seekrai 0.5.2__tar.gz → 0.5.5__tar.gz

{seekrai-0.5.2 → seekrai-0.5.5}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: seekrai
-Version: 0.5.2
+Version: 0.5.5
 Summary: Python client for SeekrAI
 License: Apache-2.0
 Author: SeekrFlow
@@ -17,7 +17,7 @@ Classifier: Programming Language :: Python :: 3.13
 Requires-Dist: click (>=8.1.7,<9.0.0)
 Requires-Dist: eval-type-backport (>=0.1.3,<0.3.0)
 Requires-Dist: filelock (>=3.13.1,<4.0.0)
-Requires-Dist: httpx[http2] (>=0.27.0,<0.28.0)
+Requires-Dist: httpx[http2] (>=0.28.0,<0.29.0)
 Requires-Dist: numpy (>=1.23.5) ; python_version < "3.12"
 Requires-Dist: numpy (>=1.26.0) ; python_version >= "3.12"
 Requires-Dist: pillow (>=10.3.0,<11.0.0)

{seekrai-0.5.2 → seekrai-0.5.5}/pyproject.toml

@@ -14,7 +14,7 @@ build-backend = "poetry.core.masonry.api"
 
 [tool.poetry]
 name = "seekrai"
-version = "0.5.2"
+version = "0.5.5"
 authors = [
     "SeekrFlow <support@seekr.com>"
 ]
@@ -46,7 +46,7 @@ numpy = [
     { version = ">=1.23.5", python = "<3.12" },
     { version = ">=1.26.0", python = ">=3.12" },
 ]
-httpx = {extras = ["http2"], version = "^0.27.0"}
+httpx = {extras = ["http2"], version = "^0.28.0"}
 
 [tool.poetry.group.quality]
 optional = true

{seekrai-0.5.2 → seekrai-0.5.5}/src/seekrai/abstract/response_parsing.py

@@ -22,6 +22,8 @@ def parse_data_line(line: str) -> str:
 def parse_stream(chunks: Iterator[str]) -> Iterator[Any]:
     buffer = []
     for chunk in chunks:
+        if chunk == "data: [DONE]":
+            break
         content = parse_data_line(chunk)
 
         if content:

{seekrai-0.5.2 → seekrai-0.5.5}/src/seekrai/resources/__init__.py

@@ -4,6 +4,7 @@ from seekrai.resources.chat import AsyncChat, Chat
 from seekrai.resources.completions import AsyncCompletions, Completions
 from seekrai.resources.deployments import AsyncDeployments, Deployments
 from seekrai.resources.embeddings import AsyncEmbeddings, Embeddings
+from seekrai.resources.explainability import AsyncExplainability, Explainability
 from seekrai.resources.files import AsyncFiles, Files
 from seekrai.resources.finetune import AsyncFineTuning, FineTuning
 from seekrai.resources.images import AsyncImages, Images
@@ -41,4 +42,6 @@ __all__ = [
     "VectorDatabase",
     "AsyncVectorDatabase",
     "AgentInference",
+    "AsyncExplainability",
+    "Explainability",
 ]

{seekrai-0.5.2 → seekrai-0.5.5}/src/seekrai/resources/agents/agent_inference.py

@@ -1,8 +1,8 @@
-from typing import Any, AsyncGenerator, Iterator, Union
+from typing import Any, AsyncGenerator, Iterator, Optional, Union
 
 from seekrai.abstract import api_requestor
 from seekrai.seekrflow_response import SeekrFlowResponse
-from seekrai.types import Run, RunRequest, RunResponse, SeekrFlowRequest
+from seekrai.types import ModelSettings, Run, RunRequest, RunResponse, SeekrFlowRequest
 
 
 class AgentInference:
@@ -16,7 +16,7 @@ class AgentInference:
         thread_id: str,
         *,
         stream: bool = False,
-        **model_settings: Any,
+        model_settings: Optional[ModelSettings] = None,
     ) -> Union[RunResponse, Iterator[Any]]:
         """
         Run an inference call on a deployed agent.
@@ -25,13 +25,14 @@ class AgentInference:
             agent_id (str): The unique identifier of the deployed agent.
             thread_id (str): A thread identifier.
             stream (bool, optional): Whether to stream the response. Defaults to False.
-            **model_settings: Additional parameters (such as temperature, max_tokens, etc).
+            model_settings (optional): Additional parameters (such as temperature, max_tokens, etc).
 
         Returns:
             A dictionary with the response (if non-streaming) or an iterator over response chunks.
         """
         payload = RunRequest(agent_id=agent_id).model_dump()
-        payload.update(model_settings)
+        if model_settings is not None:
+            payload["model_settings"] = model_settings.model_dump()
         endpoint = f"threads/{thread_id}/runs"
         if stream:
             endpoint += "/stream"
@@ -146,7 +147,7 @@ class AsyncAgentInference:
         thread_id: str,
         *,
         stream: bool = False,
-        **model_settings: Any,
+        model_settings: Optional[ModelSettings] = None,
     ) -> Union[RunResponse, AsyncGenerator[Any, None]]:
         """
         Run an inference call on a deployed agent.
@@ -155,13 +156,14 @@ class AsyncAgentInference:
             agent_id (str): The unique identifier of the deployed agent.
             thread_id (str): A thread identifier.
             stream (bool, optional): Whether to stream the response. Defaults to False.
-            **model_settings: Additional parameters (such as temperature, max_tokens, etc).
+            model_settings (optional): Additional parameters (such as temperature, max_tokens, etc).
 
         Returns:
             A dictionary with the response (if non-streaming) or an iterator over response chunks.
         """
         payload = RunRequest(agent_id=agent_id).model_dump()
-        payload.update(model_settings)
+        if model_settings is not None:
+            payload["model_settings"] = model_settings.model_dump()
         endpoint = f"threads/{thread_id}/runs"
         if stream:
             endpoint += "/stream"

{seekrai-0.5.2 → seekrai-0.5.5}/src/seekrai/resources/chat/completions.py

@@ -22,6 +22,7 @@ class ChatCompletions:
         *,
         messages: List[Dict[str, str]],
         model: str,
+        max_completion_tokens: int | None = None,
         max_tokens: int | None = 512,
         stop: List[str] | None = None,
         temperature: float = 0.7,
@@ -36,7 +37,7 @@ class ChatCompletions:
         safety_model: str | None = None,
         response_format: Dict[str, str | Dict[str, Any]] | None = None,
         tools: Dict[str, str | Dict[str, Any]] | None = None,
-        tool_choice: str | Dict[str, str | Dict[str, str]] | None = "auto",
+        tool_choice: str | Dict[str, str | Dict[str, str]] | None = None,
     ) -> ChatCompletionResponse | Iterator[ChatCompletionChunk]:
         """
         Method to generate completions based on a given prompt using a specified model.
@@ -45,6 +46,7 @@ class ChatCompletions:
             messages (List[Dict[str, str]]): A list of messages in the format
                 `[{"role": seekrai.types.chat_completions.MessageRole, "content": TEXT}, ...]`
             model (str): The name of the model to query.
+            max_completion_tokens (int, optional): The maximum number of tokens the output can contain.
             max_tokens (int, optional): The maximum number of tokens to generate.
                 Defaults to 512.
             stop (List[str], optional): List of strings at which to stop generation.
@@ -99,6 +101,7 @@ class ChatCompletions:
             top_p=top_p,
             top_k=top_k,
             temperature=temperature,
+            max_completion_tokens=max_completion_tokens,
             max_tokens=max_tokens,
             stop=stop,
             repetition_penalty=repetition_penalty,
@@ -110,14 +113,16 @@ class ChatCompletions:
             safety_model=safety_model,
             response_format=response_format,
             tools=tools or [],
-            tool_choice=tool_choice,
         ).model_dump()
+        if tool_choice is not None:
+            parameter_payload["tool_choice"] = tool_choice
 
         response, _, _ = requestor.request(
             options=SeekrFlowRequest(
                 method="POST",
                 url="inference/chat/completions",
                 params=parameter_payload,
+                headers={"content-type": "application/json"},
             ),
             stream=stream,
         )
@@ -139,6 +144,7 @@ class AsyncChatCompletions:
         *,
         messages: List[Dict[str, str]],
         model: str,
+        max_completion_tokens: int | None = None,
         max_tokens: int | None = 512,
         stop: List[str] | None = None,
         temperature: float = 0.7,
@@ -162,6 +168,7 @@ class AsyncChatCompletions:
             messages (List[Dict[str, str]]): A list of messages in the format
                 `[{"role": seekrai.types.chat_completions.MessageRole, "content": TEXT}, ...]`
             model (str): The name of the model to query.
+            max_completion_tokens (int, optional): The maximum number of tokens the output can contain.
             max_tokens (int, optional): The maximum number of tokens to generate.
                 Defaults to 512.
             stop (List[str], optional): List of strings at which to stop generation.
@@ -217,6 +224,7 @@ class AsyncChatCompletions:
             top_p=top_p,
             top_k=top_k,
             temperature=temperature,
+            max_completion_tokens=max_completion_tokens,
             max_tokens=max_tokens,
             stop=stop,
             repetition_penalty=repetition_penalty,
@@ -236,6 +244,7 @@ class AsyncChatCompletions:
                 method="POST",
                 url="inference/chat/completions",
                 params=parameter_payload,
+                headers={"content-type": "application/json"},
             ),
             stream=stream,
         )

seekrai-0.5.5/src/seekrai/resources/explainability.py

@@ -0,0 +1,84 @@
+from typing import Optional
+
+from seekrai.abstract import api_requestor
+from seekrai.resources.resource_base import ResourceBase
+from seekrai.seekrflow_response import SeekrFlowResponse
+from seekrai.types import (
+    SeekrFlowRequest,
+)
+from seekrai.types.explainability import (
+    InfluentialFinetuningDataRequest,
+    InfluentialFinetuningDataResponse,
+)
+
+
+class Explainability(ResourceBase):
+    def get_influential_finetuning_data(
+        self, model_id: str, question: str, answer: Optional[str], k: int = 5
+    ) -> InfluentialFinetuningDataResponse:
+        """
+        Retrieve influential QA pair fine tuning data for a specific model.
+        Args:
+            - model_id (str): ID of the model to explain.
+            - question (str): question from user,
+            - answer (str | None): answer of the finetuned model to the question; if None, the answer is retrieved from the finetuned model specified by model_id,
+            - k (int): the number of results to be retrieved (5 by default)
+        Returns:
+            InfluentialFinetuningDataResponse: Object containing the influential fine tuning data.
+        """
+        requestor = api_requestor.APIRequestor(
+            client=self._client,
+        )
+        # Create query parameters dictionary
+        parameter_payload = InfluentialFinetuningDataRequest(
+            question=question, answer=answer, k=k
+        ).model_dump()
+
+        # if limit is not None:
+        #     params["limit"] = limit
+        # TODO  limits =? timeout: float | None = None,  max_retries: int | None = None,
+
+        response, _, _ = requestor.request(
+            options=SeekrFlowRequest(
+                method="GET",
+                url=f"v1/flow/explain/models/{model_id}/influential-finetuning-data",
+                params=parameter_payload,
+            ),
+            stream=False,
+        )
+        assert isinstance(response, SeekrFlowResponse)
+        return InfluentialFinetuningDataResponse(**response.data)
+
+
+class AsyncExplainability(ResourceBase):
+    async def get_influential_finetuning_data(
+        self, model_id: str, question: str, answer: Optional[str], k: int = 5
+    ) -> InfluentialFinetuningDataResponse:
+        """
+        Retrieve influential QA pair finetuning data for a specific model asynchronously.
+        Args:
+            - model_id (str): ID of the model to explain.
+            - question (str): question from user,
+            - answer (str | None): answer of the finetuned model to the question; if None, the answer is retrieved from the finetuned model specified by model_id,
+            - k (int): the number of results to be retrieved (5 by default),
+        Returns:
+            InfluentialFinetuningDataResponse: Object containing the influential finetuning data.
+        """
+        requestor = api_requestor.APIRequestor(
+            client=self._client,
+        )
+        # Create query parameters dictionary
+        parameter_payload = InfluentialFinetuningDataRequest(
+            model_id=model_id, question=question, answer=answer, k=k
+        ).model_dump()
+
+        response, _, _ = await requestor.arequest(
+            options=SeekrFlowRequest(
+                method="GET",
+                url=f"v1/flow/explain/models/{model_id}/influential-finetuning-data",
+                params=parameter_payload,
+            ),
+            stream=False,
+        )
+        assert isinstance(response, SeekrFlowResponse)
+        return InfluentialFinetuningDataResponse(**response.data)

{seekrai-0.5.2 → seekrai-0.5.5}/src/seekrai/types/__init__.py

@@ -13,9 +13,11 @@ from seekrai.types.agents import (
     InputMessage,
     InputText,
     MessageUpdateRequest,
+    ModelSettings,
     OutputGuardrail,
     OutputMessage,
     OutputText,
+    ReasoningEffort,
     Run,
     RunRequest,
     RunResponse,
@@ -183,9 +185,11 @@ __all__ = [
     "RunUsage",
     "RunStatus",
     "RunStepUsage",
+    "ModelSettings",
     "Agent",
     "AgentStatus",
     "CreateAgentRequest",
+    "ReasoningEffort",
     "AgentDeleteResponse",
     "ToolBase",
     "ToolType",

{seekrai-0.5.2 → seekrai-0.5.5}/src/seekrai/types/agents/__init__.py

@@ -3,8 +3,10 @@ from seekrai.types.agents.agent import (
     AgentDeleteResponse,
     AgentStatus,
     CreateAgentRequest,
+    ReasoningEffort,
 )
 from seekrai.types.agents.runs import (
+    ModelSettings,
     Run,
     RunRequest,
     RunResponse,
@@ -51,6 +53,7 @@ __all__ = [
     "RunUsage",
     "RunStatus",
     "RunStepUsage",
+    "ModelSettings",
     "MessageUpdateRequest",
     "ThreadCreateRequest",
     "ThreadStatus",
@@ -78,6 +81,7 @@ __all__ = [
     "Agent",
     "AgentStatus",
     "CreateAgentRequest",
+    "ReasoningEffort",
     "AgentDeleteResponse",
     "ToolBase",
     "ToolType",

{seekrai-0.5.2 → seekrai-0.5.5}/src/seekrai/types/agents/agent.py

@@ -14,11 +14,17 @@ class AgentStatus(str, enum.Enum):
     FAILED = "Failed"
 
 
+class ReasoningEffort(str, enum.Enum):
+    PERFORMANCE_OPTIMIZED = "performance_optimized"
+    SPEED_OPTIMIZED = "speed_optimized"
+
+
 class CreateAgentRequest(BaseModel):
     name: str
     instructions: str
     tools: list[Tool]
     model_id: str
+    reasoning_effort: Optional[ReasoningEffort] = None
 
 
 class Agent(BaseModel):
@@ -35,6 +41,7 @@ class Agent(BaseModel):
     updated_at: datetime
     last_deployed_at: Optional[datetime] = None
     active_duration: int = Field(default=0, ge=0)
+    reasoning_effort: ReasoningEffort
 
 
 class AgentDeleteResponse(BaseModel):

{seekrai-0.5.2 → seekrai-0.5.5}/src/seekrai/types/agents/runs.py

@@ -115,3 +115,20 @@ 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-0.5.2 → seekrai-0.5.5}/src/seekrai/types/agents/tools/schemas/file_search_env.py

@@ -13,4 +13,3 @@ class FileSearchEnv(EnvConfig):
     score_threshold: float = Field(
         default=0, ge=0, lt=1.0, description="Score must be ≥ 0.0 and < 1.0"
     )
-

{seekrai-0.5.2 → seekrai-0.5.5}/src/seekrai/types/chat_completions.py

@@ -78,6 +78,7 @@ class ChatCompletionRequest(BaseModel):
     messages: List[ChatCompletionMessage]
     # model name
     model: str
+    max_completion_tokens: int | None = None
     # stopping criteria: max tokens to generate
     max_tokens: int | None = None
     # stopping criteria: list of strings to stop generation

{seekrai-0.5.2 → seekrai-0.5.5}/src/seekrai/types/deployments.py

@@ -35,6 +35,7 @@ class DeploymentProcessor(str, enum.Enum):
     H100 = "H100"
     XEON = "XEON"
     NVIDIA = "NVIDIA"  # TODO - this doesnt make sense with A100, etc.
+    AMD = "AMD"
 
 
 class NewDeploymentRequest(BaseModel):

seekrai-0.5.5/src/seekrai/types/explainability.py

@@ -0,0 +1,57 @@
+from __future__ import annotations
+
+from datetime import datetime
+from enum import Enum
+from typing import Any, Dict, List, Literal
+
+from pydantic import Field
+
+from seekrai.types.abstract import BaseModel
+
+
+class InfluentialFinetuningDataResponse(BaseModel):
+    results: List[Dict[str, Any]]
+    version: str
+
+
+class InfluentialFinetuningDataRequest(BaseModel):
+    question: str
+    answer: str = Field(
+        default="",
+        description="Response could be generated or given",
+    )
+    k: int
+
+
+class ExplainabilityJobStatus(Enum):
+    QUEUED = "queued"
+    RUNNING = "running"
+    COMPLETED = "completed"
+    FAILED = "failed"
+
+    # TODO should titles along the following get added:
+    # create_index
+    # populate_index
+    # delete_index
+    # influential-finetuning-data
+
+
+class ExplainabilityRequest(BaseModel):
+    files: List[str] = Field(
+        default=..., description="List of file ids to use for fine tuning"
+    )
+    method: str = Field(default="best", description="Method to use for explainability")
+
+
+class ExplainabilityResponse(BaseModel):
+    id: str = Field(default=..., description="Explainability job ID")
+    created_at: datetime
+    status: ExplainabilityJobStatus
+    output_files: List[str]
+
+
+class ExplainabilityList(BaseModel):
+    # object type
+    object: Literal["list"] | None = None
+    # list of fine-tune job objects
+    data: List[ExplainabilityResponse] | None = None

{seekrai-0.5.2 → seekrai-0.5.5}/src/seekrai/types/finetune.py

@@ -72,6 +72,11 @@ class FinetuneEventType(str, Enum):
     WARNING = "WARNING"
 
 
+class FineTuneType(str, Enum):
+    STANDARD = "STANDARD"
+    DPO = "DPO"
+
+
 class FinetuneEvent(BaseModel):
     """
     Fine-tune event type
@@ -105,6 +110,8 @@ class TrainingConfig(BaseModel):
     # wandb_key: str | None = None
     # IFT by default
     pre_train: bool = False
+    # fine-tune type
+    fine_tune_type: FineTuneType = FineTuneType.STANDARD
 
 
 class AcceleratorType(str, Enum):
@@ -138,6 +145,8 @@ class FinetuneResponse(BaseModel):
 
     # job ID
     id: str | None = None
+    # fine-tune type
+    fine_tune_type: FineTuneType = FineTuneType.STANDARD
     # training file id
     training_files: List[str] | None = None
     # validation file id