Skip to content

feat:Add Nebius Token Factory provider to the catalog#429

Open
amrrs wants to merge 2 commits into
huggingface:mainfrom
amrrs:feat/nebius-token-factory-catalog
Open

feat:Add Nebius Token Factory provider to the catalog#429
amrrs wants to merge 2 commits into
huggingface:mainfrom
amrrs:feat/nebius-token-factory-catalog

Conversation

@amrrs

@amrrs amrrs commented Jul 21, 2026

Copy link
Copy Markdown

Registers Nebius Token Factory (https://tokenfactory.nebius.com) as a built-in OpenAI-compatible provider, following the config-driven catalog introduced in #241 — a single catalog.toml entry, no Python changes.

What's in the entry

  • base_url = https://api.tokenfactory.nebius.com/v1, auth via NEBIUS_API_KEY (Token Factory endpoint, not the legacy AI Studio).
  • The 25 chat models the Token Factory /v1/models endpoint currently serves (embedding model omitted).
  • reasoning_effort thinking levels — verified accepted by the endpoint across models.
  • Default deepseek-ai/DeepSeek-V4-Pro.

Per-model context_windows are intentionally left out to keep the entry minimal (the field is optional; Tau falls back to its default window).

Tests / docs

  • Adds a focused golden test for the entry and registers the provider in the expected built-in roster (test_provider_catalog.py, test_provider_config.py).
  • Renames the test_provider_catalog.py user-catalog placeholder from nebius to example-vendor so it no longer collides with the new built-in (this also removes a stale api.studio.nebius.ai reference).
  • Documents the provider in the providers-and-models guide.

Verification

Ran end-to-end against Token Factory: tau --provider nebius returns live completions on the default model. Full suite passes (81 passed); ruff clean.

🤖 Generated with Claude Code

@amrrs
amrrs requested a review from alejandro-ao as a code owner July 21, 2026 13:30
Register Nebius Token Factory (https://tokenfactory.nebius.com) as a
built-in OpenAI-compatible provider via a catalog.toml entry, following the
config-driven catalog: no Python changes. Exposes the 25 chat models the
Token Factory /v1/models endpoint currently serves (embedding model omitted),
declares reasoning_effort thinking levels (verified accepted by the endpoint),
and authenticates with NEBIUS_API_KEY. Defaults to deepseek-ai/DeepSeek-V4-Pro.

Add a focused golden test for the entry, add the provider to the expected
built-in roster in the catalog/config tests, and rename the user-catalog test
placeholder from "nebius" to "example-vendor" so it no longer collides with
the new built-in (this also drops a stale api.studio.nebius.ai reference).
Document the provider in the providers-and-models guide.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
@amrrs
amrrs force-pushed the feat/nebius-token-factory-catalog branch from aaf26f4 to 7c22a2c Compare July 21, 2026 13:34
@amrrs amrrs changed the title Add Nebius Token Factory provider to the catalog feat:Add Nebius Token Factory provider to the catalog Jul 21, 2026
@amrrs

amrrs commented Jul 21, 2026

Copy link
Copy Markdown
Author

Sorry @alejandro-ao there was an unncessary addition of model context windows added to the TOML file, which now has been fixed and then changes are pushed

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