inference
inference copied to clipboard
BUG: 400-Bad Request if no role:user message or 500-Internal Server Error
Describe the bug
API: POST /v1/chat/completions got an error if no role:user message
To Reproduce
the latest xinference version : v0.10.1
- POST /v1/chat/completions
{
"messages": [
{
"role": "system",
"content": "<任务> 识别用户输入的技术术语。请用{XXX} -> {XXX}的格式展示翻译前后的技术术语对应关系。\n<输入文本>\n今天天气\n<示例>\nTransformer -> Transformer\nToken -> Token\nZero Shot -> 零样本\nFew Shot -> 少样本\n<专有名词>"
}
],
"model": "qwen1.5-chat",
"max_tokens": 8192,
"stream": false,
"temperature": 0.7,
"top_p": 1,
"user": "46e3da56-7079-4845-9e5d-77e2262607dd"
}
returns: HTTP-500: Internal Server Error
- POST /v1/chat/completions
{
"messages": [
{
"role": "system",
"content": "<任务> 识别用户输入的技术术语。请用{XXX} -> {XXX}的格式展示翻译前后的技术术语对应关系。\n<输入文本>\n今天天气\n<示例>\nTransformer -> Transformer\nToken -> Token\nZero Shot -> 零样本\nFew Shot -> 少样本\n<专有名词>"
},
{
"role": "user",
"content": ""
}
],
"model": "qwen1.5-chat",
"max_tokens": 8192,
"stream": false,
"temperature": 0.7,
"top_p": 1,
"user": "46e3da56-7079-4845-9e5d-77e2262607dd"
}
returns HTTP:400 -
{
"detail": "Invalid input. Please specify the prompt."
}
Expected behavior
returns the correct data
Additional context
this version should put none empty user message, for example put an '\n' character to make it runs.
{
"messages": [
{
"role": "system",
"content": "<任务> 识别用户输入的技术术语。请用{XXX} -> {XXX}的格式展示翻译前后的技术术语对应关系。\n<输入文本>\n今天天气\n<示例>\nTransformer -> Transformer\nToken -> Token\nZero Shot -> 零样本\nFew Shot -> 少样本\n<专有名词>"
},
{
"role": "user",
"content": "\n"
}
],
"model": "qwen1.5-chat",
"max_tokens": 8192,
"stream": false,
"temperature": 0.7,
"top_p": 1,
"user": "46e3da56-7079-4845-9e5d-77e2262607dd"
}
openAI API don't need role user message:
Thank you for your report, we will see how to address this issue ASAP.
@codingl2k1 Do you have time to address this issue?
This issue is stale because it has been open for 7 days with no activity.
fixed.