oobabooga completion path ignores `custom_prompt_dict` (parameter is never used and not forwarded from main)

[dario-fumarola]

What happened

The oobabooga completion integration exposes a custom_prompt_dict parameter, but the current call chain never applies it.

Evidence

1. litellm/main.py (elif custom_llm_provider == "oobabooga": branch)

• Calls oobabooga.completion(...) without passing custom_prompt_dict.

2. litellm/llms/oobabooga/chat/oobabooga.py

• Function signature includes custom_prompt_dict={}.
• Inside the function, custom_prompt_dict is never referenced.
• transform_request(...) is called with model/messages/optional_params/litellm_params/headers, but no custom prompt handling exists in this path.

3. Contrast with other providers

• litellm/llms/vllm/completion/handler.py explicitly checks and applies custom_prompt_dict:
  • if model in custom_prompt_dict: ... prompt = custom_prompt(...)

Why this matters

Users can reasonably expect registered custom prompt templates to affect provider formatting consistently. For oobabooga, the API surface suggests support, but behavior is currently a no-op.

Expected behavior

For oobabooga requests, custom_prompt_dict should either:

• be forwarded and applied in prompt construction, or
• be removed from the signature/documented as unsupported to avoid misleading behavior.

Suggested fix

In main.py, forward custom_prompt_dict in the oobabooga provider call, and implement usage in the oobabooga request/prompt transformation path (similar to vllm/replicate/sagemaker-style integrations that apply custom_prompt).