seekrai 0.5.9__tar.gz → 0.5.12__tar.gz

{seekrai-0.5.9 → seekrai-0.5.12}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: seekrai
-Version: 0.5.9
+Version: 0.5.12
 Summary: Python client for SeekrAI
 License: Apache-2.0
 Author: SeekrFlow

{seekrai-0.5.9 → seekrai-0.5.12}/pyproject.toml

@@ -14,7 +14,7 @@ build-backend = "poetry.core.masonry.api"
 
 [tool.poetry]
 name = "seekrai"
-version = "0.5.9"
+version = "0.5.12"
 authors = [
     "SeekrFlow <support@seekr.com>"
 ]

{seekrai-0.5.9 → seekrai-0.5.12}/src/seekrai/resources/agents/agent_inference.py

@@ -3,6 +3,7 @@ from typing import Any, AsyncGenerator, Iterator, Optional, Union
 from seekrai.abstract import api_requestor
 from seekrai.seekrflow_response import SeekrFlowResponse
 from seekrai.types import ModelSettings, Run, RunRequest, RunResponse, SeekrFlowRequest
+from seekrai.types.agents.runs import ResponseFormat
 
 
 class AgentInference:
@@ -16,7 +17,8 @@ class AgentInference:
         thread_id: str,
         *,
         stream: bool = False,
-        model_settings: Optional[ModelSettings] = None,
+        model_settings: ModelSettings = ModelSettings(),
+        response_format: Optional[Any] = None,
     ) -> Union[RunResponse, Iterator[Any]]:
         """
         Run an inference call on a deployed agent.
@@ -26,13 +28,18 @@ class AgentInference:
             thread_id (str): A thread identifier.
             stream (bool, optional): Whether to stream the response. Defaults to False.
             model_settings (optional): Additional parameters (such as temperature, max_tokens, etc).
+            response_format: Optional structured output specification. If provided, the LLM will be constrained to return JSON matching this schema.
 
         Returns:
             A dictionary with the response (if non-streaming) or an iterator over response chunks.
         """
-        payload = RunRequest(agent_id=agent_id).model_dump()
-        if model_settings is not None:
-            payload["model_settings"] = model_settings.model_dump()
+        payload = RunRequest(
+            agent_id=agent_id,
+            model_settings=model_settings,
+            response_format=ResponseFormat.from_value(response_format)
+            if response_format
+            else None,
+        ).model_dump()
         endpoint = f"threads/{thread_id}/runs"
         if stream:
             endpoint += "/stream"
@@ -147,7 +154,8 @@ class AsyncAgentInference:
         thread_id: str,
         *,
         stream: bool = False,
-        model_settings: Optional[ModelSettings] = None,
+        model_settings: ModelSettings = ModelSettings(),
+        response_format: Optional[Any] = None,
     ) -> Union[RunResponse, AsyncGenerator[Any, None]]:
         """
         Run an inference call on a deployed agent.
@@ -157,13 +165,18 @@ class AsyncAgentInference:
             thread_id (str): A thread identifier.
             stream (bool, optional): Whether to stream the response. Defaults to False.
             model_settings (optional): Additional parameters (such as temperature, max_tokens, etc).
+            response_format: Optional structured output specification. If provided, the LLM will be constrained to return JSON matching this schema.
 
         Returns:
             A dictionary with the response (if non-streaming) or an iterator over response chunks.
         """
-        payload = RunRequest(agent_id=agent_id).model_dump()
-        if model_settings is not None:
-            payload["model_settings"] = model_settings.model_dump()
+        payload = RunRequest(
+            agent_id=agent_id,
+            model_settings=model_settings,
+            response_format=ResponseFormat.from_value(response_format)
+            if response_format
+            else None,
+        ).model_dump()
         endpoint = f"threads/{thread_id}/runs"
         if stream:
             endpoint += "/stream"

{seekrai-0.5.9 → seekrai-0.5.12}/src/seekrai/types/agents/runs.py

@@ -1,16 +1,63 @@
 import datetime
 from enum import Enum
-from typing import Any, Optional, Union
+from typing import Any, Dict, Optional, Union
 
+import pydantic
 from pydantic import Field
 
 from seekrai.types.abstract import BaseModel
 
 
+class ModelSettings(BaseModel):
+    """Settings to use when calling an LLM.
+
+    This class holds optional model configuration parameters (e.g. temperature,
+    top_p, penalties, truncation, etc.).
+
+    Not all models/providers support all of these parameters, so please check the API documentation
+    for the specific model and provider you are using.
+    """
+
+    temperature: float = Field(default=1.0, ge=0.0, le=2.0)
+    top_p: float = Field(default=1.0, ge=0.0, le=1.0)
+    frequency_penalty: float = Field(default=0.0, ge=-2.0, le=2.0)
+    presence_penalty: float = Field(default=0.0, ge=-2.0, le=2.0)
+    max_tokens: Optional[int] = None
+
+
+class ResponseFormat(BaseModel):
+    """Specifies a JSON schema for the response format.
+
+    When provided, the LLM will be constrained to return a JSON response
+    that matches the specified schema.
+
+    Can be instantiated with:
+    - A JSON schema dictionary
+    - A Pydantic model class
+    - An existing ResponseFormat instance
+    """
+
+    json_schema: Dict[str, Any]
+
+    @classmethod
+    def from_value(cls, value: Any) -> "ResponseFormat":
+        if isinstance(value, cls):
+            return value
+        if isinstance(value, dict):
+            return cls(json_schema=value)
+        if isinstance(value, type) and issubclass(value, pydantic.BaseModel):
+            return cls(json_schema=value.model_json_schema())
+        raise ValueError(
+            "ResponseFormat configuration is invalid. Expected ResponseFormat, a valid schema or a Pydantic BaseModel."
+        )
+
+
 class RunRequest(BaseModel):
     """Request model for creating a run."""
 
     agent_id: str = Field(default="default_agent")
+    model_settings: ModelSettings = ModelSettings()
+    response_format: Optional[Union[ResponseFormat, Dict[str, Any], type]] = None
 
 
 class RunResponse(BaseModel):
@@ -115,20 +162,3 @@ class RunStep(BaseModel):
     completed_at: Optional[datetime.datetime] = None
     meta_data: dict[str, Any] = Field(default_factory=dict)
     usage: Optional[RunStepUsage] = None
-
-
-class ModelSettings(BaseModel):
-    """Settings to use when calling an LLM.
-
-    This class holds optional model configuration parameters (e.g. temperature,
-    top_p, penalties, truncation, etc.).
-
-    Not all models/providers support all of these parameters, so please check the API documentation
-    for the specific model and provider you are using.
-    """
-
-    temperature: float = Field(default=1.0, ge=0.0, le=2.0)
-    top_p: float = Field(default=1.0, ge=0.0, le=1.0)
-    frequency_penalty: float = Field(default=0.0, ge=-2.0, le=2.0)
-    presence_penalty: float = Field(default=0.0, ge=-2.0, le=2.0)
-    max_tokens: Optional[int] = None

{seekrai-0.5.9 → seekrai-0.5.12}/src/seekrai/types/files.py

@@ -11,6 +11,7 @@ from seekrai.types.common import (
 
 
 class FilePurpose(str, Enum):
+    ReinforcementFineTune = "reinforcement-fine-tune"
     FineTune = "fine-tune"
     PreTrain = "pre-train"
     Alignment = "alignment"

{seekrai-0.5.9 → seekrai-0.5.12}/src/seekrai/types/finetune.py

@@ -77,6 +77,7 @@ class FinetuneEventType(str, Enum):
 class FineTuneType(str, Enum):
     STANDARD = "STANDARD"
     DPO = "DPO"
+    GRPO = "GRPO"
 
 
 class FinetuneEvent(BaseModel):
@@ -108,6 +109,8 @@ class TrainingConfig(BaseModel):
     batch_size: int = Field(..., ge=1, le=1024)
     # up to 40 character suffix for output model name
     experiment_name: str | None = None
+    # sequence length
+    max_length: int = 2500
     # # weights & biases api key
     # wandb_key: str | None = None
     # IFT by default