fix(gateway): handle deepinfra in streaming transform

[steebchen]

Problem

Streaming requests routed to DeepInfra logged a warning on every chunk:

WARN (llmgateway): [streaming] Unknown provider using OpenAI fallback
    provider: "deepinfra"
    model: "glm-5.1"

DeepInfra uses the OpenAI-compatible /chat/completions endpoint, but deepinfra was missing from the provider switch in transform-streaming-to-openai.ts. Every streamed chunk fell through to the default case, which still works (it applies the same OpenAI transform) but emits a noisy warning per chunk.

Fix

Add deepinfra to the OpenAI-compatible case list alongside the other OpenAI-compatible providers (together-ai, inference.net, etc.), so chunks are handled explicitly and the warning is no longer logged.

🤖 Generated with Claude Code

Summary by CodeRabbit

Bug Fixes

• Improved streaming response handling for the DeepInfra provider by normalizing finish reason values to align with OpenAI's standard format, ensuring consistent behavior across all supported providers.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Repository UI

Review profile: CHILL

Plan: Pro

Run ID: d61c51c4-773e-4ba6-a3a0-a72705036f24

📥 Commits

Reviewing files that changed from the base of the PR and between 0b4f031 and 1588c1a.

📒 Files selected for processing (1)

• apps/gateway/src/chat/tools/transform-streaming-to-openai.ts

---

Walkthrough

The PR adds the "deepinfra" provider to the OpenAI-compatible streaming transformation in the switch case, allowing it to use transformOpenaiStreaming and normalize non-standard finish_reason values the same way as other compatible providers.

Changes

DeepInfra streaming provider integration

Layer / File(s)	Summary
DeepInfra OpenAI streaming integration apps/gateway/src/chat/tools/transform-streaming-to-openai.ts	DeepInfra provider is added to the OpenAI streaming transformation, applying finish-reason normalization (end_turn→stop, abort→canceled, tool_use→tool_calls).

Estimated code review effort

🎯 1 (Trivial) | ⏱️ ~2 minutes

Possibly related PRs

• theopenco/llmgateway#2244: Both PRs adjust finish_reason handling for upstream "abort" during OpenAI-compatible streaming in apps/gateway/src/chat/tools/transform-streaming-to-openai.ts (one adds deepinfra to the same transform/mapping flow; the other adds warning logging when "abort" becomes "canceled").

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 0.00% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly and concisely describes the main change: adding DeepInfra provider handling to the streaming transform function.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch fix-deepinfra-streaming-fallback

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}