synth-ai 0.1.0.dev28__py3-none-any.whl → 0.1.0.dev30__py3-none-any.whl

public_tests/test_agent.py

@@ -311,7 +311,7 @@ You will be given a code_prompt_for_answer, which contains imports and the funct
 
 Your next actions / thought:  
 
-Structured output: reasoning="I need to implement the function 'task_func' that replaces values in a DataFrame based on a dictionary and calculates the Pearson correlation coefficient between the columns. I will also ensure to handle the case where the input is not a DataFrame by raising a ValueError." action_name='edit_submission' action_args=[ActionArgument(key='first_line', value=4), ActionArgument(key='last_line', value=4), ActionArgument(key='new_code', value="    if not isinstance(df, pd.DataFrame):\n        raise ValueError('Input must be a DataFrame')\n    df.replace(dct, inplace=True)\n    return df.corr(method='pearson')")]
+Structured output: reasoning="I need to implement the function 'task_func' that replaces values in a DataFrame based on a dictionary and calculates the Pearson correlation coefficient between each pair of columns. I will also ensure to handle the case where the input is not a DataFrame by raising a ValueError." action_name='edit_submission' action_args=[ActionArgument(key='first_line', value=4), ActionArgument(key='last_line', value=4), ActionArgument(key='new_code', value="    if not isinstance(df, pd.DataFrame):\n        raise ValueError('Input must be a DataFrame')\n    df.replace(dct, inplace=True)\n    return df.corr(method='pearson')")]
 <System Message> 
 # Premise
 You are a software engineer
@@ -489,11 +489,11 @@ class TestLMStructuredOutputs(unittest.TestCase):
                     user_message=user_message,
                     response_model=ReAct,
                 )
-                self.assertIsInstance(result, ReAct)
-                self.assertIsInstance(result.reasoning, str)
-                self.assertIsInstance(result.action_name, str)
-                self.assertIsInstance(result.action_args, list)
-                for arg in result.action_args:
+                self.assertIsInstance(result.structured_output, ReAct)
+                self.assertIsInstance(result.structured_output.reasoning, str)
+                self.assertIsInstance(result.structured_output.action_name, str)
+                self.assertIsInstance(result.structured_output.action_args, list)
+                for arg in result.structured_output.action_args:
                     self.assertIsInstance(arg, ActionArgument)
                     self.assertIsInstance(arg.key, str)
                     # self.assertIsInstance(arg.value, str)
@@ -512,11 +512,11 @@ class TestLMStructuredOutputs(unittest.TestCase):
                     user_message=user_message,
                     response_model=ReAct,
                 )
-                self.assertIsInstance(result, ReAct)
-                self.assertIsInstance(result.reasoning, str)
-                self.assertIsInstance(result.action_name, str)
-                self.assertIsInstance(result.action_args, list)
-                for arg in result.action_args:
+                self.assertIsInstance(result.structured_output, ReAct)
+                self.assertIsInstance(result.structured_output.reasoning, str)
+                self.assertIsInstance(result.structured_output.action_name, str)
+                self.assertIsInstance(result.structured_output.action_args, list)
+                for arg in result.structured_output.action_args:
                     self.assertIsInstance(arg, ActionArgument)
                     self.assertIsInstance(arg.key, str)
                     # self.assertIsInstance(arg.value, str)

public_tests/test_all_structured_outputs.py

@@ -3,8 +3,7 @@ from typing import Any, Dict, Optional
 import pytest
 from pydantic import BaseModel
 
-from synth_ai.zyk.lms.core.main import LM
-
+from synth_ai.zyk import LM, BaseLMResponse
 
 class StateUpdate(BaseModel):
     """Response model for state updates from LLM"""
@@ -117,8 +116,8 @@ def current_state():
 @pytest.mark.parametrize(
     "model_name",
     [
-        "gpt-4o-mini",
-        "gemini-1.5-flash",
+       "gpt-4o-mini",
+       "gemini-1.5-flash",
         "claude-3-haiku-20240307",
         "deepseek-chat",
         "llama-3.1-8b-instant",
@@ -132,26 +131,27 @@ def test_state_delta_handling(
     state_delta_instructions = """Update the final_results to include findings about code quality issues. Add a recommendation to improve error handling."""
     user_message = f"Current state: {current_state}\nState delta instructions: {state_delta_instructions}\n\nHow should the state be updated?"
 
-    try:
-        result = models[model_name].respond_sync(
-            system_message=system_message,
-            user_message=user_message,
-            response_model=StateUpdate,
-        )
-
-        # Verify response structure
-        assert isinstance(result, StateUpdate)
-
-        # Verify only allowed fields are present and have correct types
-        if result.short_term_plan is not None:
-            assert isinstance(result.short_term_plan, str)
-        if result.objective is not None:
-            assert isinstance(result.objective, str)
-        if result.final_results is not None:
-            assert isinstance(result.final_results, dict)
-
-    except Exception as e:
-        pytest.fail(f"Model {model_name} failed: {str(e)}")
+    #try:
+    result: BaseLMResponse = models[model_name].respond_sync(
+        system_message=system_message,
+        user_message=user_message,
+        response_model=StateUpdate,
+    )
+    print("Result", result)
+    # Verify response structure
+    assert isinstance(result, BaseLMResponse)
+    assert isinstance(result.structured_output, StateUpdate)
+
+    # Verify only allowed fields are present and have correct types
+    if result.structured_output.short_term_plan is not None:
+        assert isinstance(result.structured_output.short_term_plan, str)
+    if result.structured_output.objective is not None:
+        assert isinstance(result.structured_output.objective, str)
+    if result.structured_output.final_results is not None:
+        assert isinstance(result.structured_output.final_results, dict)
+
+    # except Exception as e:
+    #     pytest.fail(f"Model {model_name} failed: {str(e)}")
 
 
 @pytest.mark.timeout(15)
@@ -186,16 +186,11 @@ def test_state_delta_protected_fields(
     state_delta_instructions = """Update the message history to include new findings and update step summaries with recent progress."""
     user_message = f"Current state: {current_state}\nState delta instructions: {state_delta_instructions}\n\nHow should the state be updated?"
 
-    try:
-        result = models[model_name].respond_sync(
-            system_message=system_message,
-            user_message=user_message,
-            response_model=StateUpdate,
-        )
-
-        # Verify no protected fields are present
-        assert not hasattr(result, "message_history")
-        assert not hasattr(result, "step_summaries")
-
-    except Exception as e:
-        pytest.fail(f"Model {model_name} failed: {str(e)}")
+    #try:
+    result = models[model_name].respond_sync(
+        system_message=system_message,
+        user_message=user_message,
+        response_model=StateUpdate,
+    )
+    # except Exception as e:
+    #     pytest.fail(f"Model {model_name} failed: {str(e)}")

public_tests/test_deepseek_tools.py

@@ -0,0 +1,64 @@
+from pydantic import BaseModel
+
+from synth_ai.zyk.lms.core.main import LM
+from synth_ai.zyk.lms.tools.base import BaseTool
+from synth_ai.zyk.lms.vendors.supported.deepseek import DeepSeekAPI
+
+
+class WeatherParams(BaseModel):
+    location: str
+
+
+weather_tool = BaseTool(
+    name="get_weather",
+    description="Get current temperature for a given location.",
+    arguments=WeatherParams,
+)
+
+
+def test_weather_tool_direct():
+    client = DeepSeekAPI()
+
+    response = client._hit_api_sync(
+        model="deepseek-chat",
+        messages=[
+            {
+                "role": "system",
+                "content": "You are a helpful assistant that uses tools when appropriate.",
+            },
+            {
+                "role": "user",
+                "content": "What's the weather in Paris? Use the tools and explain your reasoning.",
+            },
+        ],
+        tools=[weather_tool],
+        lm_config={
+            "temperature": 0,
+        },
+    )
+
+    # Check that we got a tool call
+    assert response.tool_calls is not None
+    assert len(response.tool_calls) == 1
+    assert response.tool_calls[0]["function"]["name"] == "get_weather"
+    assert "Paris" in response.tool_calls[0]["function"]["arguments"]
+
+
+def test_weather_tool_lm():
+    lm = LM(
+        model_name="deepseek-chat",
+        formatting_model_name="deepseek-chat",
+        temperature=0,
+    )
+
+    response = lm.respond_sync(
+        system_message="You are a helpful assistant that uses tools when appropriate.",
+        user_message="What's the weather in Paris? Use the tools and explain your reasoning.",
+        tools=[weather_tool],
+    )
+
+    # Check that we got a tool call
+    assert response.tool_calls is not None
+    assert len(response.tool_calls) == 1
+    assert response.tool_calls[0]["function"]["name"] == "get_weather"
+    assert "Paris" in response.tool_calls[0]["function"]["arguments"]

public_tests/test_gemini_structured_outputs.py

@@ -0,0 +1,106 @@
+import asyncio
+import unittest
+from typing import List
+
+from pydantic import BaseModel, Field
+
+from synth_ai.zyk.lms.core.main import LM
+
+
+# Define example structured output models
+class SimpleResponse(BaseModel):
+    message: str
+    confidence_between_zero_one: float = Field(
+        ..., description="Confidence level between 0 and 1"
+    )
+
+
+class ComplexResponse(BaseModel):
+    title: str
+    tags: List[str]
+    content: str
+
+
+class NestedResponse(BaseModel):
+    main_category: str
+    subcategories: List[str]
+    details: SimpleResponse
+
+
+class TestLMStructuredOutputs(unittest.TestCase):
+    @classmethod
+    def setUpClass(cls):
+        # Initialize LMs for both forced_json and stringified_json modes
+        cls.lm_forced_json = LM(
+            model_name="gpt-4o-mini",
+            formatting_model_name="gpt-4o-mini",
+            temperature=0.7,
+            max_retries="Few",
+            structured_output_mode="forced_json",
+        )
+        cls.lm_stringified_json = LM(
+            model_name="gemma3-27b-it",
+            formatting_model_name="gpt-4o-mini",
+            temperature=0.7,
+            max_retries="Few",
+            structured_output_mode="stringified_json",
+        )
+
+    def test_sync_simple_response(self):
+        for lm in [self.lm_forced_json, self.lm_stringified_json]:
+            with self.subTest(
+                mode=lm.structured_output_handler.handler.structured_output_mode
+            ):
+                result = lm.respond_sync(
+                    system_message="You are a helpful assistant.",
+                    user_message="Give me a short greeting and your confidence level.",
+                    response_model=SimpleResponse,
+                )
+                self.assertIsInstance(result.structured_output, SimpleResponse)
+                self.assertIsInstance(result.structured_output.message, str)
+                self.assertIsInstance(
+                    result.structured_output.confidence_between_zero_one, float
+                )
+                self.assertGreaterEqual(
+                    result.structured_output.confidence_between_zero_one, 0
+                )
+                self.assertLessEqual(
+                    result.structured_output.confidence_between_zero_one, 1
+                )
+
+    def test_sync_complex_response(self):
+        for lm in [self.lm_forced_json, self.lm_stringified_json]:
+            with self.subTest(
+                mode=lm.structured_output_handler.handler.structured_output_mode
+            ):
+                result = lm.respond_sync(
+                    system_message="You are a content creator.",
+                    user_message="Create a short blog post about AI.",
+                    response_model=ComplexResponse,
+                )
+                self.assertIsInstance(result.structured_output, ComplexResponse)
+                self.assertIsInstance(result.structured_output.title, str)
+                self.assertIsInstance(result.structured_output.tags, list)
+                self.assertIsInstance(result.structured_output.content, str)
+
+    async def async_nested_response(self, lm):
+        result = await lm.respond_async(
+            system_message="You are a categorization expert.",
+            user_message="Categorize 'Python' and provide a brief description.",
+            response_model=NestedResponse,
+        )
+        self.assertIsInstance(result.structured_output, NestedResponse)
+        self.assertIsInstance(result.structured_output.main_category, str)
+        self.assertIsInstance(result.structured_output.subcategories, list)
+        self.assertIsInstance(result.structured_output.details, SimpleResponse)
+
+    def test_async_nested_response(self):
+        for lm in [self.lm_forced_json, self.lm_stringified_json]:  #
+            with self.subTest(
+                mode=lm.structured_output_handler.handler.structured_output_mode
+            ):
+                asyncio.run(self.async_nested_response(lm))
+
+
+if __name__ == "__main__":
+    unittest.main()

public_tests/test_models.py

@@ -1,4 +1,3 @@
-import asyncio
 import time
 
 import pytest
@@ -57,7 +56,9 @@ def model_instances():
 
     # Set reasoning_effort to "high" for specific models
     models["o3-mini-high-reasoning"].lm_config["reasoning_effort"] = "high"
-    models["claude-3-7-sonnet-latest-high-reasoning"].lm_config["reasoning_effort"] = "high"
+    models["claude-3-7-sonnet-latest-high-reasoning"].lm_config["reasoning_effort"] = (
+        "high"
+    )
 
     return models
 
@@ -91,28 +92,27 @@ def test_model_simple_response(model_instances, model_name):
     elapsed = time.time() - start_time
 
     print(f"Response time: {elapsed:.2f} seconds")
-    print(f"Response length: {len(response)} characters")
-    print(f"Response sample: {response[:100]}...")
+    print(f"Response length: {len(response.raw_response)} characters")
+    print(f"Response sample: {response.raw_response[:100]}...")
 
     # Basic validation
-    assert isinstance(response, str)
-    assert len(response) > 0
+    assert isinstance(response.raw_response, str)
+    assert len(response.raw_response) > 0
     assert (
-        "Paris" in response
-    ), f"Expected 'Paris' in response, but got: {response[:200]}..."
+        "Paris" in response.raw_response
+    ), f"Expected 'Paris' in response, but got: {response.raw_response[:200]}..."
 
 
 @pytest.mark.asyncio
 @pytest.mark.parametrize(
     "model_name",
     [
-        "o3-mini",
-        "o3-mini-high-reasoning",
-        "claude-3-7-sonnet-latest",
+   #     "o3-mini",
+ #       "claude-3-7-sonnet-latest",
         "claude-3-7-sonnet-latest-high-reasoning",
-        "gemini-2-flash",
-        "gemma3-27b-it",
-        "gpt-4o-mini",
+        # "gemini-2-flash",
+        # "gemma3-27b-it",
+        # "gpt-4o-mini",
     ],
 )
 async def test_reasoning_question(model_instances, model_name):
@@ -131,24 +131,24 @@ async def test_reasoning_question(model_instances, model_name):
     elapsed = time.time() - start_time
 
     print(f"Response time: {elapsed:.2f} seconds")
-    print(f"Response length: {len(response)} characters")
-    print(f"Response sample: {response[:100]}...")
+    print(f"Response length: {len(response.raw_response)} characters")
+    print(f"Response sample: {response.raw_response[:100]}...")
 
     # Basic validation
-    assert isinstance(response, str)
-    assert len(response) > 0
+    assert isinstance(response.raw_response, str)
+    assert len(response.raw_response) > 0
 
 
 @pytest.mark.parametrize(
     "model_name",
     [
         "o3-mini",
-        "o3-mini-high-reasoning",
-        "claude-3-7-sonnet-latest",
+       # "o3-mini",
+        #"claude-3-7-sonnet-latest",
         "claude-3-7-sonnet-latest-high-reasoning",
-        "gemini-2-flash",
-        "gemma3-27b-it",
-        "gpt-4o-mini",
+        # "gemini-2-flash",
+        # "gemma3-27b-it",
+        # "gpt-4o-mini",
     ],
 )
 def test_model_context_and_factuality(model_instances, model_name):
@@ -171,11 +171,11 @@ def test_model_context_and_factuality(model_instances, model_name):
 
     # Check if the response contains the correct information
     assert (
-        "1968" in response
-    ), f"Expected '1968' in response for founding year, but got: {response[:200]}..."
+        "1968" in response.raw_response
+    ), f"Expected '1968' in response for founding year, but got: {response.raw_response[:200]}..."
     assert (
-        "Robert Neptune" in response
-    ), f"Expected 'Robert Neptune' in response for mayor, but got: {response[:200]}..."
+        "Robert Neptune" in response.raw_response
+    ), f"Expected 'Robert Neptune' in response for mayor, but got: {response.raw_response[:200]}..."
 
 
 if __name__ == "__main__":

public_tests/test_openai_structured_outputs.py

@@ -0,0 +1,106 @@
+import asyncio
+import unittest
+from typing import List
+
+from pydantic import BaseModel, Field
+
+from synth_ai.zyk.lms.core.main import LM
+
+
+# Define example structured output models
+class SimpleResponse(BaseModel):
+    message: str
+    confidence_between_zero_one: float = Field(
+        ..., description="Confidence level between 0 and 1"
+    )
+
+
+class ComplexResponse(BaseModel):
+    title: str
+    tags: List[str]
+    content: str
+
+
+class NestedResponse(BaseModel):
+    main_category: str
+    subcategories: List[str]
+    details: SimpleResponse
+
+
+class TestLMStructuredOutputs(unittest.TestCase):
+    @classmethod
+    def setUpClass(cls):
+        # Initialize LMs for both forced_json and stringified_json modes
+        cls.lm_forced_json = LM(
+            model_name="gpt-4o-mini",
+            formatting_model_name="gpt-4o-mini",
+            temperature=0.7,
+            max_retries="Few",
+            structured_output_mode="forced_json",
+        )
+        cls.lm_stringified_json = LM(
+            model_name="gpt-4o-mini",
+            formatting_model_name="gpt-4o-mini",
+            temperature=0.7,
+            max_retries="Few",
+            structured_output_mode="stringified_json",
+        )
+
+    def test_sync_simple_response(self):
+        for lm in [self.lm_forced_json, self.lm_stringified_json]:
+            with self.subTest(
+                mode=lm.structured_output_handler.handler.structured_output_mode
+            ):
+                result = lm.respond_sync(
+                    system_message="You are a helpful assistant.",
+                    user_message="Give me a short greeting and your confidence level.",
+                    response_model=SimpleResponse,
+                )
+                self.assertIsInstance(result.structured_output, SimpleResponse)
+                self.assertIsInstance(result.structured_output.message, str)
+                self.assertIsInstance(
+                    result.structured_output.confidence_between_zero_one, float
+                )
+                self.assertGreaterEqual(
+                    result.structured_output.confidence_between_zero_one, 0
+                )
+                self.assertLessEqual(
+                    result.structured_output.confidence_between_zero_one, 1
+                )
+
+    def test_sync_complex_response(self):
+        for lm in [self.lm_forced_json, self.lm_stringified_json]:
+            with self.subTest(
+                mode=lm.structured_output_handler.handler.structured_output_mode
+            ):
+                result = lm.respond_sync(
+                    system_message="You are a content creator.",
+                    user_message="Create a short blog post about AI.",
+                    response_model=ComplexResponse,
+                )
+                self.assertIsInstance(result.structured_output, ComplexResponse)
+                self.assertIsInstance(result.structured_output.title, str)
+                self.assertIsInstance(result.structured_output.tags, list)
+                self.assertIsInstance(result.structured_output.content, str)
+
+    async def async_nested_response(self, lm):
+        result = await lm.respond_async(
+            system_message="You are a categorization expert.",
+            user_message="Categorize 'Python' and provide a brief description.",
+            response_model=NestedResponse,
+        )
+        self.assertIsInstance(result.structured_output, NestedResponse)
+        self.assertIsInstance(result.structured_output.main_category, str)
+        self.assertIsInstance(result.structured_output.subcategories, list)
+        self.assertIsInstance(result.structured_output.details, SimpleResponse)
+
+    def test_async_nested_response(self):
+        for lm in [self.lm_forced_json, self.lm_stringified_json]:  #
+            with self.subTest(
+                mode=lm.structured_output_handler.handler.structured_output_mode
+            ):
+                asyncio.run(self.async_nested_response(lm))
+
+
+if __name__ == "__main__":
+    unittest.main()

public_tests/test_reasoning_models.py

@@ -43,7 +43,7 @@ async def test_reasoning_effort():
     high_time = time.time() - start_time
 
     print(f"Time taken: {high_time:.2f} seconds")
-    print(f"Response length: {len(high_result)} characters")
+    print(f"Response length: {len(high_result.raw_response)} characters")
     print("-" * 60)
 
     # Create a separate instance for LOW reasoning
@@ -65,7 +65,7 @@ async def test_reasoning_effort():
     low_time = time.time() - start_time
 
     print(f"Time taken: {low_time:.2f} seconds")
-    print(f"Response length: {len(low_result)} characters")
+    print(f"Response length: {len(low_result.raw_response)} characters")
     print("-" * 60)
 
     # Print comparison
@@ -75,15 +75,17 @@ async def test_reasoning_effort():
     print(
         f"Difference: {high_time - low_time:.2f} seconds ({(high_time/low_time - 1)*100:.1f}% difference)"
     )
-    print(f"High response length: {len(high_result)} characters")
-    print(f"Low response length: {len(low_result)} characters")
-    print(f"Response length ratio: {len(high_result)/len(low_result):.2f}x")
+    print(f"High response length: {len(high_result.raw_response)} characters")
+    print(f"Low response length: {len(low_result.raw_response)} characters")
+    print(
+        f"Response length ratio: {len(high_result.raw_response)/len(low_result.raw_response):.2f}x"
+    )
 
     # Print response samples
     print("\nHIGH Response Sample (first 200 chars):")
-    print(high_result[:200] + "...")
+    print(high_result.raw_response[:200] + "...")
     print("\nLOW Response Sample (first 200 chars):")
-    print(low_result[:200] + "...")
+    print(low_result.raw_response[:200] + "...")
 
 
 if __name__ == "__main__":