fix(litellm): normalize the remaining LiteLLM provider prefixes to OTel gen_ai.provider.name values

[Anai-Guo]

What

_PROVIDER_ALIASES in the LiteLLM instrumentation normalizes LiteLLM's provider prefixes onto the OTel gen_ai.provider.name well-known values. It covered six prefixes; several other prefixes that LiteLLM reports as custom_llm_provider also differ from the published values and were passing through unchanged:

LiteLLM custom_llm_provider	emitted before	emitted now
watsonx	watsonx	ibm.watsonx.ai
watsonx_text	watsonx_text	ibm.watsonx.ai
xai	xai	x_ai
cohere_chat	cohere_chat	cohere
vertex_ai_beta	vertex_ai_beta	gcp.vertex_ai
azure_text	azure_text	azure.ai.openai
bedrock_mantle	bedrock_mantle	aws.bedrock

Why

This split one provider across two keys depending on the route taken. A watsonx call made through LiteLLM was recorded as watsonx, while the same call through this repo's watsonx instrumentation is recorded as ibm.watsonx.ai (GenAiSystemValues.IBM_WATSONX_AI) — so spans and metrics for the same provider don't aggregate together, which is exactly what the comment above the table says the normalization is there to prevent. The same applies to cohere_chat vs the cohere instrumentation's cohere, and to the vertex_ai_beta / azure_text / bedrock_mantle variants of prefixes that were already normalized.

All seven are real, distinct custom_llm_provider values, verified against LiteLLM 1.94.0:

>>> from litellm import get_llm_provider
>>> get_llm_provider(model="watsonx/ibm/granite-13b-chat-v2")[1]
'watsonx'
>>> get_llm_provider(model="xai/grok-4")[1]
'xai'
>>> get_llm_provider(model="bedrock_mantle/openai.gpt-5.4")[1]
'bedrock_mantle'

Each maps onto an existing GenAiProviderNameValues member, so no new values are invented. Unknown prefixes still pass through unchanged, as before.

Testing

tests/test_provider_resolution.py is parametrized over the new prefixes. Verified old-vs-new against the same test file:

• On main (fix reverted, new tests in place): 7 failed, 10 passed — each failure is the raw prefix leaking through, e.g. assert 'xai' == 'x_ai'.
• With the fix: 17 passed.

Full package suite: 37 passed, 1 skipped (the skip is the pre-existing test_both_openai_and_litellm_instrumented, which needs opentelemetry-instrumentation-openai installed). ruff check clean under the package's config.

🤖 Generated with Claude Code

Summary by CodeRabbit

• Bug Fixes
  • Improved provider detection for additional LiteLLM aliases, including Azure Text, Bedrock Mantle, Cohere Chat, Vertex AI Beta, WatsonX, and xAI.
  • Ensured these provider variants are consistently reported with the correct OpenTelemetry provider values.

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: 74088a35-59f6-4672-b5cf-73bc263ca434

📥 Commits

Reviewing files that changed from the base of the PR and between c2f3f45 and 73106f8.

📒 Files selected for processing (2)

• packages/opentelemetry-instrumentation-litellm/opentelemetry/instrumentation/litellm/__init__.py
• packages/opentelemetry-instrumentation-litellm/tests/test_provider_resolution.py

---

📝 Walkthrough

Walkthrough

The LiteLLM instrumentation now normalizes additional provider aliases. Parameterized tests verify the expected OpenTelemetry provider values.

Changes

LiteLLM provider aliases

Layer / File(s)	Summary
Provider alias normalization and coverage packages/opentelemetry-instrumentation-litellm/opentelemetry/instrumentation/litellm/__init__.py, packages/opentelemetry-instrumentation-litellm/tests/test_provider_resolution.py	Added aliases for Azure Text, Bedrock Mantle, Cohere Chat, Vertex AI Beta, WatsonX, and XAI. Added tests for each normalized provider value.

Estimated code review effort: 1 (Trivial) | ~3 minutes

Possibly related PRs

• traceloop/openllmetry#4322: Introduced the provider alias normalization logic and tests extended by this PR.

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 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: normalizing remaining LiteLLM provider prefixes to OpenTelemetry provider names.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
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.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

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.}