Skip to content

Conversation

@m-misiura
Copy link
Contributor

Description

llm-d deployments differ from vllm ones and are usually of the form:

<llm-d-ai-gateway>/<namespace>/<model-name>

When configuring the orchestrator with such a model, the current implementation only supports hostname:port and doesn't handle path prefixes, making it impossible to connect to llm-d services.

Error: invalid hostname: 'openai' has an invalid hostname.

To tackle this, I added optional path_prefix field to ServiceConfig to support endpoints with path components before the API paths.

Test plan

Built an image: quay.io/rh-ee-mmisiura/fms-orchestr8-nlp:path_prefix_4_llmd_v0.1 and

Copy link
Collaborator

@evaline-ju evaline-ju left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thanks for the contribution! One clarifying question, and could you make sure to sign off for the DCO (https://github.com/foundation-model-stack/fms-guardrails-orchestrator/pull/523/checks?check_run_id=59587610923)

Copy link
Collaborator

@evaline-ju evaline-ju left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM if DCO is resolved

Signed-off-by: m-misiura <mmisiura@redhat.com>
@m-misiura m-misiura force-pushed the path-prefix-for-llmd branch from 06f5b2b to 43a69f2 Compare January 7, 2026 09:18
@evaline-ju evaline-ju merged commit 95b3b4d into foundation-model-stack:main Jan 7, 2026
2 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants