrip("/")) or uri.startswith(c_uri)
-        ):
-            # Update URI to the correct one from container (likely has /v1)
-            if len(c_uri) > len(uri.rstrip("/")):
-                logger.info(f"Updating URI from {uri} to {c_uri}")
-                uri = c_uri
 
+        if c_uri and uri == normalize_uri(c_uri):
             # Infer model if not provided
             if model is None and c_model:
                 logger.info(f"Inferred model {c_model} from container")

{vlmparse-0.1.6 → vlmparse-0.1.7}/vlmparse.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: vlmparse
-Version: 0.1.6
+Version: 0.1.7
 Requires-Python: >=3.11.0
 Description-Content-Type: text/markdown
 License-File: LICENSE