fix(bedrock): encode model arns for OpenAI compatible bedrock imported models by ta-stripe · Pull Request #21701 · BerriAI/litellm

ta-stripe · 2026-02-20T21:00:36Z

Relevant issues

Fixes error in #17097 when OpenAI compatible Bedrock imported models were introduced.

Pre-Submission checklist

Please complete all items before asking a LiteLLM maintainer to review your PR

I have Added testing in the tests/litellm/ directory, Adding at least 1 test is a hard requirement - see details
My PR passes all unit tests on make test-unit
My PR's scope is as isolated as possible, it only solves 1 specific problem
I have requested a Greptile review by commenting @greptileai and received a Confidence Score of at least 4/5 before requesting a maintainer review

CI (LiteLLM team)

CI status guideline:

50-55 passing tests: main is stable with minor issues.

45-49 passing tests: acceptable but needs attention

<= 40 passing tests: unstable; be careful with your merges and assess the risk.

Branch creation CI run
Link:
CI run for the last commit
Link:
Merge / cherry-pick CI run
Links:

Type

🐛 Bug Fix

Changes

This PR fixes a bug in how Bedrock model ARNs are sent when using the bedrock/openai/ provider route. Model IDs that are ARNs (e.g. arn:aws:bedrock:...:imported-model/xyz123xyz123) must be URL-encoded when used in the request path; otherwise the AWS API misparses the path and returns an UnknownOperationException error.

# AWS Accepts
POST /bedrock/model/arn:aws:bedrock:us-west-2:123456789000:imported-model%2Fxyz123xyz123/invoke
# AWS return com.amazon.coral.service#UnknownOperationException 
POST /bedrock/model/arn:aws:bedrock:us-west-2:123456789000:imported-model/xyz123xyz123/invoke

This only affects Bedrock models that use the OpenAI-compatible Runtime API (e.g. imported Qwen2.5, Qwen2-VL, Qwen2.5-VL, and GPT-OSS). For models that are invoked through the Converse API, the model arns are automatically encoded in converse_handler.py; and the Runtime invoke path did not, so ARNs containing / (e.g. :imported-model/...) were sent unencoded and triggered the error.

vercel · 2026-02-20T21:00:41Z

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

Project	Deployment	Actions	Updated (UTC)
litellm	Ready	Preview, Comment	Feb 20, 2026 9:08pm

greptile-apps · 2026-02-20T21:03:35Z

Greptile Summary

This PR fixes a bug where Bedrock OpenAI-compatible imported models using ARNs (e.g., arn:aws:bedrock:us-east-1:...:imported-model/m4gc1mrfuddy) had their ARN slashes interpreted as URL path separators, causing 500 errors from the Bedrock runtime endpoint. The fix reuses the existing CommonUtils.encode_bedrock_runtime_modelid_arn utility (already used in the passthrough path) to encode the slash after the resource type (e.g., imported-model/ becomes imported-model%2F).

Adds a call to CommonUtils.encode_bedrock_runtime_modelid_arn(model_id) in AmazonBedrockOpenAIConfig.get_complete_url() before constructing the invoke URL
Updates the existing test assertion to expect the encoded %2F in the URL
No new test added in tests/litellm/ directory (required by contribution guidelines)

Confidence Score: 4/5

This PR is safe to merge — it applies a proven encoding utility to a missing code path, fixing a real bug with imported model ARNs.
The fix is minimal (3 lines of new code) and reuses an existing, well-tested utility (CommonUtils.encode_bedrock_runtime_modelid_arn) that is already used in the bedrock passthrough path. The regex patterns in the utility correctly handle the :imported-model/ pattern. The existing test is updated to validate the fix. Score is 4 instead of 5 because the PR lacks a new unit test in the tests/litellm/ directory as required by contribution guidelines.
No files require special attention — the change is straightforward and low-risk.

Important Files Changed

Filename	Overview
litellm/llms/bedrock/chat/invoke_transformations/amazon_openai_transformation.py	Adds ARN encoding for model IDs using `CommonUtils.encode_bedrock_runtime_modelid_arn` before constructing the invoke URL. The encoding correctly converts slashes in ARN resource types (e.g., `:imported-model/` to `:imported-model%2F`) to prevent them from being treated as path separators. Import placed at module level per project conventions.
tests/llm_translation/test_bedrock_completion.py	Updates existing test assertion to expect URL-encoded slash (`%2F`) in the imported-model ARN, matching the new encoding behavior. Test validates the fix is working correctly.

Sequence Diagram

sequenceDiagram
    participant Client
    participant LiteLLM as LiteLLM (AmazonBedrockOpenAIConfig)
    participant Encoder as CommonUtils.encode_bedrock_runtime_modelid_arn
    participant Bedrock as AWS Bedrock Runtime

    Client->>LiteLLM: completion(model="bedrock/openai/arn:...:imported-model/abc123")
    LiteLLM->>LiteLLM: _get_openai_model_id() strips prefixes
    Note right of LiteLLM: model_id = "arn:...:imported-model/abc123"
    LiteLLM->>Encoder: encode_bedrock_runtime_modelid_arn(model_id)
    Encoder-->>LiteLLM: "arn:...:imported-model%2Fabc123"
    LiteLLM->>LiteLLM: Build URL: /model/{encoded_model_id}/invoke
    LiteLLM->>Bedrock: POST /model/arn:...:imported-model%2Fabc123/invoke
    Bedrock-->>Client: Response

_{Last reviewed commit: 4885b36}

greptile-apps

_{36 files reviewed, 1 comment}

_{Edit Code Review Agent Settings | Greptile}

…arn encoded

ta-stripe · 2026-02-20T21:07:56Z

@greptileai

greptile-apps

_{2 files reviewed, no comments}

_{Edit Code Review Agent Settings | Greptile}

vercel Bot deployed to Preview February 20, 2026 21:02 View deployment

greptile-apps Bot reviewed Feb 20, 2026

View reviewed changes

Comment thread debug_responses_import.py Outdated

fix: bedrock openai-compatible imported-model should also have model …

4885b36

…arn encoded

ta-stripe force-pushed the fix/bedrock-openai-imported-model-name-encoding branch from 96a3554 to 4885b36 Compare February 20, 2026 21:07

vercel Bot deployed to Preview February 20, 2026 21:08 View deployment

greptile-apps Bot reviewed Feb 20, 2026

View reviewed changes

Sameerlite merged commit 55ee8cd into BerriAI:main Feb 23, 2026
30 checks passed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

fix(bedrock): encode model arns for OpenAI compatible bedrock imported models#21701

fix(bedrock): encode model arns for OpenAI compatible bedrock imported models#21701
Sameerlite merged 1 commit intoBerriAI:mainfrom
ta-stripe:fix/bedrock-openai-imported-model-name-encoding

ta-stripe commented Feb 20, 2026 •

edited

Loading

Uh oh!

vercel Bot commented Feb 20, 2026 •

edited

Loading

Uh oh!

greptile-apps Bot commented Feb 20, 2026 •

edited

Loading

Important Files Changed

Uh oh!

greptile-apps Bot left a comment

Uh oh!

Uh oh!

ta-stripe commented Feb 20, 2026

Uh oh!

greptile-apps Bot left a comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Uh oh!

Conversation

ta-stripe commented Feb 20, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Relevant issues

Pre-Submission checklist

CI (LiteLLM team)

Type

Changes

Uh oh!

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

Uh oh!

Uh oh!

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

Uh oh!

Greptile Summary

Confidence Score: 4/5

Important Files Changed

Sequence Diagram

Uh oh!

greptile-apps Bot left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

ta-stripe commented Feb 20, 2026

Uh oh!

greptile-apps Bot left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

ta-stripe commented Feb 20, 2026 •

edited

Loading

vercel Bot commented Feb 20, 2026 •

edited

Loading

greptile-apps Bot commented Feb 20, 2026 •

edited

Loading