Skip to content

docs: document OrcaRouter as an OpenAI-compatible sampling provider - #9995

Open
nissrin2020ali-ux wants to merge 1 commit into
modelscope:mainfrom
nissrin2020ali-ux:docs/add-orcarouter-provider
Open

docs: document OrcaRouter as an OpenAI-compatible sampling provider#9995
nissrin2020ali-ux wants to merge 1 commit into
modelscope:mainfrom
nissrin2020ali-ux:docs/add-orcarouter-provider

Conversation

@nissrin2020ali-ux

Copy link
Copy Markdown

PR type

  • Bug Fix
  • New Feature
  • Document Updates
  • More Models or Datasets Support

PR information

Summary

This PR documents OrcaRouter as a named, OpenAI-compatible sampling provider in the "Sampling From Large Model" section of the Sampling guide (English and Chinese), mirroring the existing DashScope base_url example.

SWIFT already ships a fully generic OpenAI-compatible sampling path: swift sample --sampler_engine client uses OpenAIEngine (swift/pipelines/sampling/distill_sampler.py), which constructs an openai.OpenAI client from --engine_kwargs '{"base_url": ...}'. Any OpenAI-compatible endpoint works today, so OrcaRouter requires no code change. This PR only surfaces it as a first-class provider in the docs, in the same way DashScope is documented today.

OrcaRouter is an OpenAI-compatible AI gateway built for both models and agents. Like OpenRouter, it exposes a provider/model namespace across many models — but it also combines adaptive routing, automatic failover, zero-markup inference, observability, guardrails, and agent-tool governance behind the same endpoint. Adding orcarouter as a first-class provider means this project's users can use that stack directly, without treating OrcaRouter as an anonymous custom base URL. It also runs gateway-level, zero-trust security for AI agents on the same endpoint — screening every prompt/response and governing every tool call on a default-deny basis, with no application code changes.

Changes

  • docs/source_en/Instruction/Sample.md: added an OrcaRouter example after the DashScope one, using --engine_kwargs '{"base_url":"https://api.orcarouter.ai/v1"}' and the gateway model deepseek/deepseek-reasoner (mirroring the existing DeepSeek-R1 note about reasoning_content).
  • docs/source/Instruction/Sample.md: same change in Chinese.

Verification

  • pre-commit run --files on both changed files: all hooks pass.
  • L3 live test against https://api.orcarouter.ai/v1 using the OpenAI client exactly as OpenAIEngine does (non-stream and --stream true): both return HTTP 200 with a valid completion, and the deepseek/deepseek-reasoner model returns reasoning_content matching the existing doc note.

Discord: discord.gg/YEubt8enRA · X: https://x.com/OrcaRouter

I'm an engineer on the OrcaRouter team.

Add OrcaRouter as a named example alongside the existing DashScope
base_url in the Sampling From Large Model section of both the English
and Chinese Sampling docs, mirroring the current OpenAI API example.

The swift sample OpenAIEngine client already supports arbitrary
OpenAI-compatible base_url via --engine_kwargs, so OrcaRouter works
out of the box; this change documents the provider and its endpoint.

Co-Authored-By: Claude <noreply@anthropic.com>
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.

1 participant