Change to openvinotoolkit/model_server as base image for intel GPU #7459
Description
Is your feature request related to a problem? Please describe.
I'm starting to play with locaio in my new server (Intel Core Ultra 5 235 and and intel arc b580) ... with CPU images its okey, but with integ gpu (now in final 2025 with debian 13 seems not work).
I follow (with difficult) the instalation of the intel repositories (need still write down the steps to be clear for me) but the case is... if i use this image https://github.com/openvinotoolkit/model_server/blob/main/Dockerfile.ubuntu and the steps in openvino documentation i can see how LLM works and use NPU and GPU witn nvtop ... but with images of localio give me an error ...
pi-1 | 9:15AM INF BackendLoader starting backend=diffusers modelID=stablediffusion o.model=Lykon/dreamshaper-8 api-1 | 9:15AM DBG Loading model in memory from file: /models/Lykon/dreamshaper-8 api-1 | 9:15AM DBG Loading Model stablediffusion with gRPC (file: /models/Lykon/dreamshaper-8) (backend: diffusers): {backendString:diffusers model:Lykon/dreamshaper-8 modelID:stablediffusion context:{emptyCtx:{}} gRPCOptions:0xc000762908 externalBackends:map[] grpcAttempts:20 grpcAttemptsDelay:2 parallelRequests:false} api-1 | 9:15AM DBG Loading external backend: /backends/intel-diffusers/run.sh api-1 | 9:15AM DBG external backend is file: &{name:run.sh size:410 mode:493 modTime:{wall:0 ext:63899786926 loc:0x4c9f5a0} sys:{Dev:66306 Ino:5527396 Nlink:1 Mode:33261 Uid:0 Gid:0 X__pad0:0 Rdev:0 Size:410 Blksize:4096 Blocks:8 Atim:{Sec:1764190126 Nsec:0} Mtim:{Sec:1764190126 Nsec:0} Ctim:{Sec:1765098815 Nsec:467664192} X__unused:[0 0 0]}} api-1 | 9:15AM DBG Loading GRPC Process: /backends/intel-diffusers/run.sh api-1 | 9:15AM DBG GRPC Service for stablediffusion will be running at: '127.0.0.1:33649' api-1 | 9:15AM DBG GRPC Service state dir: /tmp/go-processmanager3362510419 api-1 | 9:15AM DBG GRPC Service Started api-1 | 9:15AM DBG Wait for the service to start up api-1 | 9:15AM DBG Options: ContextSize:1024 Seed:1467615107 NBatch:512 F16Memory:true MMap:true NGPULayers:9999999 Threads:14 PipelineType:"StableDiffusionPipeline" SchedulerType:"k_dpmpp_2m" CUDA:true FlashAttention:"auto" api-1 | 9:15AM DBG GRPC(stablediffusion-127.0.0.1:33649): stdout Initializing libbackend for intel-diffusers api-1 | 9:15AM DBG GRPC(stablediffusion-127.0.0.1:33649): stdout Using portable Python api-1 | 9:15AM DBG GRPC(stablediffusion-127.0.0.1:33649): stderr Traceback (most recent call last): api-1 | 9:15AM DBG GRPC(stablediffusion-127.0.0.1:33649): stderr File "/backends/intel-diffusers/backend.py", line 13, in <module> api-1 | 9:15AM DBG GRPC(stablediffusion-127.0.0.1:33649): stderr import torch api-1 | 9:15AM DBG GRPC(stablediffusion-127.0.0.1:33649): stderr File "/backends/intel-diffusers/venv/lib/python3.10/site-packages/torch/__init__.py", line 367, in <module> api-1 | 9:15AM DBG GRPC(stablediffusion-127.0.0.1:33649): stderr from torch._C import * # noqa: F403 api-1 | 9:15AM DBG GRPC(stablediffusion-127.0.0.1:33649): stderr ImportError: libpti_view.so.0.10: cannot open shared object file: No such file or directory
Describe the solution you'd like
I'm not an expert :) but, seems that the images for localio use ubuntu22, could be good idea make this images from openvinotoolkit/model_server that seems more near to ubuntu24 an in this way more near to intel (openvino 2025 libraries) ...
Describe alternatives you've considered
Maybe follow the steps in model_server o upgrade actual docker images with still ubuntu base, but maybe more usefull the initial aproach.
Additional context
I'll try for my self.. .and i can do a PR with this if it's interesening. All the proyect is great and sure work good with nvidia but for a cheaper intel arc seems stil need some tunning :)
Thanks for this great proyect.