[Bug]: The OpenAI-compatible endpoint does not work with sglang

[jvstme]

Steps to reproduce

Apply this configuration:

type: service
name: deepseek-r1

image: lmsysorg/sglang:latest
env:
  - MODEL_ID=deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B
commands:
  - python3 -m sglang.launch_server
    --model-path $MODEL_ID
    --port 8000
    --trust-remote-code

port: 8000
# Register the model

model:
  name: deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B
  type: chat
  format: openai


# Uncomment to cache downloaded models
#volumes:
#  - /root/.cache/huggingface/hub:/root/.cache/huggingface/hub

# Disable authorization
auth: false

resources:
    gpu: 24GB

Try requesting the OpenAI-compatible endpoint.

Actual behaviour

curl http://localhost:3000/proxy/models/bihan/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer <Token>" \
  -d '{
  "model": "deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B",
  "messages": [
    {
      "role": "user",
      "content": "Hello world"
    }
  ],
  "stream": true,
  "max_tokens": 512
}'

{"detail":"Invalid chunk in model stream: 1 validation error for ChatCompletionsChunkResponse\nchoices -> 0 -> finish_reason\n  unexpected value; permitted: 'stop', 'length', 'tool_calls', 'eos_token' (type=value_error.const; given=; permitted=('stop', 'length', 'tool_calls', 'eos_token'))"}

Expected behaviour

The request returns chat completions chunks.

dstack version

0.18.37

Server logs

Additional information

No response