[Bug] gemini-pro-vision 无响应
MuRo-J opened this issue · 15 comments
为了提高交流效率,我们设立了官方 QQ 群和 QQ 频道,如果你在使用或者搭建过程中遇到了任何问题,请先第一时间加群或者频道咨询解决,除非是可以稳定复现的 Bug 或者较为有创意的功能建议,否则请不要随意往 Issue 区发送低质无意义帖子。
反馈须知
请在下方中括号内输入 x 来表示你已经知晓相关内容。
- 我确认已经在 常见问题 中搜索了此次反馈的问题,没有找到解答;
- 我确认已经在 Issues 列表(包括已经 Close 的)中搜索了此次反馈的问题,没有找到解答。
- 我确认已经在 Vercel 使用教程 中搜索了此次反馈的问题,没有找到解答。
描述问题
请在此描述你遇到了什么问题。
作者大大,最新的版本部署服务器后,GeminiProVision没有响应,也没有报错提示,等到了空内容提示,但是实在很难用Curl测试,我只用python测试了下有模型权限,能得到结果:
如何复现
请告诉我们你是通过什么操作触发的该问题。
服务端网站日志:
[20/Feb/2024:12:53:28 +0800] "POST /api/google/v1beta/models/gemini-pro-vision:streamGenerateContent HTTP/2.0" 400 158 " Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/121.0.0.0 Safari/537.36"
功能日志:
0|muro-lea | [S3] S3链接没啥用..
0|muro-lea | [Server Config] using 1 of 1 api key
0|muro-lea | [Google Route] params {
0|muro-lea | path: [ 'v1beta', 'models', 'gemini-pro-vision:streamGenerateContent' ]
0|muro-lea | }
0|muro-lea | [Server Config] using 1 of 1 api key
0|muro-lea | [Proxy] v1beta/models/gemini-pro-vision:streamGenerateContent
0|muro-lea | [Base Url] https://generativelanguage.googleapis.com
0|muro-lea | [Server Config] using 1 of 1 api key
没什么有用的信息
测试图片(Google官方的测试用例):
Write a short, engaging blog post based on this picture.
截图
请在此提供控制台截图、屏幕截图或者服务端的 log 截图。
一些必要的信息
- 系统:[比如 windows 10/ macos 12/ linux / android 11 / ios 16]
- 浏览器: [比如 chrome, safari]
- 版本: [填写设置页面的版本号]
- 部署方式:[比如 vercel、docker 或者服务器部署]
Bot detected the issue body's language is not English, translate it automatically.
Title: [Bug]
In order to improve communication efficiency, we have set up an official QQ group and QQ channel. If you encounter any problems during use or construction, please join the group or channel for consultation as soon as possible, unless it is a bug that can be stably reproduced or More creative feature suggestions, otherwise please do not send low-quality and meaningless posts to the Issue area.
Feedback Instructions
Please enter x in the square brackets below to indicate that you already know the relevant content.
- I confirm that I have searched for the feedback issue in FAQ, and there is no find the solution;
- I confirm that I have searched the Issues list (including those that have been Closed) for this feedback issue, but no answer has been found.
- I confirm that I have searched for the feedback issue in Vercel usage tutorial. No answer found.
Describe the problem
Please describe what problem you encountered here.
Author: After deploying the latest version of the server, GeminiProVision did not respond, and there was no error message. I waited until I got an empty content prompt, but it was really difficult to test with Curl. I only used python to test it with model permissions, and I got the result:
How to reproduce
Please tell us what action triggered this issue.
Server website log:
[20/Feb/2024:12:53:28 +0800] "POST /api/google/v1beta/models/gemini-pro-vision:streamGenerateContent HTTP/2.0" 400 158 "Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/121.0.0.0 Safari/537.36"
Function log:
0|muro-lea | [S3] S3 link is useless..
0|muro-lea | [Server Config] using 1 of 1 api key
0|muro-lea | [Google Route] params {
0|muro-lea | path: [ 'v1beta', 'models', 'gemini-pro-vision:streamGenerateContent' ]
0|muro-lea | }
0|muro-lea | [Server Config] using 1 of 1 api key
0|muro-lea | [Proxy] v1beta/models/gemini-pro-vision:streamGenerateContent
0|muro-lea | [Base Url] https://generativelanguage.googleapis.com
0|muro-lea | [Server Config] using 1 of 1 api key
No useful information
It should be no problem to test S3, it is available on gpt-4-version
Test pictures (Google official test cases):
Write a short, engaging blog post based on this picture.
screenshot
Please provide console screenshots, screenshots, or server-side log screenshots here.
Some necessary information
- System: [such as windows 10/ macos 12/ linux / android 11 / ios 16]
- Browser: [such as chrome, safari]
- Version: [Fill in the version number of the settings page]
- Deployment method: [such as vercel, docker or server deployment]
找到了Curl 测试的方法,有点
request.json
curl https://generativelanguage.googleapis.com/v1beta/models/gemini-pro-vision:generateContent\?key\=**APIKEY**\
-H 'Content-Type: application/json'
-d @request.json 2> /dev/null | grep "text"
Bot detected the issue body's language is not English, translate it automatically.
I found a way to test Curl, which is a bit
request.json
curl https://generativelanguage.googleapis.com/v1beta/models/gemini-pro-vision:generateContent\?key\=**APIKEY**\
-H 'Content-Type: application/json'
-d @request.json 2> /dev/null | grep "text"
Bot detected the issue body's language is not English, translate it automatically.
I tried it on the main branch code deployed in vercel and it was normal, except for the summary summary where there is a problem.
Did you configure gemini on the ui or use environment variables?
It is best to provide the request information of the interface, such as the content of the body.
Bot detected the issue body's language is not English, translate it automatically.
disturb
I tried it on the main branch code deployed in vercel and it was normal, except for the summary summary where there is a problem. Did you configure gemini on the ui or use environment variables? It is best to provide the request information of the interface, such as the content of the body. ! [Image] (https://private- user-images.githubusercontent.com/14031260/306181511-d05539dd-4127-6bbf7d28c.png?jwt=eyjhbgci Oijiuzi1niisini5CCI6IKPXVCJ9.EYJPC3MIOJNAXRODWIUY29TIIWIYXVKICMF3LMDPDGH1YNVZXJJW50LMNVBSIMTLETUILCJLEHAIOJE3 Mdg0mtcznzqsim5izii6mTCWODQXNZA3NCWICGF0ACI6II8XNDAZMC8ZMDYXODYXDA1NTM5ZGQTYJVMTI3LWIXODKMDLKMJLNBUZZZ 9YLUFTEI1BBGDCML09QVDTNC1ITUFDLVNIQTI1NIZYLUFTEI1DCMVKZW50FSPUFWQ0UXTQTUFFLNFPBJGMJALMKZ1lyxn0ltel Mkzzmuyrmf3czrfcmvxdwvzdczyluftei1eyxrptiWMJIWVDA4MTC1BBXOTRXHWAXJLCZMDAMWC1BXJLPWM4OGNIMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMVIN NDCWNGU2ZGZHNHNWUXZWQ3YZU2ZDBKZWYXYTMZZJLHMTA5MDAXZTVHNJYNZUMWC1bbxotu2LnbmvHzgvycz1LKJMFJDG9YX2V5X2LKP TAMCMVWB19PZD0WIN0.fg-OZD_A1O4J6Y0CMOJBCZ1AK0BVDSIPMIPMTAHC0G)
I found out that I can't submit it for the second time... Author, please continue to ask the next question and you will know. You can only ask once.
Bot detected the issue body's language is not English, translate it automatically.
All the things I mentioned earlier are useless. They are here, and gemini-pro-vision requires that pictures must be brought when asking questions. Therefore, during general testing, if the first item is a greeting, the content will no longer be returned, and then if the second item is a greeting, the content will not be returned. When Erjo starts posting pictures, it prompts that multiple rounds of dialogue are not allowed. For some dishes, I am just curious about the reason for the test. This should be the current feature. Ridiculous.
应该是 gemini 的视觉模型不支持多轮对话,后面谷歌应该会支持,我先改下把这个异常抛出来。
Bot detected the issue body's language is not English, translate it automatically.
It should be that gemini's visual model does not support multi-turn dialogue. Google should support it later. I will change it first and throw this exception.
首次提问需要带图片也可以抛出来,需要不? 辛苦您啦~hh被google乐到了
Bot detected the issue body's language is not English, translate it automatically.
If you ask a question for the first time, you need to bring a picture, but you can also throw it out. Do you need it? Thank you for your hard work~ hh is very happy with Google
我在文档中标注一下吧,谷歌这个模型还在测试,添加太多逻辑去判断也不太合适。
Bot detected the issue body's language is not English, translate it automatically.
Let me mark it in the document. Google's model is still being tested. It is not appropriate to add too much logic to judge.