Autocomplete Gives Multiline Autocomplete, but VSCode only shows One line and not the rest

[Gurneet1928]

Before submitting your bug report

• I've tried using the "Ask AI" feature on the Continue docs site to see if the docs have an answer
• I believe this is a bug. I'll try to join the Continue Discord for questions
• I'm not able to find an open issue that reports the same bug
• I've seen the troubleshooting guide on the Continue Docs

Relevant environment info

- OS: Windows10
- Continue version: 1.2.2 (all the way from 1.16.x)
- IDE version: VSCode 1.103
- Model: vLLM w/ Starcoder2 15B
- config:
  
models:
  - name: Autcomplete
    provider: vllm
    model: Starcoder2-15B
    apiBase: https://vllm-deployment-link/
    requestOptions:
      verifySsl: false
      timeout: 50000
    roles:
      - autocomplete
    autocompleteOptions:
      onlyMyCode: true
      useImports: true
      maxPromptTokens: 500

Description

We are currently using the Starcoder 2 15B model deployed with vLLM, integrated into Continue. When I begin receiving autocomplete suggestions (which typically take around 1-1.5 seconds to appear), only a single line of the suggestion is displayed. Interestingly, the Continue Console shows that the model is generating multiline suggestions. However, after accepting the first line by pressing 'TAB', the subsequent lines do not appear. If I press 'SPACE' or 'TAB' again, Continue sends a new autocomplete request, and the process repeats. I have experimented with various configurations using the same model, but the issue persists. Any guidance or solutions would be greatly appreciated.

Pasting some images as sample:

1. Getting first line suggestion (Good)

2. After Clicking TAB

3. After accepting suggestion with TAB. Console shows multiline response, but not VSCode

To reproduce

No response

Log output

This is an internal project for our company, and may contain some confidential data. So, pasting any logs is not possible for me.

[Mobious]

I had a similar issue. Autocomplete would show multiple lines of completion, but would truncate the model output anytime it encountered two or more sequential newlines. Adding transform: false under autocompleteOptions in the config fixed the issue for me.