BUG: Could not start v0.11.0 from Docker Compose

[Minamiyama]

No description provided.

[XiaoCC]

我也是最新版的镜像起不来

[yanmao2023]

me too.+1

[ChengjieLi28]

@Minamiyama @yanmao2023 @XiaoCC Could you please help me test this:
Build a new image based on our offical image:

FROM xprobe/xinference:v0.11.0

RUN pip install torchvision==0.17.1

And then test it?

In my own machine with two GPUs, I can use xinference it normally with above method.

[Minamiyama]

@Minamiyama @yanmao2023 @XiaoCC Could you please help me test this: Build a new image based on our offical image:

FROM xprobe/xinference:v0.11.0

RUN pip install torchvision==0.17.1

And then test it?

In my own machine with two GPUs, I can use xinference it normally with above method.

It seems not work for me

[ChengjieLi28]

@Minamiyama @yanmao2023 @XiaoCC Could you please help me test this: Build a new image based on our offical image:

FROM xprobe/xinference:v0.11.0

RUN pip install torchvision==0.17.1

And then test it?
In my own machine with two GPUs, I can use xinference it normally with above method.

It seems not work for me

Paste your error stack. And your related commands.

[Minamiyama]

@Minamiyama @yanmao2023 @XiaoCC Could you please help me test this: Build a new image based on our offical image:

FROM xprobe/xinference:v0.11.0

RUN pip install torchvision==0.17.1

And then test it?
In my own machine with two GPUs, I can use xinference it normally with above method.

It seems not work for me

Paste your error stack. And your related commands.

[ChengjieLi28]

@Minamiyama @yanmao2023 @XiaoCC Could you please help me test this: Build a new image based on our offical image:

FROM xprobe/xinference:v0.11.0

RUN pip install torchvision==0.17.1

And then test it?
In my own machine with two GPUs, I can use xinference it normally with above method.

It seems not work for me

Paste your error stack. And your related commands.

What's the error here? You can also add --log-level debug in your entrypoint command.
Could you just test:

1. build the new image
2. run it

docker run -p 9997:9997 --gpus all <the new image> xinference-local --log-level debug -H 0.0.0.0

3. And then use it

[Minamiyama]

@Minamiyama @yanmao2023 @XiaoCC Could you please help me test this: Build a new image based on our offical image:

FROM xprobe/xinference:v0.11.0

RUN pip install torchvision==0.17.1

And then test it?
In my own machine with two GPUs, I can use xinference it normally with above method.

It seems not work for me

Paste your error stack. And your related commands.

What's the error here? You can also add --log-level debug in your entrypoint command. Could you just test:

1. build the new image
2. run it

docker run -p 9997:9997 --gpus all <the new image> xinference-local --log-level debug -H 0.0.0.0

3. And then use it

is this message useful?

[ChengjieLi28]

@Minamiyama @yanmao2023 @XiaoCC Could you please help me test this: Build a new image based on our offical image:

FROM xprobe/xinference:v0.11.0

RUN pip install torchvision==0.17.1

And then test it?
In my own machine with two GPUs, I can use xinference it normally with above method.

It seems not work for me

Paste your error stack. And your related commands.

What's the error here? You can also add --log-level debug in your entrypoint command. Could you just test:

1. build the new image
2. run it

docker run -p 9997:9997 --gpus all <the new image> xinference-local --log-level debug -H 0.0.0.0

3. And then use it

is this message useful?

What's this? It seems no relation with xinference and it may be the issue with your cuda environment. The docker image uses pytorch image as the base image. You can try that whether you can use this image directly:

pytorch/pytorch:2.1.2-cuda12.1-cudnn8-devel

[Minamiyama]

@Minamiyama @yanmao2023 @XiaoCC Could you please help me test this: Build a new image based on our offical image:

FROM xprobe/xinference:v0.11.0

RUN pip install torchvision==0.17.1

And then test it?
In my own machine with two GPUs, I can use xinference it normally with above method.

It seems not work for me

Paste your error stack. And your related commands.

What's the error here? You can also add --log-level debug in your entrypoint command. Could you just test:

1. build the new image
2. run it

docker run -p 9997:9997 --gpus all <the new image> xinference-local --log-level debug -H 0.0.0.0

3. And then use it

is this message useful?

What's this? It seems no relation with xinference and it may be the issue with your cuda environment. The docker image uses pytorch image as the base image. You can try that whether you can use this image directly:
pytorch/pytorch:2.1.2-cuda12.1-cudnn8-devel

cause by adding --log-level, mayby it's my wrong usage

[Minamiyama]

@Minamiyama @yanmao2023 @XiaoCC Could you please help me test this: Build a new image based on our offical image:

FROM xprobe/xinference:v0.11.0

RUN pip install torchvision==0.17.1

And then test it?
In my own machine with two GPUs, I can use xinference it normally with above method.

It seems not work for me

Paste your error stack. And your related commands.

What's the error here? You can also add --log-level debug in your entrypoint command. Could you just test:

1. build the new image
2. run it

docker run -p 9997:9997 --gpus all <the new image> xinference-local --log-level debug -H 0.0.0.0

3. And then use it

is this message useful?

What's this? It seems no relation with xinference and it may be the issue with your cuda environment. The docker image uses pytorch image as the base image. You can try that whether you can use this image directly:
pytorch/pytorch:2.1.2-cuda12.1-cudnn8-devel

cause by adding --log-level, mayby it's my wrong usage

nothing new shown, and auto shut down as well

[ChengjieLi28]

pytorch/pytorch:2.1.2-cuda12.1-cudnn8-devel

Just run

docker run --gpus all pytorch/pytorch:2.1.2-cuda12.1-cudnn8-devel tail -f /dev/null

still auto shutdown?

[ChengjieLi28]

@Minamiyama @yanmao2023 @XiaoCC Could you please help me test this: Build a new image based on our offical image:

FROM xprobe/xinference:v0.11.0

RUN pip install torchvision==0.17.1

And then test it?
In my own machine with two GPUs, I can use xinference it normally with above method.

It seems not work for me

Paste your error stack. And your related commands.

What's the error here? You can also add --log-level debug in your entrypoint command. Could you just test:

1. build the new image
2. run it

docker run -p 9997:9997 --gpus all <the new image> xinference-local --log-level debug -H 0.0.0.0

3. And then use it

is this message useful?

What's this? It seems no relation with xinference and it may be the issue with your cuda environment. The docker image uses pytorch image as the base image. You can try that whether you can use this image directly:
pytorch/pytorch:2.1.2-cuda12.1-cudnn8-devel

cause by adding --log-level, mayby it's my wrong usage

nothing new shown, and auto shut down as well

The host machine is windows OS. May cannot use 0.0.0.0. I haven't tried windows. Remove -H 0.0.0.0 and try again.

[Minamiyama]

pytorch/pytorch:2.1.2-cuda12.1-cudnn8-devel

Just run

docker run --gpus all pytorch/pytorch:2.1.2-cuda12.1-cudnn8-devel tail -f /dev/null

still auto shutdown?

running normally

[ChengjieLi28]

pytorch/pytorch:2.1.2-cuda12.1-cudnn8-devel

Just run

docker run --gpus all pytorch/pytorch:2.1.2-cuda12.1-cudnn8-devel tail -f /dev/null

still auto shutdown?

running normally

Cannot reproduce.

docker pull xprobe/xinference:nightly-bug_torchvision_version

This image is built by #1485 . And I can use it normally on my ubuntu machine.

[Minamiyama]

pytorch/pytorch:2.1.2-cuda12.1-cudnn8-devel

Just run

docker run --gpus all pytorch/pytorch:2.1.2-cuda12.1-cudnn8-devel tail -f /dev/null

still auto shutdown?

running normally

Cannot reproduce.

docker pull xprobe/xinference:nightly-bug_torchvision_version

This image is built by #1485 . And I can use it normally on my ubuntu machine.

failed as well

[ChengjieLi28]

@Minamiyama Try this image:

docker pull xprobe/xinference:nightly-docker_crash_due_to_llama

[Minamiyama]

0.11.1 is ok