Conversation
|
Add the vgr route type to runner config, publish the local tier's live context window through /v1/models, strip Anthropic strict tool fields on NVIDIA Inference API, and document the route.
gburachas
force-pushed
the
vgr/review/03-runtime
branch
from
September 30, 2026 19:21
88a3946 to
f8bfb5f
Compare
gburachas
force-pushed
the
vgr/review/04-public-surface
branch
from
September 30, 2026 19:22
5923512 to
b16ce42
Compare
This was referenced Oct 1, 2026
This branch has not been deployed
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What
Makes VGR available through deployment TOML as
type = "vgr", wiring local, cloud, and optional verifier targets into the existing runner and server.It also adds live context-window discovery for VGR routes backed by llama.cpp, two provider compatibility fixes, and a configuration guide.
Why
The runtime from #718 needs a deployment configuration and an end-to-end server path. Clients also need the local backend’s actual context capacity, and context-overflow errors need to reach the existing fallback handling.
Notes for reviewers
Review this against #718. Start with
VgrRouteConfigandbuild_vgrincrates/switchyard-runner/src/algorithm.rs, then follow route construction inconfig.rsand the server integration test.AlgorithmSpec, target-name resolution, model categories, and the existing route/client construction path. Local and cloud targets populate the efficient/capable categories; verifier targets are registered as judge dependencies.Vgr::new./v1/modelsprobes the VGR local backend’s/propsendpoint and readsdefault_generation_settings.n_ctx. Each probe has a two-second timeout; unavailable or invalid properties leave the configured capabilities unchanged. The request reuses backend authentication and extra headers, with redirects disabled.inference-api.nvidia.com, it removes the unsupported toolstrictfield. These changes affect the shared client, so they merit review separately from the VGR wiring.The configuration example and serving-mode descriptions are in
docs/routing_algorithms/vgr_routing.md. #881 builds on this surface with agentic handoff controls.