解决查询命中缓存时流式响应未遵循Ollma规范的问题
- rag返回结果未字符串时,响应分两次发送 - 第一次发送查询内容 - 第二次发送统计信息
This commit is contained in:
parent
e978a15593
commit
9632a8f0dc
2 changed files with 19 additions and 4 deletions
|
|
@ -709,16 +709,31 @@ def create_app(args):
|
|||
try:
|
||||
# 确保 response 是异步生成器
|
||||
if isinstance(response, str):
|
||||
# 如果是字符串,作为单个完整响应发送
|
||||
# 如果是字符串,分两次发送
|
||||
# 第一次发送查询内容
|
||||
data = {
|
||||
"model": LIGHTRAG_MODEL,
|
||||
"created_at": LIGHTRAG_CREATED_AT,
|
||||
"message": {
|
||||
"role": "assistant",
|
||||
"role": "assistant",
|
||||
"content": response,
|
||||
"images": None
|
||||
},
|
||||
"done": True
|
||||
"done": False
|
||||
}
|
||||
yield f"{json.dumps(data, ensure_ascii=False)}\n"
|
||||
|
||||
# 第二次发送统计信息
|
||||
data = {
|
||||
"model": LIGHTRAG_MODEL,
|
||||
"created_at": LIGHTRAG_CREATED_AT,
|
||||
"done": True,
|
||||
"total_duration": 1,
|
||||
"load_duration": 1,
|
||||
"prompt_eval_count": 999,
|
||||
"prompt_eval_duration": 1,
|
||||
"eval_count": 999,
|
||||
"eval_duration": 1
|
||||
}
|
||||
yield f"{json.dumps(data, ensure_ascii=False)}\n"
|
||||
else:
|
||||
|
|
|
|||
|
|
@ -119,7 +119,7 @@ DEFAULT_CONFIG = {
|
|||
"test_cases": {
|
||||
"basic": {
|
||||
"query": "孙悟空",
|
||||
"stream_query": "孙悟空有什么法力,性格特征是什么"
|
||||
"stream_query": "孙悟空"
|
||||
}
|
||||
}
|
||||
}
|
||||
|
|
|
|||
Loading…
Add table
Reference in a new issue