chore: Sync updates from stainless branch: ehhuang/dev

src/llama_stack_client/types/chat_completion_response_stream_chunk.py

@@ -1,7 +1,6 @@
 # File generated from our OpenAPI spec by Stainless. See CONTRIBUTING.md for details.
 
-from typing import Dict, List, Union, Optional
-from datetime import datetime
+from typing import List, Optional
 from typing_extensions import Literal
 
 from .._models import BaseModel
@@ -31,19 +30,9 @@ class Event(BaseModel):
 class Metric(BaseModel):
     metric: str
 
-    span_id: str
-
-    timestamp: datetime
-
-    trace_id: str
-
-    type: Literal["metric"]
-
-    unit: str
-
     value: float
 
-    attributes: Optional[Dict[str, Union[str, float, bool, None]]] = None
+    unit: Optional[str] = None
 
 
 class ChatCompletionResponseStreamChunk(BaseModel):

src/llama_stack_client/types/completion_response.py

@@ -6,7 +6,15 @@
 from .._models import BaseModel
 from .token_log_probs import TokenLogProbs
 
-__all__ = ["CompletionResponse"]
+__all__ = ["CompletionResponse", "Metric"]
+
+
+class Metric(BaseModel):
+    metric: str
+
+    value: float
+
+    unit: Optional[str] = None
 
 
 class CompletionResponse(BaseModel):
@@ -18,3 +26,5 @@ class CompletionResponse(BaseModel):
 
     logprobs: Optional[List[TokenLogProbs]] = None
     """Optional log probabilities for generated tokens"""
+
+    metrics: Optional[List[Metric]] = None

src/llama_stack_client/types/shared/chat_completion_response.py

@@ -1,8 +1,6 @@
 # File generated from our OpenAPI spec by Stainless. See CONTRIBUTING.md for details.
 
-from typing import Dict, List, Union, Optional
-from datetime import datetime
-from typing_extensions import Literal
+from typing import List, Optional
 
 from ..._models import BaseModel
 from ..token_log_probs import TokenLogProbs
@@ -14,19 +12,9 @@
 class Metric(BaseModel):
     metric: str
 
-    span_id: str
-
-    timestamp: datetime
-
-    trace_id: str
-
-    type: Literal["metric"]
-
-    unit: str
-
     value: float
 
-    attributes: Optional[Dict[str, Union[str, float, bool, None]]] = None
+    unit: Optional[str] = None
 
 
 class ChatCompletionResponse(BaseModel):

src/llama_stack_client/types/shared/tool_response_message.py

@@ -1,6 +1,5 @@
 # File generated from our OpenAPI spec by Stainless. See CONTRIBUTING.md for details.
 
-from typing import Union
 from typing_extensions import Literal
 
 from ..._models import BaseModel
@@ -18,6 +17,3 @@ class ToolResponseMessage(BaseModel):
 
     role: Literal["tool"]
     """Must be "tool" to identify this as a tool response"""
-
-    tool_name: Union[Literal["brave_search", "wolfram_alpha", "photogen", "code_interpreter"], str]
-    """Name of the tool that was called"""

src/llama_stack_client/types/shared_params/tool_response_message.py

@@ -2,7 +2,6 @@
 
 from __future__ import annotations
 
-from typing import Union
 from typing_extensions import Literal, Required, TypedDict
 
 from .interleaved_content import InterleavedContent
@@ -19,6 +18,3 @@ class ToolResponseMessage(TypedDict, total=False):
 
     role: Required[Literal["tool"]]
     """Must be "tool" to identify this as a tool response"""
-
-    tool_name: Required[Union[Literal["brave_search", "wolfram_alpha", "photogen", "code_interpreter"], str]]
-    """Name of the tool that was called"""