fix(ui): keep completion-mode models in the playground chat dropdown (backport to rc/1.98.0)

[yuneng-berri]

TLDR

Problem this solves:

• Completion-mode models vanish from the playground model dropdown
• rc/1.98.0 shipped that regression, so the line needs the fix
• The line's committed UI bundle predates the fix

How it solves it:

• Backports fix(ui): keep completion-mode models in the playground chat dropdown #37954 onto rc/1.98.0, verbatim
• Rebuilds the shipped Next.js bundle from this branch's tip

User Flow

Before: an admin who registered a legacy text-completion deployment cannot select it in the playground, even though the proxy routes it fine

1. They register davinci-002 and confirm it is live: POST https://litellm-domain/v1/chat/completions with "model": "davinci-002" returns 200 with a normal completion
2. They open https://litellm-domain/ui/?page=llm-playground
3. The endpoint selector is on /chat/completions
4. They open the model dropdown and davinci-002 is absent, while their chat models are all listed
5. They switch the endpoint to /responses and it is absent there too, so there is no way to try the model from the UI

After: the same deployment shows up and is selectable

1. They register davinci-002 and confirm it is live: POST https://litellm-domain/v1/chat/completions with "model": "davinci-002" returns 200 with a normal completion
2. They open https://litellm-domain/ui/?page=llm-playground
3. The endpoint selector is on /chat/completions
4. They open the model dropdown and davinci-002 is listed alongside their chat models
5. They switch the endpoint to /responses and it is listed there too, and sending a prompt returns a completion
6. Speech and transcription endpoints still do not list it, which is correct

Relevant issues

Backport of #37954

Linear ticket

Pre-Submission checklist

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

• I have added meaningful tests
• My PR passes all CI/CD checks (e.g., lint, format, unit tests)
• My PR's scope is as isolated as possible; it only solves 1 specific problem
• I have received a Greptile Confidence Score of at least 4/5 before requesting a maintainer review (Greptile reviews automatically once the PR is opened; only comment @greptileai to re-request a review after pushing changes)

Screenshots / Proof of Fix

Shared setup, run once against a proxy started from this branch:

cat > /tmp/completion_mode.yaml <<'EOF'
model_list:
  - model_name: davinci-002
    litellm_params:
      model: text-completion-openai/davinci-002
      api_key: os.environ/OPENAI_API_KEY
    model_info:
      mode: completion
  - model_name: gpt-4o
    litellm_params:
      model: openai/gpt-4o
      api_key: os.environ/OPENAI_API_KEY
EOF

python litellm/proxy/proxy_cli.py --config /tmp/completion_mode.yaml --detailed_debug 2>&1 | tee litellm.log

Confirm the deployment really routes, so the dropdown is the only thing at fault:

curl -s http://localhost:4000/v1/chat/completions -H "Authorization: Bearer sk-1234" -H "Content-Type: application/json" -d '{"model":"davinci-002","messages":[{"role":"user","content":"say hi"}]}'

Before (bc6e7df)

Check out v1.98.0-rc.1, boot the proxy against the config above, then:

1. Open http://localhost:4000/ui/?page=llm-playground
2. Leave the endpoint selector on /chat/completions
3. Open the model dropdown, screenshot it: davinci-002 is missing, gpt-4o is present
4. Switch the endpoint selector to /responses, screenshot the dropdown: davinci-002 is missing there too

After (2f39f5c)

Check out this branch, boot the proxy against the same config, then:

1. Open http://localhost:4000/ui/?page=llm-playground
2. Leave the endpoint selector on /chat/completions
3. Open the model dropdown, screenshot it: davinci-002 is now listed next to gpt-4o
4. Select davinci-002, send "say hi", screenshot the reply
5. Switch the endpoint selector to /responses, screenshot the dropdown: davinci-002 is listed
6. Switch to /audio/speech, screenshot the dropdown: davinci-002 is correctly absent

Screenshots to follow on this PR once the runs above are captured

Type

🐛 Bug Fix

Caveats (if any)

• Base is rc/1.98.0, the release line, not staging
• Second commit is generated bundle output, no hand edits

Final Attestation

• The tests check the right things, including the edge cases, and regressions in the respective real-world customer use-cases are not possible after this PR

[greptile-apps]

Too many files changed for review (445 files, 100 file limit).

Bypass the limit by tagging @greptile-apps to review.