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

Project	Deployment	Actions	Updated (UTC)
litellm	Ready	Preview, Comment	Mar 20, 2026 6:17am

All committers have signed the CLA.

Greptile Summary

This PR fixes a long-standing silent bug where LiteLLM was sending task_type (snake_case) to the Gemini embeddings API instead of the correct taskType (camelCase). The API was silently ignoring the field, meaning task type had no effect. The fix adds the missing task_type → taskType key conversion to both transform_openai_input_gemini_content (batchEmbedContents path) and transform_openai_input_gemini_embed_content (embedContent path), mirroring the already-present dimensions → outputDimensionality pattern.

Changes:

• batch_embed_content_transformation.py: Adds task_type → taskType renaming in both transformation functions (4 lines, following existing idiom)
• test_gemini_batch_embeddings.py: Three new mock-only unit tests covering snake_case conversion for each path and camelCase passthrough preservation

Notes:

• The fix is minimal, correct, and consistent with the existing dimensions mapping pattern
• All new tests use mocks (no real network calls), satisfying CI/CD requirements
• Minor coverage gap: there is no camelCase taskType passthrough test for the batch (transform_openai_input_gemini_content) path — only for transform_openai_input_gemini_embed_content — though this is low risk given the code path is identical

Confidence Score: 5/5

• This PR is safe to merge — it is a minimal, targeted bug fix with proper test coverage and no backward-compatibility concerns.
• The change is 4 lines in the source file, follows an established pattern already in the codebase, correctly fixes an API contract mismatch, and is covered by new mock-based tests. No custom rules are violated, no real network calls are added to the test suite, and the existing 23 tests continue to pass.
• No files require special attention.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[User calls litellm.embedding with task_type param] --> B{Which path?}
    B -->|text-only, non-multimodal| C[transform_openai_input_gemini_content\nbatchEmbedContents path]
    B -->|multimodal / gemini-embedding-2-preview| D[transform_openai_input_gemini_embed_content\nembedContent path]

    C --> E{task_type in gemini_params?}
    E -->|Yes - BEFORE fix| F1["❌ Sends task_type\n(silently ignored by API)"]
    E -->|Yes - AFTER fix| G1["✅ Renames to taskType\n(accepted by API)"]
    E -->|No| H1[Pass through unchanged]

    D --> I{task_type in gemini_params?}
    I -->|Yes - BEFORE fix| F2["❌ Sends task_type\n(silently ignored by API)"]
    I -->|Yes - AFTER fix| G2["✅ Renames to taskType\n(accepted by API)"]
    I -->|No| H2[Pass through unchanged]

    G1 --> J1[EmbedContentRequest with taskType field]
    H1 --> J1
    G2 --> J2[embedContent request body with taskType field]
    H2 --> J2

_{Last reviewed commit: "fix(gemini-embedding..."}

Merging this PR will not alter performance

✅ 16 untouched benchmarks

_{Comparing christopherbaer:fix/gemini-embedding-task-type-camelcase (ed764f7) with main (30589b0)}