Modernize dependencies, remove DeepSeek support, and add dynamic model detection

.gitignore

@@ -14,4 +14,6 @@ wheels/
 .env.*
 
 # Virtual environments
-.venv
+.venv
+
+uv.lock

README.md

@@ -121,7 +121,7 @@ markdown_pages = parser.convert_pdf(pdf_path)
 > [!TIP]
 > Please refer to [FAQs](docs/faq.md) for more details on how to improve the performance of locally hosted vision models.
 
-### API Models Usage (OpenAI, Azure OpenAI, Gemini, DeepSeek)
+### API Models Usage (OpenAI, Azure OpenAI, Gemini)
 
 ```python
 from vision_parse import VisionParser
@@ -155,7 +155,7 @@ parser = VisionParser(
 
 # Initialize parser with Google Gemini model
 parser = VisionParser(
-    model_name="gemini-1.5-flash",
+    model_name="gemini-2.5-flash",
     api_key="your-gemini-api-key", # Get the Gemini API key from https://aistudio.google.com/app/apikey
     temperature=0.7,
     top_p=0.4,
@@ -164,16 +164,6 @@ parser = VisionParser(
     enable_concurrency=True,
 )
 
-# Initialize parser with DeepSeek model
-parser = VisionParser(
-    model_name="deepseek-chat",
-    api_key="your-deepseek-api-key", # Get the DeepSeek API key from https://platform.deepseek.com/api_keys
-    temperature=0.7,
-    top_p=0.4,
-    image_mode="url",
-    detailed_extraction=False, # Set to True for more detailed extraction
-    enable_concurrency=True,
-)
 ```
 
 ## ✅ Supported Models
@@ -184,15 +174,14 @@ This package supports the following Vision LLM models:
 |:------------:|:----------:|
 | gpt-4o | OpenAI |
 | gpt-4o-mini | OpenAI |
-| gemini-1.5-flash | Google |
-| gemini-2.0-flash-exp | Google |
-| gemini-1.5-pro | Google |
+| gemini-2.5-pro | Google |
+| gemini-2.5-flash | Google |
+| gemini-2.0-flash | Google |
 | llava:13b | Ollama |
 | llava:34b | Ollama |
 | llama3.2-vision:11b | Ollama |
 | llama3.2-vision:70b | Ollama |
 | deepseek-r1:32b | Ollama |
-| deepseek-chat | DeepSeek |
 
 ## 🔧 Customization Parameters
 
@@ -203,7 +192,7 @@ Vision Parse offers several customization parameters to enhance document process
 | model_name | Name of the Vision LLM model to use | str |
 | custom_prompt | Define custom prompt for the model and it will be used as a suffix to the default prompt | str |
 | ollama_config | Specify custom configuration for Ollama client initialization | dict |
-| openai_config | Specify custom configuration for OpenAI, Azure OpenAI or DeepSeek client initialization | dict |
+| openai_config | Specify custom configuration for OpenAI or Azure OpenAI client initialization | dict |
 | gemini_config | Specify custom configuration for Gemini client initialization | dict |
 | image_mode | Sets the image output format for the model i.e. if you want image url in markdown content or base64 encoded image | str |
 | detailed_extraction | Enable advanced content extraction to extract complex information such as LaTeX equations, tables, images, etc. | bool |

docs/examples/gemini_demo.ipynb

@@ -25,92 +25,10 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 3,
+   "execution_count": null,
    "metadata": {},
-   "outputs": [
-    {
-     "name": "stderr",
-     "output_type": "stream",
-     "text": [
-      "/Users/arunbrahma/anaconda3/envs/mp/lib/python3.13/site-packages/tqdm/auto.py:21: TqdmWarning: IProgress not found. Please update jupyter and ipywidgets. See https://ipywidgets.readthedocs.io/en/stable/user_install.html\n",
-      "  from .autonotebook import tqdm as notebook_tqdm\n",
-      "Converting pages in PDF file into markdown format: 100%|██████████| 1/1 [00:15<00:00, 15.82s/it]"
-     ]
-    },
-    {
-     "name": "stdout",
-     "output_type": "stream",
-     "text": [
-      "\n",
-      "--- Page 1 ---\n",
-      "# TEXAS A&M\n",
-      "# AGRILIFE\n",
-      "# EXTENSION\n",
-      "\n",
-      "March 3, 2020\n",
-      "\n",
-      "TO: JR Sprague, TCAAA D1 Chairman\n",
-      "\n",
-      "FROM: Andy Holloway, CEA ANR Hemphill County\n",
-      "\n",
-      "Regarding the three chairmanships appointed to me none have any action\n",
-      "needed at this time except for the rules to be submitted to be posted on the\n",
-      "TCAAA website for the Texas Holdem Tournament. Please note the following:\n",
-      "\n",
-      "# Texas Hold'em Basic Poker Rules\n",
-      "\n",
-      "Texas Hold'em is played with a standard deck of 52 cards. The object of the game is to make the best\n",
-      "possible five-card poker hand using any combination of the two card's in a player's hand (known as\n",
-      "the \"hole cards\") and the five community cards that the dealer places in the middle of the table, which\n",
-      "can be used by all players. The standard hand rankings are used, as follows (from best to worst):\n",
-      "\n",
-      "* Straight Flush: Any five cards of the same suit and consecutive rank; the best of these hands,\n",
-      "AKQJT of a single suit, is known as a Royal Flush. For these and regular straights, aces may be\n",
-      "played as high or low cards.\n",
-      "* Four of a Kind: Any four cards of the same rank.\n",
-      "* Full House: Any three cards from a single rank combined with a pair from a different rank (i.e.,\n",
-      "TT55).\n",
-      "* Flush: Any five cards of the same suit.\n",
-      "* Straight: Any five cards of consecutive ranks (i.e., 34567).\n",
-      "* Three of a Kind: Any three cards of the same rank.\n",
-      "* Two Pair: Any two pairs of cards from the same ranks (ie, 5599A).\n",
-      "* One Pair: Any two cards of the same rank.\n",
-      "* High Card: Hands that do not fit any of the above categories are ranked based on the highest card\n",
-      "in their hand (aces are high), then by the second highest card, and so on.\n",
-      "\n",
-      "\n"
-     ]
-    },
-    {
-     "name": "stderr",
-     "output_type": "stream",
-     "text": [
-      "\n"
-     ]
-    }
-   ],
-   "source": [
-    "from vision_parse import VisionParser\n",
-    "import os\n",
-    "\n",
-    "# Initialize parser\n",
-    "parser = VisionParser(\n",
-    "    model_name=\"gemini-1.5-flash\",\n",
-    "    api_key=os.getenv(\"GEMINI_API_KEY\"),\n",
-    "    temperature=0.9,\n",
-    "    top_p=0.4,\n",
-    "    max_output_tokens=2048,\n",
-    "    image_mode=\"url\",\n",
-    "    detailed_extraction=True,\n",
-    ")\n",
-    "\n",
-    "pdf_path = \"../tests/Texas-Holdem-Rules.pdf\"\n",
-    "markdown_pages = parser.convert_pdf(pdf_path)\n",
-    "\n",
-    "# Print the markdown pages\n",
-    "for i, page_content in enumerate(markdown_pages):\n",
-    "    print(f\"\\n--- Page {i+1} ---\\n{page_content}\")"
-   ]
+   "outputs": [],
+   "source": "from vision_parse import VisionParser\nimport os\n\n# Initialize parser\nparser = VisionParser(\n    model_name=\"gemini-2.5-flash\",\n    api_key=os.getenv(\"GEMINI_API_KEY\"),\n    temperature=0.9,\n    top_p=0.4,\n    max_output_tokens=2048,\n    image_mode=\"url\",\n    detailed_extraction=True,\n)\n\npdf_path = \"../tests/Texas-Holdem-Rules.pdf\"\nmarkdown_pages = parser.convert_pdf(pdf_path)\n\n# Print the markdown pages\nfor i, page_content in enumerate(markdown_pages):\n    print(f\"\\n--- Page {i+1} ---\\n{page_content}\")"
   }
  ],
  "metadata": {
@@ -134,4 +52,4 @@
  },
  "nbformat": 4,
  "nbformat_minor": 2
-}
+}

docs/faq.md

@@ -8,7 +8,7 @@
 
     This is a known limitation with locally hosted Ollama models. Here are some solutions:
 
-    - **Use API-based Models**: For better performance, consider using API-based models like OpenAI, DeepSeek, or Gemini, which are significantly faster and more accurate.
+    - **Use API-based Models**: For better performance, consider using API-based models like OpenAI or Gemini, which are significantly faster and more accurate.
     - **Enable Concurrency**: Set `enable_concurrency` to `True` so that multiple pages are processed in parallel, thereby reducing latency. You can also increase the value of `OLLAMA_NUM_PARALLEL` to maximize the number of pages that can be processed in parallel.
     - **Disable Detailed Extraction**: Disable the `detailed_extraction` parameter for simpler PDF documents, which can improve latency.
 

docs/model_config.md

@@ -16,7 +16,7 @@ Pass the following configuration settings to `ollama_config` parameter while ini
 
 For model-specific parameters (like temperature, top_p, etc.), please refer to the [Ollama Model Parameters documentation](https://github.com/ollama/ollama/blob/main/docs/modelfile.md#valid-parameters-and-values). You can pass model-specific parameters as additional kwargs to the `VisionParser` class.
 
-### OpenAI and DeepSeek Clients:
+### OpenAI Client:
 
 Pass the following configuration settings to `openai_config` parameter while initializing the VisionParser class.
 

pyproject.toml

@@ -38,7 +38,8 @@ dependencies = [
     "ollama>=0.4.4",
     "opencv-python>=4.10.0.84",
     "pydantic>=2.0.0",
-    "pymupdf>=1.22.0",
+    "pypdfium2>=4.0.0",
+    "Pillow>=10.4.0",
     "tenacity>=9.0.0",
     "tqdm>=4.65.0",
 ]
@@ -58,13 +59,13 @@ dev = [
     "twine>=6.0.1",
 ]
 gemini = [
-    "google-generativeai==0.8.3",
+    "google-genai>=0.2.0",
 ]
 openai = [
     "openai==1.59.8",
 ]
 all = [
-    "google-generativeai==0.8.3",
+    "google-genai>=0.2.0",
     "openai==1.59.8",
 ]
 

src/vision_parse/constants.py

@@ -2,14 +2,23 @@
 
 SUPPORTED_MODELS: Dict[str, str] = {
     "llama3.2-vision:11b": "ollama",
-    "llama3.2-vision:70b": "ollama",
-    "llava:13b": "ollama",
-    "llava:34b": "ollama",
-    "deepseek-r1:32b": "ollama",
     "gpt-4o": "openai",
     "gpt-4o-mini": "openai",
-    "gemini-1.5-flash": "gemini",
-    "gemini-2.0-flash-exp": "gemini",
-    "gemini-1.5-pro": "gemini",
-    "deepseek-chat": "deepseek",
+    "gemini-2.5-pro": "gemini",
+    "gemini-2.5-flash": "gemini",
+    "gemini-2.0-flash": "gemini",
 }
+
+
+def discover_ollama_vision_models() -> Dict[str, str]:
+    try:
+        import ollama
+        from .model_detector import ModelDetector
+
+        client = ollama.Client()
+        detector = ModelDetector(client)
+        vision_models = detector.get_vision_models()
+
+        return {model: "ollama" for model in vision_models}
+    except Exception:
+        return {}