_LLM Operations and Integration Platform_
Project description
LLM Operations and Integration Platform
Orquesta Python SDK
Contents
Installation
pip install orquesta-sdk
Creating a client instance
You can get your workspace API key from the settings section in your Orquesta workspace. https://my.orquesta.dev/<workspace>/settings/developers
Initialize the Orquesta client with your API key:
import os
from orquesta_sdk import OrquestaClient, OrquestaClientOptions
api_key = os.environ.get("ORQUESTA_API_KEY", "__API_KEY__")
options = OrquestaClientOptions(
api_key=api_key,
ttl=3600,
environment="production"
)
client = OrquestaClient(options)
When creating a client instance, the following connection settings can be adjusted using the OrquestaClientOptions class:
OrquestaClientOptions
api_key: str - your workspace API key to use for authentication.environment: Optional[str] - the environment to use for the client. Not required but recommended to use so it"s added to the evaluation context automatically.ttl: Optional[int] - the time to live in seconds for the local cache. Default is 3600 seconds (1 hour).
Usage - Endpoints
Use the Endpoints API to query or stream your endpoints from Orquesta.
Using endpoints to generate a LLM response based on your use case with Orquesta provides a low-latency, secure connection to the Endpoints API online prediction service. Getting out of the box metrics and logging for your LLMs.
Endpoints API support streaming and querying. We recommend to use the code snippets provided in the Orquesta Admin panel to reduce risk of errors and improve ease of use.
Example: Querying an endpoint
from orquesta_sdk.endpoints import OrquestaEndpointRequest
request = OrquestaEndpointRequest(
key="customer_service",
context=context: { environments: "production", country: "NLD" },,
variables: { firstname: "John", city: "New York" },
metadata: { customer_id: "Qwtqwty90281" },
)
endpointRef = client.endpoints.query(
request
)
print(endpointRef.content)
Example: Streaming your endpoints
request = OrquestaEndpointRequest(
key="customer_service",
context=context: { environments: "production", country: "NLD" },,
variables: { firstname: "John", city: "New York" },
metadata: { customer_id: "Qwtqwty90281" },
)
endpointRef = None
def handle_next(chunk):
endpointRef = chunk
print(f"Received {chunk.content}")
def handle_error(e):
print(f"Error Occurred: {e}")
def handle_completed():
print("Stream completed!")
const stream = client.endpoints.stream(
request
).subscribe(
on_next=handle_next,
on_error=handle_error,
on_completed=handle_completed,
)
Logging score and metadata for endpoints
After every query, Orquesta will generate a log with the result of the evaluation. You can add metadata and score to the endpoint by using the addMetrics method.
If you need to cancel a stream, you can call stream.unsubscribe() method.
metrics = OrquestaEndpointMetrics(
score=85,
metadata={
"custom": "custom_metadata",
"chain_id": "ad1231xsdaABw",
},
)
endpointRef.addMetrics(metrics);
Usage - Prompts
Use the Prompts API to query your prompts from Orquesta.
You can use Orquesta in prompt management mode by consuming our Prompts API. The prompt value type is OrquestaPrompt. We recommend to use the code snippets provided in the Orquesta Admin panel to reduce risk of errors and improve ease of use.
We support an unified data model structure for all our prompts and provide helper functions that map the returned value from Orquesta to the specific provider.
The query method receives an object of type OrquestaPromptRequest as parameter.
Example: Querying a prompt
from orquesta_sdk.helpers import orquesta_openai_parameters_mapper
prompt = client.prompts.query(
key="prompt_key",
context={"environments": "production", "workspaceId": "soql1odAABC2"},
variables={"firstname": "John", "city": "New York"},
metadata={"chain_id": "ad1231xsdaABw"},
)
openai_api_parameters = orquesta_openai_parameters_mapper(prompt.value)
Helper functions per LLM provider
We provide helper functions that map the returned value from Orquesta to a dict following the definitions of the specific provider, so it"s easy for you to forward the Prompt to your different LLM providers.
| Provider | Helper |
|---|---|
| Anthropic | orquesta_anthropic_parameters_mapper |
| Cohere | orquesta_cohere_parameters_mapper |
orquesta_google_parameters_mapper |
|
| Hugging Face | orquesta_huggingface_parameters_mapper |
| OpenAI | ⚠️ Work in progres |
| Replicate | ⚠️ Work in progres |
Logging metrics and metadata for prompts
After every query, Orquesta will generate a log with the result of the evaluation. You can add metadata and information about the interaction with the LLM to the log by using the add_metrics method.
The properties score, latency, llm_response and economics are reserved and used to generate your real-time dashboards. metadata is a set of key-value pairs that you can use to add custom information to the log.
Example: Add metrics to your request log
from orquesta_sdk.prompts import OrquestaPromptMetricsEconomics, OrquestaPromptMetrics
economics = OrquestaPromptMetricsEconomics(
prompt_tokens=1200,
completion_tokens=750,
total_tokens=1950,
)
metrics = OrquestaPromptMetrics(
score=100,
latency=40,
llm_response="Orquesta is awesome!",
economics=economics,
metadata={
"custom": "custom_metadata",
"chain_id": "ad1231xsdaABw",
"total_interactions": 200,
}
)
prompt.add_metrics(metrics)
Usage - Remote Configurations
Orquesta also comes with a powerful Remote Configurations API that allows you to dynamically configure and run all your environments and services remotely.
Orquesta has a powerful Remote Configurations API that allows you to configure and run all your environments and services remotely dynamically. Orquesta supports different Class of remote configurations, and we recommend always typing the query method to help Classcript infer the correct type.
Supported Class: bool, float, str, dict, list
Example: Querying a configuration of type boolean
config = client.remoteconfigs.query(
key="boolean_config",
default_value=False,
context={"environments": "production", "role": "admin"},
metadata={"user_id": 450}
)
Example: Querying a configuration of type str
config = client.remoteconfigs.query(
key="str_config",
default_value="str_value",
context={"environments": "production", "country": "NL"},
metadata={"timestamp": 1623345600}
)
Example: Querying a configuration of type int
config = client.remoteconfigs.query(
key="int_config",
default_value=1990,
context={"environments": "production", "market": "US" },
metadata={"domain": "ecommerce"}
)
Example: Querying a configuration of type array
config = client.remoteconfigs.query(
key="list_config",
default_value=["USA", "NL"],
context={"environments": "acceptance", "is_enable": True},
metadata={"domain": "ecommerce"}
)
Example: Querying a configuration of type JSON
config = client.remoteconfigs.query(
key="json_config",
default_value=dict,
contenxt={"environments": "develop", "platform": "mobile"},
)
Additional metadata logging
After every query, Orquesta will generate a log with data about the request. You can add metadata to the log using the add_metrics method anytime.
metadata is a set of key-value pairs that you can use to add custom information to the log.
Example: Add metrics to your request log
from orquesta_sdk.remoteconfigs import OrquestaRemoteConfigMetrics
metrics = OrquestaRemoteConfigMetrics(
metadata={
"custom": "custom_metadata",
"user_clicks": 20,
"selected_option": "option1"
}
)
config.add_metrics(metrics)
Orquesta API
Endpoints API
Class:
Methods:
client.endpoints.query({ ...params }) -> OrquestaEndpointclient.endpoints.stream({ ...params }) -> Observable[OrquestaEndpoint]
Prompts API
Class:
Methods:
client.prompts.query({ ...params }) -> OrquestaPrompt
RemoteConfigs API
Class:
OrquestaRemoteConfigKindOrquestaRemoteConfigOrquestaRemoteConfigMetricsOrquestaRemoteConfigRequest
Methods:
client.remoteconfigs.query({ ...params }) -> OrquestaRemoteConfig
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
