LiteLLM: `image_generation()`, `transcription()`, `speech()`, and `rerank()` not instrumented

[braintrust-bot]

Summary

The Braintrust LiteLLM wrapper (wrap_litellm() and patch_litellm()) instruments completion, acompletion, responses, aresponses, embedding, and moderation. LiteLLM also exposes several other documented API functions that are not instrumented at all:

• litellm.image_generation() / litellm.aimage_generation() — image generation (DALL-E, etc.)
• litellm.transcription() / litellm.atranscription() — audio transcription (Whisper, etc.)
• litellm.speech() / litellm.aspeech() — text-to-speech
• litellm.rerank() / litellm.arerank() — reranking for RAG pipelines (Cohere, etc.)

Additionally, litellm.aembedding() (async embedding) is not wrapped — the EmbeddingWrapper class only accepts a sync function.

What is missing

No tracing spans are created when users call the above functions through either wrap_litellm() or patch_litellm(). The LiteLLMWrapper class in py/src/braintrust/wrappers/litellm.py (lines 484-518) only exposes: completion, acompletion, responses, aresponses, embedding, moderation. The patch_litellm() function (lines 630-667) only patches: completion, acompletion, responses, aresponses.

Braintrust docs status

not_found — The Braintrust LiteLLM integration docs at https://www.braintrust.dev/docs do not mention image generation, transcription, speech, or rerank support.

Upstream sources

• LiteLLM image generation: https://docs.litellm.ai/docs/image_generation
• LiteLLM audio transcription: https://docs.litellm.ai/docs/audio_transcription
• LiteLLM text-to-speech: https://docs.litellm.ai/docs/text_to_speech
• LiteLLM rerank: https://docs.litellm.ai/docs/rerank
• LiteLLM async embedding: https://docs.litellm.ai/docs/embedding/supported_embedding

Local repo files inspected

• py/src/braintrust/wrappers/litellm.py — full wrapper implementation (668 lines)
• py/src/braintrust/wrappers/test_litellm.py — tests (no tests for the missing functions)
• py/noxfile.py — LiteLLM test sessions

Relationship to existing issues

• Issue LiteLLM: patch_litellm() does not patch embedding or moderation; aembedding missing entirely #115 covers patch_litellm() not patching the existing embedding and moderation wrappers. This issue covers API functions that have no wrapper implementation at all.
• Issue OpenAI: Images API (client.images.generate()) not instrumented #124 covers the OpenAI Images API specifically. This issue covers the LiteLLM multi-provider equivalents.