pyconverters-openai_vision 0.5.3__tar.gz → 0.5.7__tar.gz

{pyconverters_openai_vision-0.5.3 → pyconverters_openai_vision-0.5.7}/Dockerfile

@@ -1,4 +1,4 @@
-FROM python:3.8-slim-buster
+FROM python:3.8-slim-bookworm
 # Install prerequisites
 RUN apt-get update -y && \
     apt-get install -y \

{pyconverters_openai_vision-0.5.3 → pyconverters_openai_vision-0.5.7}/Jenkinsfile

@@ -176,7 +176,7 @@ pipeline {
             }
             if (sendEmailNotif("${PATH_HOME}/${JOB_NAME}", "${BUILD_NUMBER}")) {
               println 'sending Success Build notification'
-              CUSTOM_SUBJECT = '[CI - Jenkinzz SUCCESS] ' + CUSTOM_SUBJECT
+              def CUSTOM_SUBJECT = '[CI - Jenkinzz SUCCESS] ' + CUSTOM_SUBJECT
               emailext(
                   mimeType: 'text/html',
                   subject: CUSTOM_SUBJECT,
@@ -211,7 +211,7 @@ pipeline {
               println 'Exception occurred: ' + e.toString()
             }
             println 'sending Failure Build notification'
-            CUSTOM_SUBJECT = '[CI - Jenkinzz FAILURE] ' + CUSTOM_SUBJECT
+            def CUSTOM_SUBJECT = '[CI - Jenkinzz FAILURE] ' + CUSTOM_SUBJECT
             emailext(
                 mimeType: 'text/html',
                 subject: CUSTOM_SUBJECT,
@@ -235,7 +235,7 @@ pipeline {
 
 // return FLIT_USERNAME from given file
 def getUserName(path) {
-  USERNAME = sh(
+  def USERNAME = sh(
                  script: "grep FLIT_USERNAME ${path}|cut -d '=' -f2",
                  returnStdout: true
                ).trim()
@@ -244,7 +244,7 @@ def getUserName(path) {
 
 // return FLIT_PASSWORD from given file
 def getUserPass(path) {
-  USERPASS = sh(
+  def USERPASS = sh(
                  script: "grep FLIT_PASSWORD ${path}|cut -d '=' -f2",
                  returnStdout: true
                ).trim()
@@ -264,11 +264,11 @@ def switchEmailNotif(toggle, build) {
 
 // return true if emailNotif file present
 boolean sendEmailNotif(path, build) {
-  emailNotif = sh(
+  def emailNotif = sh(
                  script: "find ${path} -name '.emailNotif'|wc -l",
                  returnStdout: true
                ).trim()
-  emailContent = ''
+  def emailContent = ''
   if (emailNotif == '1') {
     emailContent = sh(
                      script: "cat ${path}/.emailNotif",
@@ -279,7 +279,7 @@ boolean sendEmailNotif(path, build) {
 }
 
 def analyseBuildCause() {
-  upstreamProjects = ['pymultirole_plugins']
+  String[] upstreamProjects = ['pymultirole_plugins']
   boolean upstreamRunning = false
   String jobName
   // iterate over upstreamProjects

{pyconverters_openai_vision-0.5.3 → pyconverters_openai_vision-0.5.7}/PKG-INFO

@@ -1,6 +1,6 @@
-Metadata-Version: 2.1
+Metadata-Version: 2.4
 Name: pyconverters-openai_vision
-Version: 0.5.3
+Version: 0.5.7
 Summary: OpenAIVision converter
 Home-page: https://kairntech.com/
 Author: Olivier Terrier

{pyconverters_openai_vision-0.5.3 → pyconverters_openai_vision-0.5.7}/pyconverters_openai_vision/__init__.py

@@ -1,2 +1,2 @@
 """OpenAIVision converter"""
-__version__ = "0.5.3"
+__version__ = "0.5.7"

{pyconverters_openai_vision-0.5.3 → pyconverters_openai_vision-0.5.7}/pyconverters_openai_vision/openai_vision.py

@@ -1,7 +1,9 @@
 import base64
 import os
+import re
 from enum import Enum
 from logging import Logger
+from re import Pattern
 from typing import List, cast, Type, Dict, Any
 
 import filetype as filetype
@@ -24,12 +26,13 @@ class OpenAIVisionBaseParameters(ConverterParameters):
         None, extra="internal"
     )
     prompt: str = Field(
-        "Describe the image with a lot of details",
+        """If the attached file is an image: describe the image with a lot of details.",
+        If the attached file is a PDF document: convert the PDF document into Markdown format. The output must be just the markdown result without any explanation or introductory prefix.""",
         description="""Contains the prompt as a string""",
         extra="multiline",
     )
     max_tokens: int = Field(
-        256,
+        16384,
         description="""The maximum number of tokens to generate in the completion.
     The token count of your prompt plus max_tokens cannot exceed the model's context length.
     Most models have a context length of 2048 tokens (except for the newest models, which support 4096).""",
@@ -85,13 +88,14 @@ class OpenAIVisionBaseParameters(ConverterParameters):
 class OpenAIVisionModel(str, Enum):
     gpt_4o_mini = "gpt-4o-mini"
     gpt_4o = "gpt-4o"
+    o3_mini = "o3-mini"
 
 
 class OpenAIVisionParameters(OpenAIVisionBaseParameters):
     model: OpenAIVisionModel = Field(
         OpenAIVisionModel.gpt_4o_mini,
-        description="""The [OpenAI model](https://platform.openai.com/docs/models) used for speech to text transcription. Options currently available:</br>
-                        <li>`whisper-1` - state-of-the-art open source large-v2 Whisper model.
+        description="""The [OpenAI model](https://platform.openai.com/docs/models) used for vision. Options currently available:</br>
+
                         """, extra="pipeline-naming-hint"
     )
 
@@ -126,10 +130,24 @@ class OpenAIVisionConverterBase(ConverterBase):
     You input some text as a prompt, and the model will generate a text completion that attempts to match whatever context or pattern you gave it."""
     PREFIX: str = ""
 
-    def compute_args(self, params: OpenAIVisionBaseParameters, source: UploadFile
+    def compute_args(self, params: OpenAIVisionBaseParameters, source: UploadFile, kind
                      ) -> Dict[str, Any]:
         data = source.file.read()
         rv = base64.b64encode(data)
+        if kind.mime.startswith("image"):
+            binary_block = {
+                "type": "image_url",
+                "image_url": {
+                    "url": f"data:image/jpeg;base64,{rv.decode('utf-8')}"
+                }
+            }
+        else:
+            binary_block = {
+                "type": "file",
+                "file": {
+                    "filename": source.filename,
+                    "file_data": f"data:application/pdf;base64,{rv.decode('utf-8')}"}
+            }
         messages = [{"role": "system", "content": params.system_prompt}] if params.system_prompt is not None else []
         messages.append({"role": "user",
                          "content": [
@@ -137,12 +155,8 @@ class OpenAIVisionConverterBase(ConverterBase):
                                  "type": "text",
                                  "text": params.prompt
                              },
-                             {
-                                 "type": "image_url",
-                                 "image_url": {
-                                     "url": f"data:image/jpeg;base64,{rv.decode('utf-8')}"
-                                 }
-                             }]})
+                             binary_block
+                         ]})
         kwargs = {
             'model': params.model_str,
             'messages': messages,
@@ -156,11 +170,18 @@ class OpenAIVisionConverterBase(ConverterBase):
         return kwargs
 
     def compute_result(self, **kwargs):
+        pattern: Pattern = re.compile(r"```(?:markdown\s+)?(\W.*?)```", re.DOTALL)
+        """Regex pattern to parse the output."""
         response = openai_chat_completion(self.PREFIX, **kwargs)
         contents = []
         for choice in response.choices:
             if choice.message.content:
-                contents.append(choice.message.content)
+                if "```" in choice.message.content:
+                    action_match = pattern.search(choice.message.content)
+                    if action_match is not None:
+                        contents.append(action_match.group(1).strip())
+                else:
+                    contents.append(choice.message.content)
         if contents:
             result = "\n".join(contents)
         return result
@@ -178,9 +199,9 @@ class OpenAIVisionConverterBase(ConverterBase):
         try:
             kind = filetype.guess(source.file)
             source.file.seek(0)
-            if kind.mime.startswith("image"):
+            if kind.mime.startswith("image") or kind.mime.endswith("pdf"):
                 result = None
-                kwargs = self.compute_args(params, source)
+                kwargs = self.compute_args(params, source, kind)
                 if kwargs['model'] != NO_DEPLOYED_MODELS:
                     result = self.compute_result(**kwargs)
                 if result:
@@ -189,7 +210,7 @@ class OpenAIVisionConverterBase(ConverterBase):
         except BaseException as err:
             raise err
         if doc is None:
-            raise TypeError(f"Conversion of audio file {source.filename} failed")
+            raise TypeError(f"Conversion of file {source.filename} failed")
         return [doc]
 
     @classmethod

{pyconverters_openai_vision-0.5.3 → pyconverters_openai_vision-0.5.7}/setup.py

@@ -43,7 +43,7 @@ entry_points = \
                           'pyconverters_openai_vision.openai_vision:OpenAIVisionConverter']}
 
 setup(name='pyconverters-openai_vision',
-      version='0.5.3',
+      version='0.5.7',
       description='OpenAIVision converter',
       author='Olivier Terrier',
       author_email='olivier.terrier@kairntech.com',

{pyconverters_openai_vision-0.5.3 → pyconverters_openai_vision-0.5.7}/tests/test_openai_vision.py

@@ -1,3 +1,4 @@
+import os
 from pathlib import Path
 from typing import List
 
@@ -7,7 +8,7 @@ from starlette.datastructures import UploadFile
 
 from pyconverters_openai_vision.openai_vision import (
     OpenAIVisionConverter,
-    OpenAIVisionParameters, DeepInfraOpenAIVisionParameters, DeepInfraOpenAIVisionConverter
+    OpenAIVisionParameters, DeepInfraOpenAIVisionParameters, DeepInfraOpenAIVisionConverter, OpenAIVisionModel
 )
 
 
@@ -37,6 +38,19 @@ def test_openai():
         assert 'kairntech' in doc0.text.lower()
 
 
+@pytest.mark.skip(reason="Not a test")
+def test_openai_pdf():
+    converter = OpenAIVisionConverter()
+    parameters = OpenAIVisionParameters(model=OpenAIVisionModel.gpt_4o_mini)
+    testdir = Path(__file__).parent
+    source = Path(testdir, 'data/Sodexo_URD_2023_FR - 4p.pdf')
+    with source.open("rb") as fin:
+        docs: List[Document] = converter.convert(UploadFile(source.name, fin, 'application/pdf'), parameters)
+        assert len(docs) == 1
+        doc0 = docs[0]
+        assert 'une performance solide au cours' in doc0.text.lower()
+
+
 @pytest.mark.skip(reason="Not a test")
 def test_deepinfra():
     converter = DeepInfraOpenAIVisionConverter()
@@ -55,3 +69,26 @@ def test_deepinfra():
         assert len(docs) == 1
         doc0 = docs[0]
         assert 'generative ai' in doc0.text.lower()
+
+
+@pytest.mark.skip(reason="Not a test")
+def test_runpod():
+    os.environ["OPENAI_API_BASE"] = "https://api.runpod.ai/v2/vllm-9jnu8ajtktj5ay/openai/v1"
+    os.environ["OPENAI_MODEL"] = "mistralai/Pixtral-12B-2409"
+    os.environ["OPENAI_API_KEY"] = os.getenv("RUNPOD_API_KEY")
+    converter = OpenAIVisionConverter()
+    parameters = OpenAIVisionParameters()
+    testdir = Path(__file__).parent
+    source = Path(testdir, 'data/colducoq.jpg')
+    with source.open("rb") as fin:
+        docs: List[Document] = converter.convert(UploadFile(source.name, fin, 'image/jpeg'), parameters)
+        assert len(docs) == 1
+        doc0 = docs[0]
+        assert 'dent de crolles' in doc0.text.lower()
+
+    source = Path(testdir, 'data/webinar.png')
+    with source.open("rb") as fin:
+        docs: List[Document] = converter.convert(UploadFile(source.name, fin, 'image/png'), parameters)
+        assert len(docs) == 1
+        doc0 = docs[0]
+        assert 'generative ai' in doc0.text.lower()