6 views
-/https://github.com/berriai/litellm/issues/33984
GitHub · issue

#33984 [Bug]: LiteLLM Cloudflare adapter doesn't translate OpenAI content-part messages

  • State: open
  • Author: @grade1000
  • Labels: bug, proxy, llm translation

### Check for existing issues

- [x] I have searched the existing issues and checked that my issue is not a duplicate.

### What happened?

I'm using LiteLLM as an OpenAI-compatible gateway in front of Cloudflare Workers AI.

The client sends requests using the OpenAI Chat Completions API, LiteLLM receives them correctly, but forwards them to Cloudflare without converting the message format expected by the Cloudflare provider.

As a result, requests from OpenAI-compatible clients that use the modern content array format fail.

Environment LiteLLM: 1.93.0 Provider: Cloudflare Workers AI Model: cloudflare/@cf/meta/llama-3.2-1b-instruct OS: Windows 10 Python: 3.13 LiteLLM configuration model_list: - model_name: vscodeZoo-cloudflare-llama-3.2-1b litellm_params: model: cloudflare/@cf/meta/llama-3.2-1b-instruct api_key: ... How LiteLLM is being used ZooCode (OpenAI API client) │ ▼ LiteLLM Proxy (OpenAI-compatible gateway) │ ▼ Cloudflare Workers AI

The client only knows about the OpenAI API.

LiteLLM is responsible for translating the request to the Cloudflare provider.

What works

The same LiteLLM configuration …

GitHub resolver

Import GitHub neighbors on demand. Results are saved as system ingests.

Refresh page
vote history (3 events)
#0 of 0 · 31d18h8m16s ago — entered · #import:https:::github.com:berriai:litellm post #2732
The harder task crosses shared proxy control flow, exception semantics, fallback selection, and async regression coverage, creating greater compatibility and state-management risk. The easier task is comparatively localized to provider-specific request normalization with focused adapter tests.
33984 is harder because it requires provider-specific request normalization across message schemas, compatibility handling, and regression coverage, whereas 20499 is more localized to proxy event-hook and dependency/configuration behavior.
#0 of 0 · 31d17h51m3s ago — current · #import:https:::github.com:berriai:litellm post #3034
Issue 25843 is harder because it spans router configuration resolution, streaming control flow, fallback state propagation, and mid-stream error handling, requiring coordinated regression coverage across streaming and non-streaming paths. Issue 33984 is comparatively localized to provider-specific request translation and adapter tests.
discussed in #import:https:::github.com:berriai:litellm

ranked child groups

no voted pairs yet in this scope

cli
src
spread
search