[bot] LiteLLM: `text_completion()` / `atext_completion()` not instrumented

[braintrust-bot]

Summary

The Braintrust LiteLLM integration instruments completion, acompletion, responses, aresponses, image_generation, aimage_generation, embedding, aembedding, moderation, speech, aspeech, transcription, atranscription, rerank, and arerank. However, LiteLLM also exposes text_completion() and atext_completion() for legacy OpenAI-style text completions (the /completions endpoint), and these are completely uninstrumented.

text_completion() is a documented, actively supported LiteLLM function that provides access to text completion models like gpt-3.5-turbo-instruct and text-davinci-003. It accepts a prompt string (not chat messages) and returns a TextCompletionResponse. LiteLLM translates this across all supported providers. The async variant atext_completion() also exists (confirmed via LiteLLM Router and GitHub issue BerriAI/litellm#7105).

The Braintrust LiteLLM docs page states the integration traces "Chat and text completion / acompletion calls", but the code only instruments completion()/acompletion() (chat completions), not the separate text_completion()/atext_completion() functions.

What is missing

No tracing spans are created when users call litellm.text_completion() or litellm.atext_completion() through either wrap_litellm() or patch_litellm().

There are:

• No text_completion or atext_completion patchers in py/src/braintrust/integrations/litellm/patchers.py
• No _text_completion_wrapper or _atext_completion_wrapper_async in py/src/braintrust/integrations/litellm/tracing.py
• No text_completion test coverage in py/src/braintrust/integrations/litellm/test_litellm.py
• Zero matches for text_completion anywhere in py/src/braintrust/integrations/litellm/

Current _ALL_LITELLM_PATCHERS in patchers.py (lines 125–141):

Function	Patched?
litellm.completion / acompletion	Yes
litellm.responses / aresponses	Yes
litellm.image_generation / aimage_generation	Yes
litellm.embedding / aembedding	Yes
litellm.moderation	Yes (sync only)
litellm.speech / aspeech	Yes
litellm.transcription / atranscription	Yes
litellm.rerank / arerank	Yes
litellm.text_completion / atext_completion	No

Implementation notes

text_completion() returns a TextCompletionResponse with a similar structure to chat completions but uses choices[].text instead of choices[].message. The existing _completion_wrapper logic in tracing.py could likely be adapted with minor changes to handle the text completion response shape. Token usage fields (prompt_tokens, completion_tokens, total_tokens) are the same.

Braintrust docs status

The Braintrust LiteLLM integration page mentions "text completion" in its description, but the SDK does not actually instrument text_completion().

Upstream sources

• LiteLLM text completion docs: https://docs.litellm.ai/docs/text_completion
• LiteLLM text completion tutorial: https://docs.litellm.ai/docs/tutorials/text_completion
• atext_completion confirmed via: [Bug]: atext_completion for Codestral streaming response BerriAI/litellm#7105

Local repo files inspected

• py/src/braintrust/integrations/litellm/patchers.py — 15 patchers defined; none for text_completion or atext_completion
• py/src/braintrust/integrations/litellm/tracing.py — no text_completion wrapper functions
• py/src/braintrust/integrations/litellm/test_litellm.py — no text_completion tests
• py/src/braintrust/integrations/litellm/__init__.py

Relationship to existing issues

• LiteLLM: patch_litellm() does not patch embedding or moderation; aembedding missing entirely #115 (closed): covered patch_litellm() gaps for embedding/moderation — did not mention text_completion
• LiteLLM: speech() / aspeech() not instrumented #165 (closed): covered speech/aspeech — did not mention text_completion
• [bot] LiteLLM: rerank() / arerank() not instrumented #266 (closed): covered rerank/arerank — did not mention text_completion
• No existing open or closed issue mentions text_completion in this repo