Skip to content

Leveraging Orchestration Capabilities to Enhance Responses #24080

@frankmjlao

Description

@frankmjlao

Tutorials: https://developers.sap.com/tutorials/ai-core-orchestration-consumption-opt.html

We found some block leading the python sdk script not able to enable the filter condition, some of them i can fix, but some of them not able to fix

  1. How these AzureContentSafty class coming into the picture is not mentioned (not able to fixed)
input_filter= AzureContentSafety(hate=AzureThreshold.ALLOW_SAFE,
                                  violence=AzureThreshold.ALLOW_SAFE,
                                  self_harm=AzureThreshold.ALLOW_SAFE,
                                  sexual=AzureThreshold.ALLOW_SAFE)
input_filter_llama = LlamaGuard38bFilter(hate=True)
output_filter = AzureContentFilter(hate=AzureThreshold.ALLOW_SAFE,
                                   violence=AzureThreshold.ALLOW_SAFE_LOW,
                                   self_harm=AzureThreshold.ALLOW_SAFE_LOW_MEDIUM,
                                   sexual=AzureThreshold.ALLOW_ALL)
output_filter_llama = LlamaGuard38bFilter(hate=True)
  1. when creating the orchestration service instance, missing the proxy_client parameter, i finally solve the issue with following code
from gen_ai_hub.orchestration.service import OrchestrationService 

from gen_ai_hub.proxy.core.proxy_clients import get_proxy_client
proxy_client = get_proxy_client('gen-ai-hub', ai_core_client=ai_core_client) # for an AI Core proxy

print("Proxy client obtained successfully.")

# Initialize an empty list to store the responses 
responses = [] 

# Iterate through each config and get the response using the filtered input 
for i, config in enumerate(configs): 
    orchestration_service = OrchestrationService(api_url=ready_deployment.deployment_url, 
                                                 config=config, 
                                                 proxy_client=proxy_client # this is not documented in the tutorial
                                                 ) 
  1. SystemMessage and UserMessage class seems no longer available in template module, but moved into message modules, like below
from gen_ai_hub.orchestration.models.message import SystemMessage, UserMessage

Write here how you think we can improve the tutorial ...

Metadata

Metadata

Assignees

Labels

No labels
No labels

Type

No type

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions