SGLang DeepSeek-V3-0324
I have been trying to run Deepseek-V3-0324 using instant clusters with 2 x (8 x H100s) and have so far been unsuccessful. I am trying to get the model to run multi-node + multi-gpu.
I have downloaded the model from Huggingface onto a persistent and attach the persistent volume to my instant cluster before launching. After launching, I then run the Pytorch demo script as presented in https://docs.runpod.io/instant-clusters/pytorch to make sure that the network is working (it does).
I then follow the instructions to get Deepseek-V3-0324 running according to: https://github.com/sgl-project/sglang/tree/main/benchmark/deepseek_v3
Instead of following the absolute default instructions and doing:
In its place, I run the following command on each node:
The issue is that this hangs. I check nvidia-smi to see the model loading and it only ever loads each GPU up to almost 1GB before it goes up no further.
Any help would be greatly appreciated.
Deploy with PyTorch | RunPod Documentation
Learn how to deploy an Instant Cluster and run a multi-node process using PyTorch.
GitHub
sglang/benchmark/deepseek_v3 at main · sgl-project/sglang
SGLang is a fast serving framework for large language models and vision language models. - sgl-project/sglang
130 Replies
Before discussing this problem
Does a 600+ parameter fp16 model fit in 16xH100s?
With reasonable context length?
Or is it fp8?
Idk
Anyways why are you doing tensor parallel over network
Unknown User•7mo ago
Message Not Public
Sign In & Join Server To View
Its tensor parallel tho
Over network you should pipeline parallel
Wow u r rich lol
Apparently ots fp8 in the official repo
So it should work
Tensor Parallelism - NADDOD Blog
Tensor parallelism alleviates memory issues in large-scale training. RoCE enables efficient communication for GPU tensor parallelism, accelerating computations.
Maybe becuz its cmd
Try bash -c 'command'
Unknown User•7mo ago
Message Not Public
Sign In & Join Server To View
Yeah, I just want to run this thing. I'm happy to spend on GPUs for a period of time to get it running. But I can't even get the basics to work unfortunately... has anyone seen any example on any infrastructure setup of this working multi-node / pipeline parallelism? If not on RunPod than anywhere else? It seems that no one has got this running anywhere.
Unknown User•7mo ago
Message Not Public
Sign In & Join Server To View
@frogsbody
Escalated To Zendesk
The thread has been escalated to Zendesk!
Unknown User•7mo ago
Message Not Public
Sign In & Join Server To View
Yes, I have exactly this issue.
They claim to support pipeline parallelism:
GitHub
sglang/benchmark/deepseek_v3 at main · sgl-project/sglang
SGLang is a fast serving framework for large language models and vision language models. - sgl-project/sglang
Unknown User•7mo ago
Message Not Public
Sign In & Join Server To View
nnodes = 2
It’s a torchrun argument that gets passed through to the equivalent in SGLang I believe.Unknown User•7mo ago
Message Not Public
Sign In & Join Server To View
@Jason have you tried vLLM with instant clusters? I believe the communication mechanism under the hood doesn't work with the way that Runpod sets up inter-node communication. I couldn't get it to work (this was a few weeks ago when it was still in beta though).
I wasn't sure where to open a ticket because I'm not sure where the error is really coming from... I think it's an SGLang issue but I wasn't clear.
are you still here?
ive never used sglang but
vllm pipeline parallel works well with multi node
even with not. that good network bandwidth
I'm still trying @riverfog7. Have you tested vLLM with Instant Cluster or do you have another solution where I can test multi-gpu in the cloud to run this?
@frogsbody but do you really need multigpu?
Yeah, I specifically need to test tensor parallelism and pipeline parallelism: 2 nodes of 8 x H100
i mean
you can host the same model in 1 node
you want deepseek v3 at fp8
right?
My requirements are to run Deepseek-V3-0324 over two nodes by whatever means - I just have to see that pipeline and tensor parallelism can work for the model
okay
is sglang required too?
It's less about actually using it - more about showing it can work
No, it can be anything
vLLM would be fine too
vllm should work
ive done it in the past
Have you got that working in Runpod Instant Clusters?
not with 2x8
but that doesnt matter
no in AWS
should work. with runpod tho
I had trouble running the basic torchrun script:
https://docs.runpod.io/instant-clusters/pytorch
Deploy with PyTorch | RunPod Documentation
Learn how to deploy an Instant Cluster and run a multi-node process using PyTorch.
It failed to run it when I tried with VLLM
Yeah, I tried AWS but they wouldn't give me any GPUs so now just trying with runpod
I will try vllm again
can you ping the other pod
Yeah I can ping it
with ip
It's something to do with Ray, which vLLM uses under the hood
do u use vllm docker image
or sth else?
I was using something else - but I can use that docker image
i suceededd with the docker image
soo
Thanks for letting me know, I'll try that out and let you know how it goes
@frogsbody https://docs.vllm.ai/en/latest/serving/distributed_serving.html
here's the multinode docs
GitHub
vllm/examples/online_serving/run_cluster.sh at main · vllm-project...
A high-throughput and memory-efficient inference and serving engine for LLMs - vllm-project/vllm
the cluster making script
docker run \
--entrypoint /bin/bash \
--network host \
--name node \
--shm-size 10.24g \
--gpus all \
-v "${PATH_TO_HF_HOME}:/root/.cache/huggingface" \
"${ADDITIONAL_ARGS[@]}" \
"${DOCKER_IMAGE}" -c "${RAY_START_CMD}"
this is the docker run command so
you can modify this and run it
Where do I run this? When I create the pod?
no so waht the docs says is
you have two physical machines
then you create a ray container on both physical machines and make a cluster.
but in your case you have no access to physical machines
Yeah, the issue I had is that runpod doesn't have that
I have to work within those bounds
I don't have AWS or anything to work with
so you should translate the docker run command to runpod's template
docker run --entrypoint /bin/bash --network host --name node --shm-size 10.24g --gpus all -v /path/to/the/huggingface/home/in/this/node:/root/.cache/huggingface -e VLLM_HOST_IP=ip_of_this_node vllm/vllm-openai -c ray start --block --address=ip_of_head_node:6379
Yeah, this was my next idea - I just have to figure out how to modify runpod to work with this since "docker" can't be run inside of a pod once I start it
It has to be part of a template or something. I'm pretty new to this side of Runpod.
should be
image name: vllm/vllm-openai
CMD: python3 -m vllm.entrypoints.openai.api_server -c ray start --block --address=ip_of_head_node:6379
mount nw volume to ~/.cache/
for the worker
image name: vllm/vllm-openai
CMD: python3 -m vllm.entrypoints.openai.api_server -c ray start --block --head --port=6379
env: VLLM_HOST_IP=ip_of_this_node
for the head
Are you starting these as two separate pods or using Instant Cluster?
In this case it looks like you're using two separate pods with global networking or something
can you apply diff images
Not with instant cluster
for the two pods in clusters?
or diff setting at least
Doesn't look likeit
lol
should we write a script?
its solvable
Yeah I'd love to lol, been trying to run DeepSeek V3 across two nodes for a while now
vllm serve /path/to/the/model/in/the/container \
--tensor-parallel-size 8 \
--pipeline-parallel-size 2
I was thinking this should just work
If I spin up a cluster
And go into each node and run this
And make sure to pass in the right information for the host...
But then it fails because of Ray
yeah but we need a script
to build the ray cluster first
it should run INSIDE a ray cluster
Yeah, I feel like that's outside the default scope of instant clusters. Are you suggesting we set up a ray cluster inside of our non-ray cluster?
that's what i meant
😄
the writing a script part was for that
That would be cool... solve a lot of problems lol
ill try with global networking first
just to see if it works
I'll try form a ray cluster in Instant Cluster again
try
python3 -m vllm.entrypoints.openai.api_server -c ray start --block --address=ip_of_head_node:6379
this inside a vllm container
Sure will try now
so you have seperate ssh access to
the two nodes right?
Yes
good
Will send sc in a sec
api_server.py: error: argument --block-size: expected one argument
isnt it --block?
I copied what you sent and that's what it gave me
In my case:
python3 -m vllm.entrypoints.openai.api_server -c ray start --block --address=ip_of_head_node:6379
this?
Yeah, that's what I did ^
hnm
vllm/vllm-openai
the image is this
Yep
i think il ltest in mine first
wait a sec
oh
it was
just
ray start --block --address=10.65.0.2:6379
or bash -c "ray start ...."
Yeah, I'm doing that right now actually
didnt see the --entrypoint /bin/bash
But can't get the worker to connect
xD
Unknown User•7mo ago
Message Not Public
Sign In & Join Server To View
wdym?
Unknown User•7mo ago
Message Not Public
Sign In & Join Server To View
Tried 6379 and couldn't get that to work
Unknown User•7mo ago
Message Not Public
Sign In & Join Server To View
So tried a port I knew was exposed 29400 since I can ping between the nodes with that
ray start --block --address=10.65.0.2:6379
this?
This works
is adding --block
make a diff?
But connecting from worker doesn't
Let me try with block
ray start --block --head --port=6379
for the head
ray start --block --address=10.65.0.2:6379
for the worker
maybe it doesnt work bc u already started a vllm process in the start command
I actually haven't started anything in this one
I am not using the vLLM image this time. I started a new cluster that doesn't have vLLM. I pip installed it.
Regardless, Ray should work independently
yeah
same thought
but had nothing to blame other than that
check
ufw
just in case
What is UFW?
and other firewalls too
ubuntu firewall
Ah okay let me check
that was the problem in my last attempt
@frogsbody i have one question
Trying to check but have to install packages
@riverfog7 yeah what's up?
why does the first pic say 172.xx
but second pic says10.60.sth
That's the "master address"
master addr?
Overview | RunPod Documentation
Instant Clusters enable high-performance computing across multiple GPUs with high-speed networking capabilities.
NODE_ADDR is the address of the individual node
That's the one that ray uses
vLLM uses Ray under the hood and it isn't playing nicely
That's why I was hoping SGLang would work since it uses pytorch
But then we have that weird bug where it hangs model loading at 1% lol
I suspect that it's actually only loading the pytorch stuff
And never actually loads any of the weights in
maybe it binds to the wrong nic?
We use eth1 I think here
and recieves from the public ip
but not from private ip
Issue is that I'm not sure if that's something we can even fix under the hood with vLLM... I just don't know enough about how vLLM works
if ray works vllm works
And vLLM uses that same ray cluster?
yeah
Hmm
you can just use the ray cluster
as one computer
vllm does the finicky things by itself
I actually can't even ping between each machine now
maybe ray start --block --head --port 6379 --node-ip-address 10.65.0.2
in the head?
wut
can u just ping it
ping 10.sth
ufw status?
Interesting
My environment is messed up now
I can't run the default torch script here
Deploy with PyTorch | RunPod Documentation
Learn how to deploy an Instant Cluster and run a multi-node process using PyTorch.
lol
So I messed something up with whatever we tried
hmm
maybe start with a fresh pytorch image
I'
and install everything (ray and vllm)
I'm going to have to refund this account lol, it won't let me start another pod
Not enough money in the account lol
I may sleep for a bit and get back to this, interseting problem to solve
great
the community says
--node-ip-address providing this should make it bind to the proper address
so maybe try that next time
in the head node
Yeah will do, I'll post any findings here