Revert "fix(embedding): omit null encoding_format for openai requests"

[Sameerlite]

Reverts #25395
This breaks test_encoding_format_none_not_omitted_from_openai_sdk. And even if you remove that line, float gets passed as format in every call to all providers, most of them don't support this, causing errors

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
litellm	Building	Preview, Comment	Apr 14, 2026 3:07pm

[greptile-apps]

Greptile Summary

This PR reverts #25395 ("fix(embedding): omit null encoding_format for openai requests") because the original fix caused the OpenAI Python SDK to inject its default "float" encoding format into calls that should have been format-neutral, breaking providers that don't support the parameter. The code change in main.py is correct — restoring the explicit optional_params["encoding_format"] = None in the OpenAI provider path prevents the SDK from silently adding its default.

• The test file test_openai_embeddings_encoding_format.py is deleted without a replacement, leaving the reverted behavior fully untested. The encoding_format=None → SDK behavior is non-obvious and easy to regress; a mock unit test asserting "encoding_format": None is present in optional_params before the SDK call should be added.

Confidence Score: 4/5

• The code revert is correct, but the deletion of the only test for this behavior leaves a regression risk that should be addressed before merge.
• The main.py change correctly restores the encoding_format=None guard for the OpenAI provider path and fixes the regression introduced by fix(embedding): omit null encoding_format for openai requests #25395. However, the accompanying test file is deleted without a replacement, which violates the project's test coverage guidelines and leaves the behavior unprotected against future regressions. This warrants a 4/5 until a replacement test is added.
• tests/test_litellm/llms/openai/embeddings/test_openai_embeddings_encoding_format.py — needs a replacement mock test verifying the encoding_format=None behavior in the OpenAI path.

Important Files Changed

Filename	Overview
litellm/main.py	Reverts the removal of the else branch that explicitly sets optional_params["encoding_format"] = None for the OpenAI provider path. The logic is correct — it prevents the OpenAI Python SDK from injecting its default "float" value — but the comment has a typo ("Omiting").
tests/test_litellm/llms/openai/embeddings/test_openai_embeddings_encoding_format.py	Deleted as part of the revert. No replacement test is provided, leaving the encoding_format=None behavior completely untested and vulnerable to future regressions.

Sequence Diagram

sequenceDiagram
    participant Caller
    participant litellm_main as litellm.embedding()
    participant get_optional_params as get_optional_params_embeddings()
    participant openai_llm as OpenAIChatCompletion.embedding()
    participant OpenAI_SDK as OpenAI Python SDK

    Caller->>litellm_main: "embedding(model, input, encoding_format=None)"
    litellm_main->>get_optional_params: "encoding_format=None (default)"
    Note over get_optional_params: None == default → omitted from non_default_params
    get_optional_params-->>litellm_main: optional_params (no encoding_format key)

    Note over litellm_main: OpenAI provider path (reverted)
    litellm_main->>litellm_main: "optional_params["encoding_format"] = None (explicit override)"

    litellm_main->>openai_llm: "data = {model, input, encoding_format: None, ...}"
    openai_llm->>OpenAI_SDK: "embeddings.create(**data)"
    Note over OpenAI_SDK: explicit None prevents SDK from injecting default "float"
    OpenAI_SDK-->>openai_llm: EmbeddingResponse
    openai_llm-->>litellm_main: EmbeddingResponse
    litellm_main-->>Caller: EmbeddingResponse

Loading

Comments Outside Diff (1)

1. tests/test_litellm/llms/openai/embeddings/test_openai_embeddings_encoding_format.py

   Test file deleted without a replacement

   This file is being deleted as part of the revert, leaving zero test coverage for the encoding_format=None → OpenAI SDK behavior. The CLAUDE.md guidelines require at least one test when changing functionality, and the custom rule on test coverage requires that tests not be removed in a way that masks regressions.

   The reverted behavior — explicitly setting optional_params["encoding_format"] = None to prevent the OpenAI SDK from defaulting to "float" — is non-obvious and easy to regress. Consider adding a unit/mock test (this folder is mock-only per project rules) that asserts:

   1. When encoding_format=None, the key is still present in optional_params before the SDK call (i.e., "encoding_format": None).
   2. When encoding_format="base64", it is forwarded correctly.

   Without a test, the next PR touching this block can unknowingly re-introduce the same regression that prompted fix(embedding): omit null encoding_format for openai requests #25395.

   Rule Used: What: Flag any modifications to existing tests and... (source)

_{Reviews (1): Last reviewed commit: "Revert "fix(embedding): omit null encodi..." | Re-trigger Greptile}

[greptile-apps]

Comment typo: "Omiting" → "Omitting"

The inline comment has a spelling error that will be seen by every future reader of this block.

Suggested change

	if encoding_format is not None:
	optional_params["encoding_format"] = encoding_format
	else:
	# Omiting causes openai sdk to add default value of "float"
	optional_params["encoding_format"] = None
	if encoding_format is not None:
	optional_params["encoding_format"] = encoding_format
	else:
	# Omitting causes openai sdk to add default value of "float"
	optional_params["encoding_format"] = None