synth-ai 0.1.0.dev37__py3-none-any.whl → 0.1.0.dev39__py3-none-any.whl

synth_ai/zyk/lms/caching/handler.py

@@ -18,6 +18,8 @@ def map_params_to_key(
     temperature: float,
     response_model: Optional[Type[BaseModel]],
     tools: Optional[List] = None,
+    reasoning_effort: Optional[str] = None,
+    max_tokens: Optional[int] = None,
 ) -> str:
     if not all([isinstance(msg["content"], str) for msg in messages]):
         normalized_messages = "".join([str(msg["content"]) for msg in messages])
@@ -26,6 +28,8 @@ def map_params_to_key(
     normalized_model = model
     normalized_temperature = f"{temperature:.2f}"[:4]
     normalized_response_model = str(response_model.schema()) if response_model else ""
+    normalized_reasoning_effort = str(reasoning_effort) if reasoning_effort is not None else ""
+    normalized_max_tokens = str(max_tokens) if max_tokens is not None else ""
 
     # Normalize tools if present
     normalized_tools = ""
@@ -57,6 +61,8 @@ def map_params_to_key(
             + normalized_temperature
             + normalized_response_model
             + normalized_tools
+            + normalized_reasoning_effort
+            + normalized_max_tokens
         ).encode()
     ).hexdigest()
 
@@ -83,6 +89,7 @@ class CacheHandler:
         messages: List[Dict[str, Any]],
         lm_config: Dict[str, Any],
         tools: Optional[List] = None,
+        reasoning_effort: Optional[str] = None,
     ) -> Optional[BaseLMResponse]:
         """Hit the cache with the given key."""
         self._validate_messages(messages)
@@ -93,6 +100,8 @@ class CacheHandler:
             lm_config.get("temperature", 0.0),
             lm_config.get("response_model", None),
             tools,
+            reasoning_effort,
+            lm_config.get("max_tokens"),
         )
         if self.use_persistent_store:
             return persistent_cache.hit_cache(
@@ -112,6 +121,7 @@ class CacheHandler:
         lm_config: Dict[str, Any],
         output: BaseLMResponse,
         tools: Optional[List] = None,
+        reasoning_effort: Optional[str] = None,
     ) -> None:
         """Add the given output to the cache."""
         self._validate_messages(messages)
@@ -123,6 +133,8 @@ class CacheHandler:
             lm_config.get("temperature", 0.0),
             lm_config.get("response_model", None),
             tools,
+            reasoning_effort,
+            lm_config.get("max_tokens"),
         )
         if self.use_persistent_store:
             persistent_cache.add_to_cache(key, output)

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

@@ -64,7 +64,7 @@ class AnthropicAPI(VendorBase):
         ), "response_model is not supported for standard calls"
         used_cache_handler = get_cache_handler(use_ephemeral_cache_only)
         cache_result = used_cache_handler.hit_managed_cache(
-            model, messages, lm_config=lm_config, tools=tools
+            model, messages, lm_config=lm_config, tools=tools, reasoning_effort=reasoning_effort
         )
         if cache_result:
             return cache_result
@@ -135,7 +135,7 @@ class AnthropicAPI(VendorBase):
         )
 
         used_cache_handler.add_to_managed_cache(
-            model, messages, lm_config=lm_config, output=lm_response, tools=tools
+            model, messages, lm_config=lm_config, output=lm_response, tools=tools, reasoning_effort=reasoning_effort
         )
         return lm_response
 
@@ -162,7 +162,7 @@ class AnthropicAPI(VendorBase):
             use_ephemeral_cache_only=use_ephemeral_cache_only
         )
         cache_result = used_cache_handler.hit_managed_cache(
-            model, messages, lm_config=lm_config, tools=tools
+            model, messages, lm_config=lm_config, tools=tools, reasoning_effort=reasoning_effort
         )
         if cache_result:
             return cache_result
@@ -235,7 +235,7 @@ class AnthropicAPI(VendorBase):
         )
 
         used_cache_handler.add_to_managed_cache(
-            model, messages, lm_config=lm_config, output=lm_response, tools=tools
+            model, messages, lm_config=lm_config, output=lm_response, tools=tools, reasoning_effort=reasoning_effort
         )
         return lm_response
 
@@ -249,6 +249,17 @@ class AnthropicAPI(VendorBase):
         reasoning_effort: str = "high",
         **vendor_params: Dict[str, Any],
     ) -> BaseLMResponse:
+        used_cache_handler = get_cache_handler(use_ephemeral_cache_only)
+        lm_config = {"temperature": temperature, "response_model": response_model}
+        cache_result = used_cache_handler.hit_managed_cache(
+            model=model,
+            messages=messages,
+            lm_config=lm_config,
+            reasoning_effort=reasoning_effort,
+        )
+        if cache_result:
+            return cache_result
+        
         try:
             # First try with Anthropic
             reasoning_effort = vendor_params.get("reasoning_effort", reasoning_effort)
@@ -283,6 +294,13 @@ class AnthropicAPI(VendorBase):
                 structured_output=response_model(**parsed),
                 tool_calls=None,
             )
+            used_cache_handler.add_to_managed_cache(
+                model=model,
+                messages=messages,
+                lm_config=lm_config,
+                output=lm_response,
+                reasoning_effort=reasoning_effort,
+            )
             return lm_response
         except (json.JSONDecodeError, pydantic.ValidationError):
             # If Anthropic fails, fallback to OpenAI
@@ -306,6 +324,17 @@ class AnthropicAPI(VendorBase):
         reasoning_effort: str = "high",
         **vendor_params: Dict[str, Any],
     ) -> BaseLMResponse:
+        used_cache_handler = get_cache_handler(use_ephemeral_cache_only)
+        lm_config = {"temperature": temperature, "response_model": response_model}
+        cache_result = used_cache_handler.hit_managed_cache(
+            model=model,
+            messages=messages,
+            lm_config=lm_config,
+            reasoning_effort=reasoning_effort,
+        )
+        if cache_result:
+            return cache_result
+
         try:
             # First try with Anthropic
             reasoning_effort = vendor_params.get("reasoning_effort", reasoning_effort)
@@ -342,6 +371,13 @@ class AnthropicAPI(VendorBase):
                 structured_output=response_model(**parsed),
                 tool_calls=None,
             )
+            used_cache_handler.add_to_managed_cache(
+                model=model,
+                messages=messages,
+                lm_config=lm_config,
+                output=lm_response,
+                reasoning_effort=reasoning_effort,
+            )
             return lm_response
         except (json.JSONDecodeError, pydantic.ValidationError):
             # If Anthropic fails, fallback to OpenAI

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

@@ -237,7 +237,7 @@ class GeminiAPI(VendorBase):
         ), "response_model is not supported for standard calls"
         used_cache_handler = get_cache_handler(use_ephemeral_cache_only)
         cache_result = used_cache_handler.hit_managed_cache(
-            model, messages, lm_config=lm_config, tools=tools
+            model, messages, lm_config=lm_config, tools=tools, reasoning_effort=reasoning_effort
         )
         if cache_result:
             return cache_result
@@ -256,7 +256,7 @@ class GeminiAPI(VendorBase):
         )
 
         used_cache_handler.add_to_managed_cache(
-            model, messages, lm_config=lm_config, output=lm_response, tools=tools
+            model, messages, lm_config=lm_config, output=lm_response, tools=tools, reasoning_effort=reasoning_effort
         )
         return lm_response
 
@@ -282,7 +282,7 @@ class GeminiAPI(VendorBase):
             use_ephemeral_cache_only=use_ephemeral_cache_only
         )
         cache_result = used_cache_handler.hit_managed_cache(
-            model, messages, lm_config=lm_config, tools=tools
+            model, messages, lm_config=lm_config, tools=tools, reasoning_effort=reasoning_effort
         )
         if cache_result:
             return cache_result
@@ -301,6 +301,6 @@ class GeminiAPI(VendorBase):
         )
 
         used_cache_handler.add_to_managed_cache(
-            model, messages, lm_config=lm_config, output=lm_response, tools=tools
+            model, messages, lm_config=lm_config, output=lm_response, tools=tools, reasoning_effort=reasoning_effort
         )
         return lm_response

synth_ai/zyk/lms/vendors/openai_standard.py

@@ -87,7 +87,7 @@ class OpenAIStandard(VendorBase):
         messages = special_orion_transform(model, messages)
         used_cache_handler = get_cache_handler(use_ephemeral_cache_only)
         cache_result = used_cache_handler.hit_managed_cache(
-            model, messages, lm_config=lm_config, tools=tools
+            model, messages, lm_config=lm_config, tools=tools, reasoning_effort=reasoning_effort
         )
         if cache_result:
             return cache_result
@@ -118,7 +118,6 @@ class OpenAIStandard(VendorBase):
 
         # Add reasoning_effort only for o3-mini
         if model in openai_reasoners:
-            print("Reasoning effort:", reasoning_effort)
             api_params["reasoning_effort"] = reasoning_effort
 
         output = await self.async_client.chat.completions.create(**api_params)
@@ -145,7 +144,7 @@ class OpenAIStandard(VendorBase):
             tool_calls=tool_calls,
         )
         used_cache_handler.add_to_managed_cache(
-            model, messages, lm_config=lm_config, output=lm_response, tools=tools
+            model, messages, lm_config=lm_config, output=lm_response, tools=tools, reasoning_effort=reasoning_effort
         )
         return lm_response
 
@@ -173,7 +172,7 @@ class OpenAIStandard(VendorBase):
             use_ephemeral_cache_only=use_ephemeral_cache_only
         )
         cache_result = used_cache_handler.hit_managed_cache(
-            model, messages, lm_config=lm_config, tools=tools
+            model, messages, lm_config=lm_config, tools=tools, reasoning_effort=reasoning_effort
         )
         if cache_result:
             return cache_result
@@ -230,7 +229,7 @@ class OpenAIStandard(VendorBase):
             tool_calls=tool_calls,
         )
         used_cache_handler.add_to_managed_cache(
-            model, messages, lm_config=lm_config, output=lm_response, tools=tools
+            model, messages, lm_config=lm_config, output=lm_response, tools=tools, reasoning_effort=reasoning_effort
         )
         return lm_response
 
@@ -248,7 +247,11 @@ class OpenAIStandard(VendorBase):
         used_cache_handler = get_cache_handler(use_ephemeral_cache_only)
         cache_result: Union[BaseLMResponse, None] = (
             used_cache_handler.hit_managed_cache(
-                model, messages, lm_config=lm_config, tools=tools
+                model,
+                messages,
+                lm_config=lm_config,
+                tools=tools,
+                reasoning_effort=reasoning_effort,
             )
         )
         if cache_result is not None:
@@ -291,7 +294,12 @@ class OpenAIStandard(VendorBase):
             tool_calls=tool_calls,
         )
         used_cache_handler.add_to_managed_cache(
-            model, messages, lm_config=lm_config, output=lm_response, tools=tools
+            model,
+            messages,
+            lm_config=lm_config,
+            output=lm_response,
+            tools=tools,
+            reasoning_effort=reasoning_effort,
         )
         return lm_response
 
@@ -309,7 +317,11 @@ class OpenAIStandard(VendorBase):
         used_cache_handler = get_cache_handler(use_ephemeral_cache_only)
         cache_result: Union[BaseLMResponse, None] = (
             used_cache_handler.hit_managed_cache(
-                model, messages, lm_config=lm_config, tools=tools
+                model,
+                messages,
+                lm_config=lm_config,
+                tools=tools,
+                reasoning_effort=reasoning_effort,
             )
         )
         if cache_result is not None:
@@ -352,6 +364,11 @@ class OpenAIStandard(VendorBase):
             tool_calls=tool_calls,
         )
         used_cache_handler.add_to_managed_cache(
-            model, messages, lm_config=lm_config, output=lm_response, tools=tools
+            model,
+            messages,
+            lm_config=lm_config,
+            output=lm_response,
+            tools=tools,
+            reasoning_effort=reasoning_effort,
         )
         return lm_response

{synth_ai-0.1.0.dev37.dist-info → synth_ai-0.1.0.dev39.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: synth-ai
-Version: 0.1.0.dev37
+Version: 0.1.0.dev39
 Summary: Software for aiding the best and multiplying the will.
 Author: Josh Purtell
 Author-email: Josh Purtell <josh@usesynth.ai>

{synth_ai-0.1.0.dev37.dist-info → synth_ai-0.1.0.dev39.dist-info}/RECORD

@@ -24,7 +24,7 @@ synth_ai/zyk/lms/caching/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZ
 synth_ai/zyk/lms/caching/constants.py,sha256=fPi3x9p-yRdvixMSIyclvmwmwCRliXLXQjEm6dRnG8s,52
 synth_ai/zyk/lms/caching/dbs.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 synth_ai/zyk/lms/caching/ephemeral.py,sha256=pNMG5Rzzp2m0Ln1UYmWxz1qbXwq3iNIrhjYAS0yO3ZE,2370
-synth_ai/zyk/lms/caching/handler.py,sha256=a-4FBxXLWeHCXiGDWP8QU-LPxMAvGbJ_5lUrXnZytn0,4478
+synth_ai/zyk/lms/caching/handler.py,sha256=4h4Kywf0_-WohE1RxBt4cqPo-kHRjZv-2K50WWO91V4,5050
 synth_ai/zyk/lms/caching/initialize.py,sha256=zZls6RKAax6Z-8oJInGaSg_RPN_fEZ6e_RCX64lMLJw,416
 synth_ai/zyk/lms/caching/persistent.py,sha256=ZaY1A9qhvfNKzcAI9FnwbIrgMKvVeIfb_yCyl3M8dxE,2860
 synth_ai/zyk/lms/core/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
@@ -43,11 +43,11 @@ synth_ai/zyk/lms/tools/base.py,sha256=j7wYb1xAvaAm3qVrINphgUhGS-UjZmRpbouseQYgh7
 synth_ai/zyk/lms/vendors/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 synth_ai/zyk/lms/vendors/base.py,sha256=aK4PEtkMLt_o3qD22kW-x3HJUEKdIk06zlH4kX0VkAE,760
 synth_ai/zyk/lms/vendors/constants.py,sha256=3CCq45otD80yaLts5sFHvPgLCQNkcjHkc9cqOQ0zH4Y,320
-synth_ai/zyk/lms/vendors/openai_standard.py,sha256=tgZKV9VUos7is05wd9vJJ3EXZPol5G4uosEjmKiqA0w,12402
+synth_ai/zyk/lms/vendors/openai_standard.py,sha256=SOhm64gI65ZQD2VtvNxZz0dqK3hANEQ2WP6V0kSku20,12877
 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=Yr7xNKa2XnQe5ReEWbxt-3zHv-IQkaV8PmwZqSEB_no,13829
-synth_ai/zyk/lms/vendors/core/gemini_api.py,sha256=PcR4ZHu9NzBabIFeJc2IkT3jCat3Nk8hZm1KoD7uPRM,11390
+synth_ai/zyk/lms/vendors/core/anthropic_api.py,sha256=cgSPFgR9YM8eO1if92fch6pVv0Dxk3xigSXExGSdu-4,15329
+synth_ai/zyk/lms/vendors/core/gemini_api.py,sha256=gsWudX1RP2a4mjuyWZXT3LUek-UMUYhCfE5TMYxp0nA,11530
 synth_ai/zyk/lms/vendors/core/mistral_api.py,sha256=-EMPBEIoYxxDMxukmcmKL8AGAHPNYe4w-76gsPtmrhk,11860
 synth_ai/zyk/lms/vendors/core/openai_api.py,sha256=GDCHIc0kpCnNPj2oW8RE3Cj2U_HcbXzzA5JV1ArAQlE,6600
 synth_ai/zyk/lms/vendors/local/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
@@ -57,11 +57,11 @@ synth_ai/zyk/lms/vendors/supported/deepseek.py,sha256=BElW0NGpkSA62wOqzzMtDw8XR3
 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.dev37.dist-info/licenses/LICENSE,sha256=ynhjRQUfqA_RdGRATApfFA_fBAy9cno04sLtLUqxVFM,1069
+synth_ai-0.1.0.dev39.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.dev37.dist-info/METADATA,sha256=WksAY7eMseikIiWk5JaogfO4yyPQsUqQ6xgSLC152AU,2702
-synth_ai-0.1.0.dev37.dist-info/WHEEL,sha256=CmyFI0kx5cdEMTLiONQRbGQwjIoR1aIYB7eCAQ4KPJ0,91
-synth_ai-0.1.0.dev37.dist-info/top_level.txt,sha256=5GzJO9j-KbJ_4ppxhmCUa_qdhHM4-9cHHNU76yAI8do,42
-synth_ai-0.1.0.dev37.dist-info/RECORD,,
+synth_ai-0.1.0.dev39.dist-info/METADATA,sha256=dUv2K-LU-yTLqXf_wI1bfuLWwy1xLjsV3qnYvnkq830,2702
+synth_ai-0.1.0.dev39.dist-info/WHEEL,sha256=CmyFI0kx5cdEMTLiONQRbGQwjIoR1aIYB7eCAQ4KPJ0,91
+synth_ai-0.1.0.dev39.dist-info/top_level.txt,sha256=5GzJO9j-KbJ_4ppxhmCUa_qdhHM4-9cHHNU76yAI8do,42
+synth_ai-0.1.0.dev39.dist-info/RECORD,,