lm-deluge 0.0.62__tar.gz → 0.0.64__tar.gz

{lm_deluge-0.0.62/src/lm_deluge.egg-info → lm_deluge-0.0.64}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: lm_deluge
-Version: 0.0.62
+Version: 0.0.64
 Summary: Python utility for using LLM API models.
 Author-email: Benjamin Anderson <ben@trytaylor.ai>
 Requires-Python: >=3.10

{lm_deluge-0.0.62 → lm_deluge-0.0.64}/pyproject.toml

@@ -3,7 +3,7 @@ requires = ["setuptools", "wheel"]
 
 [project]
 name = "lm_deluge"
-version = "0.0.62"
+version = "0.0.64"
 authors = [{ name = "Benjamin Anderson", email = "ben@trytaylor.ai" }]
 description = "Python utility for using LLM API models."
 readme = "README.md"

{lm_deluge-0.0.62 → lm_deluge-0.0.64}/src/lm_deluge/client.py

@@ -70,7 +70,7 @@ class _LLMClient(BaseModel):
     top_p: float = 1.0
     json_mode: bool = False
     max_new_tokens: int = 512
-    reasoning_effort: Literal["low", "medium", "high", None] = None
+    reasoning_effort: Literal["low", "medium", "high", "minimal", "none", None] = None
     logprobs: bool = False
     top_logprobs: int | None = None
     force_local_mcp: bool = False
@@ -957,7 +957,7 @@ def LLMClient(
     top_p: float = 1.0,
     json_mode: bool = False,
     max_new_tokens: int = 512,
-    reasoning_effort: Literal["low", "medium", "high", None] = None,
+    reasoning_effort: Literal["low", "medium", "high", "minimal", "none", None] = None,
     logprobs: bool = False,
     top_logprobs: int | None = None,
     force_local_mcp: bool = False,
@@ -986,7 +986,7 @@ def LLMClient(
     top_p: float = 1.0,
     json_mode: bool = False,
     max_new_tokens: int = 512,
-    reasoning_effort: Literal["low", "medium", "high", None] = None,
+    reasoning_effort: Literal["low", "medium", "high", "minimal", "none", None] = None,
     logprobs: bool = False,
     top_logprobs: int | None = None,
     force_local_mcp: bool = False,
@@ -1014,7 +1014,7 @@ def LLMClient(
     top_p: float = 1.0,
     json_mode: bool = False,
     max_new_tokens: int = 512,
-    reasoning_effort: Literal["low", "medium", "high", None] = None,
+    reasoning_effort: Literal["low", "medium", "high", "minimal", "none", None] = None,
     logprobs: bool = False,
     top_logprobs: int | None = None,
     force_local_mcp: bool = False,

{lm_deluge-0.0.62 → lm_deluge-0.0.64}/src/lm_deluge/config.py

@@ -8,7 +8,7 @@ class SamplingParams(BaseModel):
     top_p: float = 1.0
     json_mode: bool = False
     max_new_tokens: int = 512
-    reasoning_effort: Literal["low", "medium", "high", "none", None] = None
+    reasoning_effort: Literal["low", "medium", "high", "minimal", "none", None] = None
     logprobs: bool = False
     top_logprobs: int | None = None
 

{lm_deluge-0.0.62 → lm_deluge-0.0.64}/src/lm_deluge/tool.py

@@ -239,6 +239,9 @@ async def _load_all_mcp_tools(client: Client) -> list["Tool"]:
 
     tools: list[Tool] = []
     for m in metas:
+        # Extract definitions from the schema (could be $defs or definitions)
+        definitions = m.inputSchema.get("$defs") or m.inputSchema.get("definitions")
+
         tools.append(
             Tool(
                 name=m.name,
@@ -246,6 +249,7 @@ async def _load_all_mcp_tools(client: Client) -> list["Tool"]:
                 parameters=m.inputSchema.get("properties", {}),
                 required=m.inputSchema.get("required", []),
                 additionalProperties=m.inputSchema.get("additionalProperties"),
+                definitions=definitions,
                 run=make_runner(m.name),
             )
         )
@@ -268,6 +272,8 @@ class Tool(BaseModel):
     is_built_in: bool = False
     type: str | None = None
     built_in_args: dict[str, Any] = Field(default_factory=dict)
+    # JSON Schema definitions (for $ref support)
+    definitions: dict[str, Any] | None = None
 
     @field_validator("name")
     @classmethod
@@ -562,6 +568,34 @@ class Tool(BaseModel):
         """
         return _python_type_to_json_schema_enhanced(python_type)
 
+    def _is_strict_mode_compatible(self) -> bool:
+        """
+        Check if this tool's schema is compatible with OpenAI strict mode.
+        Strict mode requires all objects to have defined properties.
+        """
+
+        def has_undefined_objects(schema: dict | list | Any) -> bool:
+            """Recursively check for objects without defined properties."""
+            if isinstance(schema, dict):
+                # Check if this is an object type without properties
+                if schema.get("type") == "object":
+                    # If additionalProperties is True or properties is missing/empty
+                    if schema.get("additionalProperties") is True:
+                        return True
+                    if "properties" not in schema or not schema["properties"]:
+                        return True
+                # Recursively check nested schemas
+                for value in schema.values():
+                    if has_undefined_objects(value):
+                        return True
+            elif isinstance(schema, list):
+                for item in schema:
+                    if has_undefined_objects(item):
+                        return True
+            return False
+
+        return not has_undefined_objects(self.parameters or {})
+
     def _json_schema(
         self, include_additional_properties=False, remove_defaults=False
     ) -> dict[str, Any]:
@@ -614,6 +648,14 @@ class Tool(BaseModel):
         else:
             processed_parameters = self.parameters
 
+        # Process definitions too
+        if self.definitions and include_additional_properties:
+            processed_definitions = _add_additional_properties_recursive(
+                self.definitions, remove_defaults
+            )
+        else:
+            processed_definitions = self.definitions
+
         res = {
             "type": "object",
             "properties": processed_parameters,
@@ -623,6 +665,10 @@ class Tool(BaseModel):
         if include_additional_properties:
             res["additionalProperties"] = False
 
+        # Include definitions if present (for $ref support)
+        if processed_definitions:
+            res["$defs"] = processed_definitions
+
         return res
 
     # ---------- dumpers ----------
@@ -631,6 +677,11 @@ class Tool(BaseModel):
     ) -> dict[str, Any]:
         if self.is_built_in:
             return {"type": self.type, **self.built_in_args, **kwargs}
+
+        # Check if schema is compatible with strict mode
+        if strict and not self._is_strict_mode_compatible():
+            strict = False
+
         if strict:
             # For strict mode, remove defaults and make all parameters required
             schema = self._json_schema(

{lm_deluge-0.0.62 → lm_deluge-0.0.64/src/lm_deluge.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: lm_deluge
-Version: 0.0.62
+Version: 0.0.64
 Summary: Python utility for using LLM API models.
 Author-email: Benjamin Anderson <ben@trytaylor.ai>
 Requires-Python: >=3.10