fix(watsonx): wrap string embedding input in array for WatsonX API by shivamrawat1 · Pull Request #30897 · BerriAI/litellm

shivamrawat1 · 2026-06-20T22:01:39Z

WatsonX embedding requests fail with 400 invalid_request_entity when input is sent as a single string (valid OpenAI format)

Issue
Calling /v1/embeddings on the proxy with a WatsonX model (e.g. multilingual-e5-large backed by watsonx/intfloat/multilingual-e5-large) and a string input returns:

WatsonxException - Failed to deserialize json: 'Mismatch type []string with value string'
The upstream payload shows "inputs":"cdsc" instead of "inputs":["cdsc"]. OpenAI clients and the LiteLLM UI commonly send "input": "cdsc" for single-text embeddings, so this breaks the normal embedding path for WatsonX.

Root cause
IBMWatsonXEmbeddingConfig.transform_embedding_request passed the OpenAI input value straight through as WatsonX inputs. OpenAI accepts both a string and a list of strings; WatsonX /ml/v1/text/embeddings always expects inputs to be []string. A string input was serialized as a JSON string, which WatsonX rejected.

Fix
Normalize input before building the WatsonX request body: wrap a single string in a one-element list, pass through an existing list of strings unchanged, and raise a clear error for token-array inputs (which WatsonX does not support).

Added regression tests in tests/test_litellm/llms/watsonx/embed/test_watsonx_embedding_transformation.py covering string input, list input, and token-array rejection.

Screenshots / Proof of Fix
Before (broken payload sent to WatsonX):

{"inputs":"cdsc","parameters":{...},"model_id":"intfloat/multilingual-e5-large","project_id":"..."}
After fix, unit test + mocked integration confirm the payload is:

{"inputs":["cdsc"],"parameters":{...},"model_id":"intfloat/multilingual-e5-large","project_id":"..."}
Live proxy repro (restart proxy after pulling this branch):

uv run litellm --config litellm/proxy/test.yaml --port 4000 --detailed_debug --reload
curl -s http://localhost:4000/v1/embeddings \
  -H "Authorization: Bearer sk-1234" \
  -H "Content-Type: application/json" \
  -d '{"model":"multilingual-e5-large","input":"cdsc"}'

Type
Bug Fix

Changes
litellm/llms/watsonx/embed/transformation.py: wrap string input into inputs: [str] before calling WatsonX
tests/test_litellm/llms/watsonx/embed/test_watsonx_embedding_transformation.py: regression tests for string, list, and token-a

WatsonX text/embeddings expects inputs as []string; OpenAI clients often send a single string. Co-authored-by: Cursor <[email protected]>

CLAassistant · 2026-06-20T22:01:46Z

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.

Shivam Rawat seems not to be a GitHub user. You need a GitHub account to be able to sign the CLA. If you have already a GitHub account, please add the email address used for this commit to your account.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

greptile-apps · 2026-06-20T22:03:53Z

Greptile Summary

This PR fixes a WatsonX embedding failure where a single-string input was forwarded directly to the WatsonX API as a JSON string, causing a 400 Mismatch type []string with value string error. The fix normalises the input in IBMWatsonXEmbeddingConfig.transform_embedding_request before building the request body, and a new test file validates the three key cases.

String input is now wrapped in a one-element list ("cdsc" → ["cdsc"]), matching the WatsonX []string requirement.
List-of-strings input is passed through unchanged, preserving existing batch-embedding behaviour.
Token-array inputs (List[int] / List[List[int]]) raise a clear ValueError, since WatsonX does not support token-based embeddings; three unit tests (no network calls) cover all branches.

Confidence Score: 5/5

Safe to merge — the change is a one-function normalisation with no effect on other code paths.

The fix is surgical: a 10-line normalisation block added before the return statement of a single method, with no changes to surrounding logic, auth handling, or response parsing. The three new unit tests directly exercise the added branches and confirm the before/after payload shape. No existing tests were modified, no production code outside the WatsonX embedding transformation was touched.

No files require special attention.

Important Files Changed

Filename	Overview
litellm/llms/watsonx/embed/transformation.py	Adds input normalization in transform_embedding_request: wraps string inputs in a list and rejects token-array inputs (List[int] / List[List[int]]) that WatsonX does not support. Change is minimal and correct.
tests/test_litellm/llms/watsonx/embed/test_watsonx_embedding_transformation.py	New test file covering string-wrapping, list pass-through, and token-array rejection. All tests are pure unit tests with no network calls, satisfying the mock-only constraint for this folder.

_{Reviews (2): Last reviewed commit: "fix(watsonx): avoid UP006 in embed trans..." | Re-trigger Greptile}

codecov · 2026-06-20T22:06:09Z

Codecov Report

❌ Patch coverage is 85.71429% with 1 line in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
litellm/llms/watsonx/embed/transformation.py	85.71%	1 Missing ⚠️

📢 Thoughts on this report? Let us know!

Co-authored-by: Cursor <[email protected]>

Use list[str] and branch-based input normalization instead of List and cast so the watsonx embedding change does not add strict ruff UP006 violations. Co-authored-by: Cursor <[email protected]>

shivamrawat1 · 2026-06-20T22:30:49Z

@greptile review again with all commits as we resolved linting issues

mateo-berri

LGTM; thanks!

…erriAI#30897) * fix(watsonx): wrap string embedding input in array for WatsonX API WatsonX text/embeddings expects inputs as []string; OpenAI clients often send a single string. Co-authored-by: Cursor <[email protected]> * style(watsonx): format watsonx embed transformation for black Co-authored-by: Cursor <[email protected]> * fix(watsonx): avoid UP006 in embed transformation strict lint gate Use list[str] and branch-based input normalization instead of List and cast so the watsonx embedding change does not add strict ruff UP006 violations. Co-authored-by: Cursor <[email protected]> --------- Co-authored-by: Shivam Rawat <[email protected]> Co-authored-by: Cursor <[email protected]>

fix(watsonx): wrap string embedding input in array for WatsonX API

70c748d

WatsonX text/embeddings expects inputs as []string; OpenAI clients often send a single string. Co-authored-by: Cursor <[email protected]>

Shivam Rawat and others added 2 commits June 20, 2026 15:24

style(watsonx): format watsonx embed transformation for black

7f740ee

Co-authored-by: Cursor <[email protected]>

fix(watsonx): avoid UP006 in embed transformation strict lint gate

8e430e6

Use list[str] and branch-based input normalization instead of List and cast so the watsonx embedding change does not add strict ruff UP006 violations. Co-authored-by: Cursor <[email protected]>

mateo-berri enabled auto-merge (squash) June 20, 2026 22:50

mateo-berri approved these changes Jun 20, 2026

View reviewed changes

mateo-berri merged commit c7efa77 into litellm_internal_staging Jun 20, 2026
116 of 120 checks passed

mateo-berri deleted the litellm_wtsonx_embedding_string branch June 20, 2026 22:50

ishaan-berri mentioned this pull request Jun 25, 2026

LiteLLM Stability Sprint Roadmap #30484

Open

13 tasks

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Uh oh!

fix(watsonx): wrap string embedding input in array for WatsonX API#30897

fix(watsonx): wrap string embedding input in array for WatsonX API#30897
mateo-berri merged 3 commits into
litellm_internal_stagingfrom
litellm_wtsonx_embedding_string

shivamrawat1 commented Jun 20, 2026

Uh oh!

CLAassistant commented Jun 20, 2026

Uh oh!

greptile-apps Bot commented Jun 20, 2026 •

edited

Loading

Important Files Changed

Uh oh!

codecov Bot commented Jun 20, 2026 •

edited

Loading

Uh oh!

shivamrawat1 commented Jun 20, 2026

Uh oh!

mateo-berri left a comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Uh oh!

Uh oh!

Conversation

shivamrawat1 commented Jun 20, 2026

Uh oh!

CLAassistant commented Jun 20, 2026

Uh oh!

greptile-apps Bot commented Jun 20, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Greptile Summary

Confidence Score: 5/5

Important Files Changed

Uh oh!

codecov Bot commented Jun 20, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

Uh oh!

shivamrawat1 commented Jun 20, 2026

Uh oh!

mateo-berri left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

greptile-apps Bot commented Jun 20, 2026 •

edited

Loading

codecov Bot commented Jun 20, 2026 •

edited

Loading