You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
对于短文章,一万字左右,点击知识图谱-自动切片,可以顺利构造知识图片并进行召回,让ai回答问题;
但是对于长文本【三体第一部】,首先是自动切片不能用,会显示报错:
CheckErrorInfo.**: Error code: 400 - {'error': {'message': "This model's maximum context length is 16385 tokens. However, your messages resulted in 267142 tokens. Please reduce the length of the messages.", 'type': 'invalid_request_error', 'param': 'messages', 'code': 'context_length_exceeded'}},即令牌数超了,
因此使用手动切片chunk size,每512token切一次,重叠50token,然后就开始处理,处理的过程中简单看了一下终端的输出,会经历切片、向量化等一系列操作,中间跳的太快看不清楚,最后出现一系列进度条
2025-02-24 16:30:27 deqing-gpu-249 dbgpt.storage.vector_store.chroma_store[3847629] INFO ChromaStore similar search with scores
Batches: 100%|███████████████████████████████████████████████████████████████████████████████| 1/1 [00:00<00:00, 20.40it/s]
但是会有一些是进度0,然后就开始报
deqing-gpu-249 dbgpt.util.api_utils[3847629] WARNING Health check failed for http://127.0.0.1:5670, error: HTTPConnectionPool(host='127.0.0.1', port=5670): Read timed out. (read timeout=10)
然后我估计是有些进度条一直是0且超时,然后程序就一直重复一直显示超时,搞到最后都不响应了
What you expected to happen
有两类进度条好像都会弹出超时的提示,因为切片是并行逻辑,所以终端弹出的内容不是很按照顺序,比较混乱,另一类是
2025-02-24 16:33:39 deqing-gpu-249 dbgpt.model.proxy.llms.chatgpt[3847629] INFO Send request to openai(1.61.1), payload: {'stream': True, 'model': 'gpt-35-turbo'}
哦对了,在这里面我还看见过一个问题,终端会显示:
Expected str but got datetime with value datetime.datetime(2025, 2, 24, 11, 10, 6, 371925) - serialized value may not be as expected
这个是什么原因,能怎么解决吗?一直有见到这个不是报错的报错
Are you willing to submit PR?
Yes I am willing to submit a PR!
The text was updated successfully, but these errors were encountered:
此外我还想知道一下,终端里面显示的:
Expected str but got datetime with value datetime.datetime(2025, 2, 24, 11, 10, 6, 371925) - serialized value may not be as expected
是什么,会不会有什么影响?
Search before asking
Operating system information
Linux
Python version information
3.10
DB-GPT version
main
Related scenes
Installation Information
Installation From Source
Docker Installation
Docker Compose Installation
Cluster Installation
AutoDL Image
Other
Device information
GPU:16G
Models information
LLM:gpt-35-turbo【微软代理】
embedding:DB-GPT/models/text2vec-large-chinese
What happened
对于短文章,一万字左右,点击知识图谱-自动切片,可以顺利构造知识图片并进行召回,让ai回答问题;
但是对于长文本【三体第一部】,首先是自动切片不能用,会显示报错:
CheckErrorInfo.**: Error code: 400 - {'error': {'message': "This model's maximum context length is 16385 tokens. However, your messages resulted in 267142 tokens. Please reduce the length of the messages.", 'type': 'invalid_request_error', 'param': 'messages', 'code': 'context_length_exceeded'}},即令牌数超了,
因此使用手动切片chunk size,每512token切一次,重叠50token,然后就开始处理,处理的过程中简单看了一下终端的输出,会经历切片、向量化等一系列操作,中间跳的太快看不清楚,最后出现一系列进度条
2025-02-24 16:30:27 deqing-gpu-249 dbgpt.storage.vector_store.chroma_store[3847629] INFO ChromaStore similar search with scores
Batches: 100%|███████████████████████████████████████████████████████████████████████████████| 1/1 [00:00<00:00, 20.40it/s]
但是会有一些是进度0,然后就开始报
deqing-gpu-249 dbgpt.util.api_utils[3847629] WARNING Health check failed for http://127.0.0.1:5670, error: HTTPConnectionPool(host='127.0.0.1', port=5670): Read timed out. (read timeout=10)
然后我估计是有些进度条一直是0且超时,然后程序就一直重复一直显示超时,搞到最后都不响应了
What you expected to happen
有两类进度条好像都会弹出超时的提示,因为切片是并行逻辑,所以终端弹出的内容不是很按照顺序,比较混乱,另一类是
2025-02-24 16:33:39 deqing-gpu-249 dbgpt.model.proxy.llms.chatgpt[3847629] INFO Send request to openai(1.61.1), payload: {'stream': True, 'model': 'gpt-35-turbo'}
后面会显示进度条,然后也会报超时。
然后我估计是有些进度条一直是0且超时,然后程序就一直重复一直显示超时,搞到最后都不响应了
How to reproduce
复现用个超长文本不知道能不能复现,我可以提供一下我的样本
santi.md
Additional context
哦对了,在这里面我还看见过一个问题,终端会显示:
Expected
str
but gotdatetime
with valuedatetime.datetime(2025, 2, 24, 11, 10, 6, 371925)
- serialized value may not be as expected这个是什么原因,能怎么解决吗?一直有见到这个不是报错的报错
Are you willing to submit PR?
The text was updated successfully, but these errors were encountered: