fix(models): add Gemini 2.5 token limits so they are not truncated to 8192

Author: ahmad-nexarapp

scrapegraphai/helpers/models_tokens.py

@@ -140,6 +140,11 @@
         "gemini-2.0-flash-latest": 1000000,
         "gemini-2.0-flash-exp": 1000000,
         "gemini-2.0-pro-exp": 2000000,
+        "gemini-2.5-flash": 1000000,
+        "gemini-2.5-flash-latest": 1000000,
+        "gemini-2.5-flash-lite": 1000000,
+        "gemini-2.5-pro": 1000000,
+        "gemini-flash-latest": 1000000,
         "models/embedding-001": 2048,
     },
     "google_vertexai": {
@@ -150,6 +155,10 @@
         "gemini-2.0-flash-exp": 1048576,
         "gemini-2.0-pro": 2000000,
         "gemini-2.0-pro-exp": 2000000,
+        "gemini-2.5-flash": 1048576,
+        "gemini-2.5-flash-lite": 1048576,
+        "gemini-2.5-pro": 1048576,
+        "gemini-flash-latest": 1048576,
     },
     "ollama": {
         "command-r": 12800,

tests/test_models_tokens.py

@@ -7,9 +7,9 @@ class TestModelsTokens:
     def test_openai_tokens(self):
         """Test that the 'openai' provider exists and its tokens are valid positive integers."""
         openai_models = models_tokens.get("openai")
-        assert openai_models is not None, (
-            "'openai' key should be present in models_tokens"
-        )
+        assert (
+            openai_models is not None
+        ), "'openai' key should be present in models_tokens"
         for model, token in openai_models.items():
             assert isinstance(model, str), "Model name should be a string"
             assert isinstance(token, int), "Token limit should be an integer"
@@ -30,19 +30,50 @@ def test_google_providers(self):
         assert google_genai is not None, "'google_genai' key should be present"
         assert google_vertexai is not None, "'google_vertexai' key should be present"
         # Check a specific key from google_genai
-        assert "gemini-pro" in google_genai, (
-            "'gemini-pro' should be in google_genai models"
-        )
+        assert (
+            "gemini-pro" in google_genai
+        ), "'gemini-pro' should be in google_genai models"
         # Validate token values types
         for provider in [google_genai, google_vertexai]:
             for token in provider.values():
                 assert isinstance(token, int), "Token limit must be an integer"
 
+    def test_gemini_2_5_models_are_registered(self):
+        """Gemini 2.5 / flash-latest must be in the table so they are not truncated to 8192.
+
+        #1121: an unknown model silently falls back to an 8192-token window.
+        google_genai/gemini-2.5-flash is the reported case; gemini-flash-latest
+        is the current flash alias. Both have a 1M input context.
+        """
+        google_genai = models_tokens["google_genai"]
+        google_vertexai = models_tokens["google_vertexai"]
+
+        for model in (
+            "gemini-2.5-flash",
+            "gemini-2.5-flash-latest",
+            "gemini-2.5-flash-lite",
+            "gemini-2.5-pro",
+            "gemini-flash-latest",
+        ):
+            assert (
+                google_genai.get(model) == 1000000
+            ), f"Expected 1M context for {model} in google_genai"
+
+        for model in (
+            "gemini-2.5-flash",
+            "gemini-2.5-flash-lite",
+            "gemini-2.5-pro",
+            "gemini-flash-latest",
+        ):
+            assert (
+                google_vertexai.get(model) == 1048576
+            ), f"Expected 1M context for {model} in google_vertexai"
+
     def test_non_existent_provider(self):
         """Test that a non-existent provider returns None."""
-        assert models_tokens.get("non_existent") is None, (
-            "Non-existent provider should return None"
-        )
+        assert (
+            models_tokens.get("non_existent") is None
+        ), "Non-existent provider should return None"
 
     def test_total_model_keys(self):
         """Test that the total number of models across all providers is above an expected count."""
@@ -59,136 +90,136 @@ def test_non_empty_model_keys(self):
         """Ensure that model token names are non-empty strings."""
         for provider, model_dict in models_tokens.items():
             for model in model_dict.keys():
-                assert model != "", (
-                    f"Model name in provider '{provider}' should not be empty."
-                )
+                assert (
+                    model != ""
+                ), f"Model name in provider '{provider}' should not be empty."
 
     def test_token_limits_range(self):
         """Test that token limits for all models fall within a plausible range (e.g., 1 to 300000)."""
         for provider, model_dict in models_tokens.items():
             for model, token in model_dict.items():
-                assert 1 <= token <= 1100000, (
-                    f"Token limit for {model} in provider {provider} is out of plausible range."
-                )
+                assert (
+                    1 <= token <= 1100000
+                ), f"Token limit for {model} in provider {provider} is out of plausible range."
 
     def test_provider_structure(self):
         """Test that every provider in models_tokens has a dictionary as its value."""
         for provider, models in models_tokens.items():
-            assert isinstance(models, dict), (
-                f"Provider {provider} should map to a dictionary, got {type(models).__name__}"
-            )
+            assert isinstance(
+                models, dict
+            ), f"Provider {provider} should map to a dictionary, got {type(models).__name__}"
 
     def test_non_empty_provider(self):
         """Test that each provider dictionary is not empty."""
         for provider, models in models_tokens.items():
-            assert len(models) > 0, (
-                f"Provider {provider} should contain at least one model."
-            )
+            assert (
+                len(models) > 0
+            ), f"Provider {provider} should contain at least one model."
 
     def test_specific_model_token_values(self):
         """Test specific expected token values for selected models from various providers."""
         # Verify a token for a selected model from the 'openai' provider
         openai = models_tokens.get("openai")
-        assert openai.get("gpt-3.5-turbo-0125") == 16385, (
-            "Expected token limit for gpt-3.5-turbo-0125 in openai to be 16385"
-        )
+        assert (
+            openai.get("gpt-3.5-turbo-0125") == 16385
+        ), "Expected token limit for gpt-3.5-turbo-0125 in openai to be 16385"
 
         # Verify a token for a selected model from the 'azure_openai' provider
         azure = models_tokens.get("azure_openai")
-        assert azure.get("gpt-3.5") == 4096, (
-            "Expected token limit for gpt-3.5 in azure_openai to be 4096"
-        )
+        assert (
+            azure.get("gpt-3.5") == 4096
+        ), "Expected token limit for gpt-3.5 in azure_openai to be 4096"
 
         # Verify a token for a selected model from the 'anthropic' provider
         anthropic = models_tokens.get("anthropic")
-        assert anthropic.get("claude_instant") == 100000, (
-            "Expected token limit for claude_instant in anthropic to be 100000"
-        )
+        assert (
+            anthropic.get("claude_instant") == 100000
+        ), "Expected token limit for claude_instant in anthropic to be 100000"
 
     def test_providers_count(self):
         """Test that the total number of providers is as expected (at least 15)."""
-        assert len(models_tokens) >= 15, (
-            "Expected at least 15 providers in models_tokens"
-        )
+        assert (
+            len(models_tokens) >= 15
+        ), "Expected at least 15 providers in models_tokens"
 
     def test_non_existent_model(self):
         """Test that a non-existent model within a valid provider returns None."""
         openai = models_tokens.get("openai")
-        assert openai.get("non_existent_model") is None, (
-            "Non-existent model should return None from a valid provider."
-        )
+        assert (
+            openai.get("non_existent_model") is None
+        ), "Non-existent model should return None from a valid provider."
 
     def test_no_whitespace_in_model_names(self):
         """Test that model names do not contain leading or trailing whitespace."""
         for provider, model_dict in models_tokens.items():
             for model in model_dict.keys():
                 # Assert that stripping whitespace does not change the model name
-                assert model == model.strip(), (
-                    f"Model name '{model}' in provider '{provider}' contains leading or trailing whitespace."
-                )
+                assert (
+                    model == model.strip()
+                ), f"Model name '{model}' in provider '{provider}' contains leading or trailing whitespace."
 
     def test_specific_models_additional(self):
         """Test specific token values for additional models across various providers."""
         # Check some models in the 'ollama' provider
         ollama = models_tokens.get("ollama")
-        assert ollama.get("llama2") == 4096, (
-            "Expected token limit for 'llama2' in ollama to be 4096"
-        )
-        assert ollama.get("llama2:70b") == 4096, (
-            "Expected token limit for 'llama2:70b' in ollama to be 4096"
-        )
+        assert (
+            ollama.get("llama2") == 4096
+        ), "Expected token limit for 'llama2' in ollama to be 4096"
+        assert (
+            ollama.get("llama2:70b") == 4096
+        ), "Expected token limit for 'llama2:70b' in ollama to be 4096"
 
         # Check a specific model from the 'mistralai' provider
         mistralai = models_tokens.get("mistralai")
-        assert mistralai.get("open-codestral-mamba") == 256000, (
-            "Expected token limit for 'open-codestral-mamba' in mistralai to be 256000"
-        )
+        assert (
+            mistralai.get("open-codestral-mamba") == 256000
+        ), "Expected token limit for 'open-codestral-mamba' in mistralai to be 256000"
 
         # Check a specific model from the 'deepseek' provider
         deepseek = models_tokens.get("deepseek")
-        assert deepseek.get("deepseek-chat") == 28672, (
-            "Expected token limit for 'deepseek-chat' in deepseek to be 28672"
-        )
+        assert (
+            deepseek.get("deepseek-chat") == 28672
+        ), "Expected token limit for 'deepseek-chat' in deepseek to be 28672"
 
         # Check a model from the 'ernie' provider
         ernie = models_tokens.get("ernie")
-        assert ernie.get("ernie-bot") == 4096, (
-            "Expected token limit for 'ernie-bot' in ernie to be 4096"
-        )
+        assert (
+            ernie.get("ernie-bot") == 4096
+        ), "Expected token limit for 'ernie-bot' in ernie to be 4096"
 
     def test_nvidia_specific(self):
         """Test specific token value for 'meta/codellama-70b' in the nvidia provider."""
         nvidia = models_tokens.get("nvidia")
         assert nvidia is not None, "'nvidia' provider should exist"
         # Verify token for 'meta/codellama-70b' equals 16384 as defined in the nvidia dictionary
-        assert nvidia.get("meta/codellama-70b") == 16384, (
-            "Expected token limit for 'meta/codellama-70b' in nvidia to be 16384"
-        )
+        assert (
+            nvidia.get("meta/codellama-70b") == 16384
+        ), "Expected token limit for 'meta/codellama-70b' in nvidia to be 16384"
 
     def test_groq_specific(self):
         """Test specific token value for 'claude-3-haiku-20240307\'' in the groq provider."""
         groq = models_tokens.get("groq")
         assert groq is not None, "'groq' provider should exist"
         # Note: The model name has an embedded apostrophe at the end in its name.
-        assert groq.get("claude-3-haiku-20240307'") == 8192, (
-            "Expected token limit for 'claude-3-haiku-20240307\\'' in groq to be 8192"
-        )
+        assert (
+            groq.get("claude-3-haiku-20240307'") == 8192
+        ), "Expected token limit for 'claude-3-haiku-20240307\\'' in groq to be 8192"
 
     def test_togetherai_specific(self):
         """Test specific token value for 'meta-llama/Meta-Llama-3.1-70B-Instruct-Turbo' in the toghetherai provider."""
         togetherai = models_tokens.get("toghetherai")
         assert togetherai is not None, "'toghetherai' provider should exist"
         expected = 128000
         model_name = "meta-llama/Meta-Llama-3.1-70B-Instruct-Turbo"
-        assert togetherai.get(model_name) == expected, (
-            f"Expected token limit for '{model_name}' in toghetherai to be {expected}"
-        )
+        assert (
+            togetherai.get(model_name) == expected
+        ), f"Expected token limit for '{model_name}' in toghetherai to be {expected}"
 
     def test_ernie_all_values(self):
         """Test that all models in the 'ernie' provider have token values exactly 4096."""
         ernie = models_tokens.get("ernie")
         assert ernie is not None, "'ernie' provider should exist"
         for model, token in ernie.items():
-            assert token == 4096, (
-                f"Expected token limit for '{model}' in ernie to be 4096, got {token}"
-            )
+            assert (
+                token == 4096
+            ), f"Expected token limit for '{model}' in ernie to be 4096, got {token}"