vlmparse 0.1.6__py3-none-any.whl → 0.1.7__py3-none-any.whl

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/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/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/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/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/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/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.dist-info → vlmparse-0.1.7.dist-info}/METADATA

@@ -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.dist-info → vlmparse-0.1.7.dist-info}/RECORD

@@ -1,22 +1,22 @@
 vlmparse/base_model.py,sha256=4U4UPe8SNArliKnUf8pp8zQugWYsnhg9okylt7mrW1U,381
 vlmparse/build_doc.py,sha256=LAWrnFrqamN5PwJo57AUtQOPrMFGnCGw4gBjEKZ6pYo,2127
-vlmparse/cli.py,sha256=gY45YZe5LanN-ozG2vVtOaB2qyNWpjO2DvPmJeBi_wA,13045
-vlmparse/constants.py,sha256=7-47S01n4MI2ebR09bpdOo3_P16d-z-NVGsm6KJP8ls,110
+vlmparse/cli.py,sha256=JfR6gk0pdYAavJgFTVx4OcgWdiLktGoKJ8TcVcD_IHw,12235
+vlmparse/constants.py,sha256=DYaK7KtTW8p9MPb3iPvoP5H1r7ICRuIFo89P01q4uCI,184
 vlmparse/converter.py,sha256=F0JSY9sFYUggCvaUCb27kKGJJpnZKW2FStMDVJoIOeQ,7383
-vlmparse/converter_with_server.py,sha256=62kcEp0NjzDR2vVmEfCeeLlwbb8E3sWcseb2jjK7DpM,4861
+vlmparse/converter_with_server.py,sha256=zpUHDpHbDBs4Cj7dcVjvUQw0-U_InRNDC5Ekb_gehRM,6022
 vlmparse/registries.py,sha256=yBVrrhy61rSoLwdNV-z0C4lqIpTbLoWab3V6u7aSyNM,5797
 vlmparse/utils.py,sha256=rcVrtPiQVj_8HAmFQOu___72uYIapp_X89yxrMNCBow,1236
 vlmparse/clients/chandra.py,sha256=EulsCZdwOtm0pQ6CDm320U96k8aWFN4wKqCm1Xo7VCE,9775
 vlmparse/clients/deepseekocr.py,sha256=Uw6tPvP2KVsPDlz1ZUgYdbgQSjmFPuYeFDrGMMOTBAo,6501
 vlmparse/clients/docling.py,sha256=SAkLsqseuWfkuiel8FWR1G0Z5s-SZU3dE2JbsOvF4SA,5328
 vlmparse/clients/dotsocr.py,sha256=uGJoYEiDkP3-rmfdkAnMeAX-T4RZyEPoh6jmow5_-J8,10336
-vlmparse/clients/granite_docling.py,sha256=EQpsv5qSJG0HtMSacmJStER2sq4TGf1EMU5_NmJsl4g,4634
+vlmparse/clients/granite_docling.py,sha256=LMJAFjpSxcgLhsVxknSqrCC35MUTmklsE9PJZvMK2O8,4691
 vlmparse/clients/hunyuanocr.py,sha256=UFqaS4b8UM9EtizyrZIxlqcYlESmxm8xrQZP7lL6tkE,1857
 vlmparse/clients/lightonocr.py,sha256=wx1Im8Z3wlRWwYbPqnSd3LqTtdAU8CnX5mzu1BuCUY8,1314
 vlmparse/clients/mineru.py,sha256=6jZ1sKn2kGwUvD8gVs4PqEDH7uUXYK8pAB5Fr1JeqnY,3617
 vlmparse/clients/nanonetocr.py,sha256=BT5vaeerCsK5agvOaHK3NvLUqWd1FfDmrMmDYbp646I,1543
 vlmparse/clients/olmocr.py,sha256=A4Vl0meYpU5QPTML_OxyyRM07xCxtfrMZedgGMYEcuU,1851
-vlmparse/clients/openai_converter.py,sha256=nMKJeWH43UxHMMLns3wjX0pYjU5Xnai6IYxFmS9I63s,6193
+vlmparse/clients/openai_converter.py,sha256=bIDpR7Yn70eEp0pmzFoG2dDwY-mxCj3kH1IZS9BvXVQ,6266
 vlmparse/clients/paddleocrvl.py,sha256=qFBDj_UQocyq3WCh24tUOx9Ud7S9DfSm-1n3ztikY2s,1402
 vlmparse/clients/prompts.py,sha256=-J60lqxgRzlkQ9VsQLxmWsIMaDt-gNqWqWoqHIw9CLc,4228
 vlmparse/clients/pipe_utils/cleaner.py,sha256=oxBkBTOkluN1lmeNbzajRIe0_D__ZGwUOBaI_Ph0uxE,2396
@@ -24,13 +24,13 @@ vlmparse/clients/pipe_utils/html_to_md_conversion.py,sha256=cFFqzD2jCNw_968_eu3W
 vlmparse/clients/pipe_utils/utils.py,sha256=935ecIO446I0pstszE_1nrIPHn1Ffrxunq7fVd0dsd8,315
 vlmparse/data_model/box.py,sha256=lJsh4qhjgYXZF5vTSJ1qMXD5GVlBi2_SBedBMlfJikU,16868
 vlmparse/data_model/document.py,sha256=xheaMeStOj2c9GZKmdtxcEl_Dj44V5JyVp6JnTrSpH0,4615
-vlmparse/servers/docker_server.py,sha256=FBW2TvtUHUQNwj0rBP92shvoiJCGlc_oAKQBXN8260E,7114
-vlmparse/servers/utils.py,sha256=qy2-rnQTCQKt6CeTV5H74tvRTXyzBV2KswQiYW8Tf-k,8908
+vlmparse/servers/docker_server.py,sha256=UVU7VDloJ8Yfqj-WUv3Trti9AODcdC9JyTzW3sCM-l4,7032
+vlmparse/servers/utils.py,sha256=tIXhgbF9EVOJy2nYEguVq69gn9ATxtya_1F4wZSt68o,9454
 vlmparse/st_viewer/fs_nav.py,sha256=7GNH68h2Loh5pQ64Pe72-D2cs2BLhqRXevEmKdFmPX0,1616
 vlmparse/st_viewer/st_viewer.py,sha256=m2rQTtk5rlwErNmivNAg-4rkHkvNkvLhoJZxFQi7Dwk,2105
-vlmparse-0.1.6.dist-info/licenses/LICENSE,sha256=3TKJHk8hPBR5dbLWZ3IpfCftl-_m-iyBwpYQGZYxj14,1080
-vlmparse-0.1.6.dist-info/METADATA,sha256=Xad3SjAYvCzUvPo6A6GKvc3daxtf5XNs1AQjDlF7RmI,5597
-vlmparse-0.1.6.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
-vlmparse-0.1.6.dist-info/entry_points.txt,sha256=gD5berP6HwE2wNIkls-Lw5goiceA8uMgPEd7ifnFJXs,47
-vlmparse-0.1.6.dist-info/top_level.txt,sha256=k4ni-GNH_iAX7liQEsk_KY_c3xgZgt8k9fsSs9IXLXs,9
-vlmparse-0.1.6.dist-info/RECORD,,
+vlmparse-0.1.7.dist-info/licenses/LICENSE,sha256=3TKJHk8hPBR5dbLWZ3IpfCftl-_m-iyBwpYQGZYxj14,1080
+vlmparse-0.1.7.dist-info/METADATA,sha256=DP--8aCeLxAgvo6vvaDog7xzzMzvZywVvCrMiAKhDbo,5597
+vlmparse-0.1.7.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
+vlmparse-0.1.7.dist-info/entry_points.txt,sha256=gD5berP6HwE2wNIkls-Lw5goiceA8uMgPEd7ifnFJXs,47
+vlmparse-0.1.7.dist-info/top_level.txt,sha256=k4ni-GNH_iAX7liQEsk_KY_c3xgZgt8k9fsSs9IXLXs,9
+vlmparse-0.1.7.dist-info/RECORD,,