synth-ai 0.1.0.dev25__py3-none-any.whl → 0.1.0.dev27__py3-none-any.whl

public_tests/test_sonnet_thinking.py

@@ -0,0 +1,178 @@
+import asyncio
+import unittest
+
+from synth_ai.zyk import LM
+
+
+class TestSonnetThinking(unittest.TestCase):
+    @classmethod
+    def setUpClass(cls):
+        cls.lm = LM(
+            model_name="claude-3-7-sonnet-latest",
+            formatting_model_name="gpt-4o-mini",
+            temperature=0,
+        )
+        # Set reasoning_effort in lm_config
+        cls.lm.lm_config["reasoning_effort"] = "high"
+
+    async def test_thinking_response(self):
+        messages = [
+            {"role": "system", "content": "You are a helpful AI assistant."},
+            {
+                "role": "user",
+                "content": "Please solve this math problem step by step: If a train travels at 60 mph for 2.5 hours, how far does it travel?",
+            },
+        ]
+
+        response = await self.lm.respond_async(messages=messages)
+        print("\n=== Math Problem Test ===")
+        print(f"Response:\n{response}\n")
+        self.assertIsInstance(response, str)
+        self.assertGreater(len(response), 0)
+
+        # Test that the response includes numerical calculation
+        self.assertTrue(any(char.isdigit() for char in response))
+
+    async def test_thinking_structured_output(self):
+        from pydantic import BaseModel
+
+        class MathSolution(BaseModel):
+            steps: list[str]
+            final_answer: float
+            units: str
+
+        messages = [
+            {"role": "system", "content": "You are a math problem solver."},
+            {
+                "role": "user",
+                "content": "If a car travels at 30 mph for 45 minutes, how far does it travel? Provide steps.",
+            },
+        ]
+
+        response = await self.lm.respond_async(
+            messages=messages, response_model=MathSolution
+        )
+
+        print("\n=== Structured Math Problem Test ===")
+        print(f"Steps:")
+        for i, step in enumerate(response.steps, 1):
+            print(f"{i}. {step}")
+        print(f"Final Answer: {response.final_answer} {response.units}\n")
+
+        self.assertIsInstance(response, MathSolution)
+        self.assertGreater(len(response.steps), 0)
+        self.assertIsInstance(response.final_answer, float)
+        self.assertIsInstance(response.units, str)
+
+    async def test_thinking_with_high_effort(self):
+        messages = [
+            {
+                "role": "system",
+                "content": "You are a problem-solving AI. Break down complex problems into detailed steps.",
+            },
+            {
+                "role": "user",
+                "content": "Design a system to automate a coffee shop's inventory management. Consider all aspects.",
+            },
+        ]
+
+        print("\n=== High Effort Thinking Test ===")
+        response = await self.lm.respond_async(messages=messages)
+        print(f"High Effort Response:\n{response}\n")
+        self.assertIsInstance(response, str)
+        self.assertGreater(len(response), 100)  # Expecting detailed response
+
+        # Test with medium effort
+        lm_medium = LM(
+            model_name="claude-3-7-sonnet-latest",
+            formatting_model_name="gpt-4o-mini",
+            temperature=0,
+        )
+        lm_medium.lm_config["reasoning_effort"] = "medium"
+        print("\n=== Medium Effort Thinking Test ===")
+        response_medium = await lm_medium.respond_async(messages=messages)
+        print(f"Medium Effort Response:\n{response_medium}\n")
+        self.assertIsInstance(response_medium, str)
+
+    async def test_thinking_blocks_attributes(self):
+        """Test to verify thinking blocks have the correct attributes and structure"""
+        messages = [
+            {"role": "system", "content": "You are a helpful AI assistant."},
+            {
+                "role": "user",
+                "content": "Please solve this math problem step by step: If a train travels at 60 mph for 2.5 hours, how far does it travel?",
+            },
+        ]
+
+        print("\n=== Testing Thinking Blocks Structure ===")
+        try:
+            response = await self.lm.respond_async(messages=messages)
+            print(f"Response received successfully: {response[:100]}...")
+            self.assertIsInstance(response, str)
+            self.assertGreater(len(response), 0)
+        except AttributeError as e:
+            if "'TextBlock' object has no attribute 'value'" in str(e):
+                self.fail(
+                    "TextBlock missing 'value' attribute - API response structure may have changed"
+                )
+            raise
+
+    async def test_thinking_blocks_with_structured_output(self):
+        """Test thinking blocks with structured output to verify attribute handling"""
+        from pydantic import BaseModel
+
+        class SimpleResponse(BaseModel):
+            answer: str
+            explanation: str
+
+        messages = [
+            {"role": "system", "content": "You are a helpful AI assistant."},
+            {"role": "user", "content": "What is 2+2? Provide answer and explanation."},
+        ]
+
+        print("\n=== Testing Thinking Blocks with Structured Output ===")
+        try:
+            response = await self.lm.respond_async(
+                messages=messages, response_model=SimpleResponse
+            )
+            print(f"Structured response received: {response}")
+            self.assertIsInstance(response, SimpleResponse)
+            self.assertTrue(hasattr(response, "answer"))
+            self.assertTrue(hasattr(response, "explanation"))
+        except AttributeError as e:
+            if "'TextBlock' object has no attribute 'value'" in str(e):
+                self.fail("TextBlock missing 'value' attribute in structured output")
+            raise
+
+    async def test_thinking_blocks_raw_response(self):
+        """Test to examine the raw response structure from the API"""
+        messages = [
+            {"role": "system", "content": "You are a helpful AI assistant."},
+            {"role": "user", "content": "Count from 1 to 3."},
+        ]
+
+        print("\n=== Testing Raw Response Structure ===")
+        try:
+            # Access the raw response if possible
+            response = await self.lm.respond_async(messages=messages)
+            print(f"Raw response type: {type(response)}")
+            print(f"Raw response content: {response}")
+            self.assertIsInstance(response, str)
+        except Exception as e:
+            print(f"Exception type: {type(e)}")
+            print(f"Exception message: {str(e)}")
+            raise
+
+    def test_all(self):
+        print("\nStarting Claude 3.7 Sonnet Thinking Tests...")
+        asyncio.run(self.test_thinking_response())
+        asyncio.run(self.test_thinking_structured_output())
+        asyncio.run(self.test_thinking_with_high_effort())
+        asyncio.run(self.test_thinking_blocks_attributes())
+        asyncio.run(self.test_thinking_blocks_with_structured_output())
+        asyncio.run(self.test_thinking_blocks_raw_response())
+        print("\nAll tests completed successfully!")
+
+
+if __name__ == "__main__":
+    unittest.main()

synth_ai/zyk/lms/vendors/core/anthropic_api.py

@@ -106,8 +106,16 @@ class AnthropicAPI(VendorBase):
         # Make the API call
         response = await self.async_client.messages.create(**api_params)
 
-        api_result = response.content[0].text
-        # Cache with original temperature from lm_config
+        # Handle both regular and thinking responses
+        if hasattr(response.content[0], "text"):
+            api_result = response.content[0].text
+        else:
+            # For thinking responses, get the final output
+            thinking_blocks = [
+                block for block in response.content if block.type == "text"
+            ]
+            api_result = thinking_blocks[-1].value if thinking_blocks else ""
+
         used_cache_handler.add_to_managed_cache(
             model, messages, lm_config=lm_config, output=api_result
         )
@@ -180,7 +188,16 @@ class AnthropicAPI(VendorBase):
         # Make the API call
         response = self.sync_client.messages.create(**api_params)
 
-        api_result = response.content[0].text
+        # Handle both regular and thinking responses
+        if hasattr(response.content[0], "text"):
+            api_result = response.content[0].text
+        else:
+            # For thinking responses, get the final output
+            thinking_blocks = [
+                block for block in response.content if block.type == "text"
+            ]
+            api_result = thinking_blocks[-1].value if thinking_blocks else ""
+
         used_cache_handler.add_to_managed_cache(
             model, messages, lm_config=lm_config, output=api_result
         )
@@ -313,21 +330,3 @@ class AnthropicAPI(VendorBase):
             use_ephemeral_cache_only=use_ephemeral_cache_only,
             **vendor_params,  # Pass all vendor-specific params
         )
-
-    def _prepare_api_params(self, messages, **kwargs):
-        # Ensure max_tokens is at least thinking budget + 4096
-        if "max_tokens" not in kwargs:
-            kwargs["max_tokens"] = 4096 * 2  # Default to 2x context window
-
-        if "thinking" in kwargs and kwargs.get("thinking", {}).get("budget_tokens"):
-            thinking_budget = kwargs["thinking"]["budget_tokens"]
-            kwargs["max_tokens"] = max(kwargs["max_tokens"], thinking_budget + 4096)
-
-        api_params = {
-            "messages": formatted_messages,
-            "model": self.model_name,
-            "max_tokens": kwargs.get("max_tokens", 4096 * 2),
-            # ... rest of params ...
-        }
-
-        return api_params

{synth_ai-0.1.0.dev25.dist-info → synth_ai-0.1.0.dev27.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: synth-ai
-Version: 0.1.0.dev25
+Version: 0.1.0.dev27
 Summary: Software for aiding the best and multiplying the will.
 Home-page: https://github.com/synth-laboratories/synth-ai
 Author: Josh Purtell

{synth_ai-0.1.0.dev25.dist-info → synth_ai-0.1.0.dev27.dist-info}/RECORD

@@ -4,6 +4,7 @@ public_tests/test_all_structured_outputs.py,sha256=x7Gj5Ykpw8Ut_XlSOEBHRLJSagYSH
 public_tests/test_models.py,sha256=7ZJ2HPDZWhcIeZDDu8Iyt5lOy1xpKpYHM8FzsyEKQmc,5703
 public_tests/test_reasoning_models.py,sha256=twKNTrWyeTgtqSC2A4V0g79Uq_SjZiBeWp6ntJIAGNM,2779
 public_tests/test_recursive_structured_outputs.py,sha256=Ne-9XwnOxN7eSpGbNHOpegR-sRj589I84T6y8Z_4QnA,5781
+public_tests/test_sonnet_thinking.py,sha256=XrV9Uxxl4T26zvgeycD7yPoPk2XR43A5bsRjejpEDKk,7003
 public_tests/test_structured_outputs.py,sha256=MZitgGedFlvxeaVFzuDQb2xXs8apwvDLTINpGBfsTdM,3653
 public_tests/test_synth_sdk.py,sha256=jqJHKpvBn9qj21P76z9onXfPg88jyUmBTKmdvCsQMk8,14885
 synth_ai/__init__.py,sha256=2siivzLbT2r-EA7m91dcJB-6Vsurc5_sX3WiKf4_o8Y,198
@@ -35,7 +36,7 @@ synth_ai/zyk/lms/vendors/constants.py,sha256=zqCOyXZqo297wboR9EKVSkvpq6JCMSJyeso
 synth_ai/zyk/lms/vendors/openai_standard.py,sha256=TJz1u6IcJ1KHjbofyHs0rlFa13smVXFTtqBSVqEYJqo,5818
 synth_ai/zyk/lms/vendors/retries.py,sha256=m-WvAiPix9ovnO2S-m53Td5VZDWBVBFuHuSK9--OVxw,38
 synth_ai/zyk/lms/vendors/core/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-synth_ai/zyk/lms/vendors/core/anthropic_api.py,sha256=EozXsU9D_f81TR-igI3bApih8QMVTvexSPnd5HmPtDg,12604
+synth_ai/zyk/lms/vendors/core/anthropic_api.py,sha256=rVAVrjgMXeih3qbISBHf1euilue9Au1n-xDDyQB81n0,12594
 synth_ai/zyk/lms/vendors/core/gemini_api.py,sha256=Cp8BpSk1yCC3SYrEK1pFOnCdUc65XIPonFEirZ6W2rA,5395
 synth_ai/zyk/lms/vendors/core/mistral_api.py,sha256=m11ItQ46VyyCUy6hv6mw5OmiqwHr07wV_NJVNnPHgiA,8080
 synth_ai/zyk/lms/vendors/core/openai_api.py,sha256=700M0QfAxDZXAURnlY--ReEwIEPJPMCwY0JIpu4vptM,5881
@@ -46,11 +47,11 @@ synth_ai/zyk/lms/vendors/supported/deepseek.py,sha256=diFfdhPMO5bLFZxnYj7VT0v6jK
 synth_ai/zyk/lms/vendors/supported/groq.py,sha256=Fbi7QvhdLx0F-VHO5PY-uIQlPR0bo3C9h1MvIOx8nz0,388
 synth_ai/zyk/lms/vendors/supported/ollama.py,sha256=K30VBFRTd7NYyPmyBVRZS2sm0UB651AHp9i3wd55W64,469
 synth_ai/zyk/lms/vendors/supported/together.py,sha256=Ni_jBqqGPN0PkkY-Ew64s3gNKk51k3FCpLSwlNhKbf0,342
-synth_ai-0.1.0.dev25.dist-info/licenses/LICENSE,sha256=ynhjRQUfqA_RdGRATApfFA_fBAy9cno04sLtLUqxVFM,1069
+synth_ai-0.1.0.dev27.dist-info/licenses/LICENSE,sha256=ynhjRQUfqA_RdGRATApfFA_fBAy9cno04sLtLUqxVFM,1069
 tests/test_agent.py,sha256=CjPPWuMWC_TzX1DkDald-bbAxgjXE-HPQvFhq2B--5k,22363
 tests/test_recursive_structured_outputs.py,sha256=Ne-9XwnOxN7eSpGbNHOpegR-sRj589I84T6y8Z_4QnA,5781
 tests/test_structured_outputs.py,sha256=J7sfbGZ7OeB5ONIKpcCTymyayNyAdFfGokC1bcUrSx0,3651
-synth_ai-0.1.0.dev25.dist-info/METADATA,sha256=0nPuKAzWgNeQ9wjobWDSepge5918YWiPdgHTC_NmSvE,2795
-synth_ai-0.1.0.dev25.dist-info/WHEEL,sha256=CmyFI0kx5cdEMTLiONQRbGQwjIoR1aIYB7eCAQ4KPJ0,91
-synth_ai-0.1.0.dev25.dist-info/top_level.txt,sha256=5GzJO9j-KbJ_4ppxhmCUa_qdhHM4-9cHHNU76yAI8do,42
-synth_ai-0.1.0.dev25.dist-info/RECORD,,
+synth_ai-0.1.0.dev27.dist-info/METADATA,sha256=JMIUzbECFL7CJ1EJv-D7Waelv_KYwhI0WyhiyJJCxfI,2795
+synth_ai-0.1.0.dev27.dist-info/WHEEL,sha256=CmyFI0kx5cdEMTLiONQRbGQwjIoR1aIYB7eCAQ4KPJ0,91
+synth_ai-0.1.0.dev27.dist-info/top_level.txt,sha256=5GzJO9j-KbJ_4ppxhmCUa_qdhHM4-9cHHNU76yAI8do,42
+synth_ai-0.1.0.dev27.dist-info/RECORD,,