Merge branch 'main' into devin/1743560186-fix-issue-2511

* ci(workflows): add Python version matrix (3.10-3.12) for tests

* refactor: remove explicit Self import from typing

Python 3.10+ natively supports Self type annotation without explicit imports

* chore: rename external_memory file test

---------

Co-authored-by: Lorenze Jay <63378463+lorenzejay@users.noreply.github.com>

.github/workflows/tests.yml

@@ -12,6 +12,9 @@ jobs:
   tests:
     runs-on: ubuntu-latest
     timeout-minutes: 15
+    strategy:
+      matrix:
+        python-version: ['3.10', '3.11', '3.12']
     steps:
       - name: Checkout code
         uses: actions/checkout@v4
@@ -21,9 +24,8 @@ jobs:
         with:
           enable-cache: true
 
-
-      - name: Set up Python
-        run: uv python install 3.12.8
+      - name: Set up Python ${{ matrix.python-version }}
+        run: uv python install ${{ matrix.python-version }}
 
       - name: Install the project
         run: uv sync --dev --all-extras

src/crewai/cli/constants.py

@@ -123,6 +123,7 @@ MODELS = {
         "claude-3-haiku-20240307",
     ],
     "gemini": [
+        "gemini/gemini-2.5-pro-exp-03-25",  # Experimental release - March 2025
         "gemini/gemini-1.5-flash",
         "gemini/gemini-1.5-pro",
         "gemini/gemini-gemma-2-9b-it",

src/crewai/llm.py

@@ -71,20 +71,26 @@ class FilteredStream:
             return self._original_stream.flush()
 
 
+PRO_CONTEXT_SIZE = 2097152  # 2M tokens
+FLASH_CONTEXT_SIZE = 1048576  # 1M tokens
+GPT4_TURBO_CONTEXT_SIZE = 128000
+CLAUDE_LARGE_CONTEXT_SIZE = 200000
+
 LLM_CONTEXT_WINDOW_SIZES = {
     # openai
     "gpt-4": 8192,
-    "gpt-4o": 128000,
-    "gpt-4o-mini": 128000,
-    "gpt-4-turbo": 128000,
-    "o1-preview": 128000,
-    "o1-mini": 128000,
-    "o3-mini": 200000,  # Based on official o3-mini specifications
+    "gpt-4o": GPT4_TURBO_CONTEXT_SIZE,
+    "gpt-4o-mini": GPT4_TURBO_CONTEXT_SIZE,
+    "gpt-4-turbo": GPT4_TURBO_CONTEXT_SIZE,
+    "o1-preview": GPT4_TURBO_CONTEXT_SIZE,
+    "o1-mini": GPT4_TURBO_CONTEXT_SIZE,
+    "o3-mini": CLAUDE_LARGE_CONTEXT_SIZE,  # Based on official o3-mini specifications
     # gemini
-    "gemini-2.0-flash": 1048576,
-    "gemini-1.5-pro": 2097152,
-    "gemini-1.5-flash": 1048576,
-    "gemini-1.5-flash-8b": 1048576,
+    "gemini-2.5-pro-exp-03-25": PRO_CONTEXT_SIZE,
+    "gemini-2.0-flash": FLASH_CONTEXT_SIZE,
+    "gemini-1.5-pro": PRO_CONTEXT_SIZE,
+    "gemini-1.5-flash": FLASH_CONTEXT_SIZE,
+    "gemini-1.5-flash-8b": FLASH_CONTEXT_SIZE,
     # deepseek
     "deepseek-chat": 128000,
     # groq
@@ -884,10 +890,16 @@ class LLM(BaseLLM):
         Derives the custom_llm_provider from the model string.
         - For example, if the model is "openrouter/deepseek/deepseek-chat", returns "openrouter".
         - If the model is "gemini/gemini-1.5-pro", returns "gemini".
-        - If there is no '/', defaults to "openai".
+        - If the model starts with "gemini-", returns "gemini" only for valid Gemini models.
+        - If there is no '/' or recognized prefix, returns None.
         """
         if "/" in self.model:
             return self.model.split("/")[0]
+        if self.model.startswith("gemini-"):
+            valid_gemini_models = ["gemini-2.5-pro-exp-03-25", "gemini-2.0-flash", 
+                                  "gemini-1.5-pro", "gemini-1.5-flash", "gemini-1.5-flash-8b"]
+            if self.model in valid_gemini_models:
+                return "gemini"
         return None
 
     def _validate_call_params(self) -> None:
@@ -951,9 +963,19 @@ class LLM(BaseLLM):
         self.context_window_size = int(
             DEFAULT_CONTEXT_WINDOW_SIZE * CONTEXT_WINDOW_USAGE_RATIO
         )
-        for key, value in LLM_CONTEXT_WINDOW_SIZES.items():
-            if self.model.startswith(key):
-                self.context_window_size = int(value * CONTEXT_WINDOW_USAGE_RATIO)
+        
+        model_name = self.model
+        if "/" in model_name:
+            model_name = model_name.split("/", 1)[1]
+            
+        if model_name in LLM_CONTEXT_WINDOW_SIZES:
+            self.context_window_size = int(LLM_CONTEXT_WINDOW_SIZES[model_name] * CONTEXT_WINDOW_USAGE_RATIO)
+        else:
+            for key, value in LLM_CONTEXT_WINDOW_SIZES.items():
+                if model_name.startswith(key):
+                    self.context_window_size = int(value * CONTEXT_WINDOW_USAGE_RATIO)
+                    break
+                    
         return self.context_window_size
 
     def set_callbacks(self, callbacks: List[Any]):

src/crewai/memory/external/external_memory.py

@@ -1,4 +1,4 @@
-from typing import TYPE_CHECKING, Any, Dict, Optional, Self
+from typing import TYPE_CHECKING, Any, Dict, Optional
 
 from crewai.memory.external.external_memory_item import ExternalMemoryItem
 from crewai.memory.memory import Memory
@@ -52,7 +52,7 @@ class ExternalMemory(Memory):
     def reset(self) -> None:
         self.storage.reset()
 
-    def set_crew(self, crew: Any) -> Self:
+    def set_crew(self, crew: Any) -> "ExternalMemory":
         super().set_crew(crew)
 
         if not self.storage:

src/crewai/memory/memory.py

@@ -1,4 +1,4 @@
-from typing import Any, Dict, List, Optional, Self
+from typing import Any, Dict, List, Optional
 
 from pydantic import BaseModel
 
@@ -38,6 +38,6 @@ class Memory(BaseModel):
             query=query, limit=limit, score_threshold=score_threshold
         )
 
-    def set_crew(self, crew: Any) -> Self:
+    def set_crew(self, crew: Any) -> "Memory":
         self.crew = crew
         return self

tests/test_gemini_2_5_support.py

@@ -0,0 +1,41 @@
+import pytest
+
+from crewai.llm import CONTEXT_WINDOW_USAGE_RATIO, LLM, PRO_CONTEXT_SIZE
+
+
+def test_get_custom_llm_provider_gemini_2_5():
+    """Test that the Gemini 2.5 model is correctly identified as a Gemini provider."""
+    llm = LLM(model="gemini/gemini-2.5-pro-exp-03-25")
+    assert llm._get_custom_llm_provider() == "gemini"
+
+def test_gemini_2_5_context_window_size():
+    """Test that the Gemini 2.5 model has the correct context window size."""
+    llm = LLM(model="gemini-2.5-pro-exp-03-25")
+    expected_size = int(PRO_CONTEXT_SIZE * CONTEXT_WINDOW_USAGE_RATIO)
+    assert llm.get_context_window_size() == expected_size
+
+def test_gemini_2_5_invalid_model_name():
+    """Test handling of invalid model name variations."""
+    llm = LLM(model="gemini-2.5-wrong")
+    assert llm._get_custom_llm_provider() != "gemini"
+
+def test_gemini_2_5_model_parameters():
+    """Test model initialization with various parameters."""
+    llm = LLM(
+        model="gemini/gemini-2.5-pro-exp-03-25", 
+        temperature=0.7, 
+        max_tokens=1000
+    )
+    assert llm.model == "gemini/gemini-2.5-pro-exp-03-25"
+    assert llm.temperature == 0.7
+    assert llm.max_tokens == 1000
+
+def test_gemini_2_5_with_and_without_prefix():
+    """Test that the model works with and without the 'gemini/' prefix."""
+    llm_with_prefix = LLM(model="gemini/gemini-2.5-pro-exp-03-25")
+    llm_without_prefix = LLM(model="gemini-2.5-pro-exp-03-25")
+    
+    assert llm_with_prefix._get_custom_llm_provider() == "gemini"
+    assert llm_without_prefix._get_custom_llm_provider() == "gemini"
+    
+    assert llm_with_prefix.get_context_window_size() == llm_without_prefix.get_context_window_size()