feat: Update Gemini model references and add rate limiting

README.md

@@ -23,7 +23,7 @@ See [below](#benchmarks) for detailed speed and accuracy benchmarks, and instruc
 
 ## Hybrid Mode
 
-For the highest accuracy, pass the `--use_llm` flag to use an LLM alongside marker.  This will do things like merge tables across pages, handle inline math, format tables properly, and extract values from forms.  It can use any gemini or ollama model.  By default, it uses `gemini-2.0-flash`.  See [below](#llm-services) for details.
+For the highest accuracy, pass the `--use_llm` flag to use an LLM alongside marker.  This will do things like merge tables across pages, handle inline math, format tables properly, and extract values from forms.  It can use any gemini or ollama model.  By default, it uses `gemini-2.5-flash`.  See [below](#llm-services) for details.
 
 Here is a table benchmark comparing marker, gemini flash alone, and marker with use_llm:
 
@@ -534,7 +534,7 @@ python benchmarks/table/table.py --max_rows 100
 Options:
 
 - `--use_llm` uses an llm with marker to improve accuracy.
-- `--use_gemini` also benchmarks gemini 2.0 flash.
+- `--use_gemini` also benchmarks gemini 2.5 flash.
 
 # How it works
 

benchmarks/overall/elo.py

@@ -143,7 +143,7 @@ def llm_response_wrapper(
         )
         try:
             responses = client.models.generate_content(
-                model="gemini-2.0-flash-001",
+                model="gemini-2.5-flash",
                 contents=prompt,
                 config={
                     "temperature": 0,

benchmarks/overall/scorers/llm.py

@@ -142,7 +142,7 @@ def llm_response_wrapper(self, prompt, response_schema, depth=0):
         )
         try:
             responses = client.models.generate_content(
-                model="gemini-2.0-flash-001",
+                model="gemini-2.5-flash",
                 contents=prompt,
                 config={
                     "temperature": 0,

benchmarks/table/gemini.py

@@ -35,7 +35,7 @@ def gemini_table_rec(image: Image.Image):
     image.save(image_bytes, format="PNG")
 
     responses = client.models.generate_content(
-        model="gemini-2.0-flash",
+        model="gemini-2.5-flash",
         contents=[types.Part.from_bytes(data=image_bytes.getvalue(), mime_type="image/png"), prompt],  # According to gemini docs, it performs better if the image is the first element
         config={
             "temperature": 0,

marker/config/parser.py

@@ -13,6 +13,7 @@
 from marker.renderers.markdown import MarkdownRenderer
 from marker.settings import settings
 from marker.util import classes_to_strings, parse_range_str, strings_to_classes
+from marker.services.gemini import GeminiModel
 
 logger = get_logger()
 
@@ -82,6 +83,12 @@ def common_options(fn):
             default=None,
             help="LLM service to use - should be full import path, like marker.services.gemini.GoogleGeminiService",
         )(fn)
+        fn = click.option(
+            "--gemini_model_name",
+            type=str,
+            default=None,
+            help="Name of the Gemini model to use (e.g., gemini-2.5-flash)",
+        )(fn)
         return fn
 
     def generate_config_dict(self) -> Dict[str, any]:
@@ -114,6 +121,13 @@ def generate_config_dict(self) -> Dict[str, any]:
         if settings.GOOGLE_API_KEY:
             config["gemini_api_key"] = settings.GOOGLE_API_KEY
 
+        if self.cli_options.get("gemini_model_name"):
+            try:
+                config["gemini_model_name"] = GeminiModel(self.cli_options["gemini_model_name"])
+            except ValueError:
+                logger.warning(f"Invalid gemini_model_name: {self.cli_options["gemini_model_name"]}. Using default.")
+                config["gemini_model_name"] = GeminiModel.DEFAULT
+
         return config
 
     def get_llm_service(self):

marker/services/gemini.py

@@ -1,7 +1,10 @@
 import json
 import time
 import traceback
+from collections import deque
+from enum import Enum
 from io import BytesIO
+from threading import Lock
 from typing import List, Annotated
 
 import PIL
@@ -17,10 +20,28 @@
 logger = get_logger()
 
 
+class GeminiModel(str, Enum):
+    GEMINI_2_5_PRO = "gemini-2.5-pro"
+    GEMINI_2_5_FLASH = "gemini-2.5-flash"
+    DEFAULT = "gemini-2.5-flash"
+
+
+# Rate limiting settings
+MODEL_LIMITS = {
+    GeminiModel.GEMINI_2_5_PRO: {"rpm": 5},
+    GeminiModel.GEMINI_2_5_FLASH: {"rpm": 10},
+    GeminiModel.DEFAULT: {"rpm": 10},  # Corresponds to gemini-2.5-flash
+}
+
+# Global request tracker and lock
+REQUEST_TIMESTAMPS = {model: deque() for model in GeminiModel}
+RATE_LIMIT_LOCK = Lock()
+
+
 class BaseGeminiService(BaseService):
     gemini_model_name: Annotated[
-        str, "The name of the Google model to use for the service."
-    ] = "gemini-2.0-flash"
+        GeminiModel, "The name of the Google model to use for the service."
+    ] = GeminiModel.DEFAULT
 
     def img_to_bytes(self, img: PIL.Image.Image):
         image_bytes = BytesIO()
@@ -52,6 +73,28 @@ def __call__(
         if timeout is None:
             timeout = self.timeout
 
+        # Proactive rate limiting
+        with RATE_LIMIT_LOCK:
+            model_name = self.gemini_model_name
+            rpm_limit = MODEL_LIMITS.get(model_name, {"rpm": 10})["rpm"]
+            request_history = REQUEST_TIMESTAMPS[model_name]
+
+            current_time = time.time()
+            # Remove timestamps older than 60 seconds
+            while request_history and current_time - request_history[0] > 60:
+                request_history.popleft()
+
+            if len(request_history) >= rpm_limit:
+                wait_time = 60 - (current_time - request_history[0])
+                if wait_time > 0:
+                    logger.warning(
+                        f"RPM limit for {model_name} reached. Waiting for {wait_time:.2f} seconds."
+                    )
+                    time.sleep(wait_time)
+
+            # Record the new request timestamp
+            request_history.append(time.time())
+
         client = self.get_google_client(timeout=timeout)
         image_parts = self.format_image_for_llm(image)
 

marker/services/vertex.py

@@ -16,7 +16,7 @@ class GoogleVertexService(BaseGeminiService):
     gemini_model_name: Annotated[
         str,
         "The name of the Google model to use for the service."
-    ] = "gemini-2.0-flash-001"
+    ] = "gemini-2.5-flash"
     vertex_dedicated: Annotated[
         bool,
         "Whether to use a dedicated Vertex AI instance."