Movinet fails in GPU mode

从发生所报告故障的 src/movinet/MovinetModel.js 开始，并将其与能够成功检测 GPU 的 bin/node src/test_gputensorflow.js 进行比较。检查 TensorFlow.js GPU 后端的堆栈跟踪，并确定 MoViNet 推理无法找到已注册平台的原因。当 GPU 模式下的视频分类能够在不出现所报告错误的情况下继续进行时，即视为完成。

Which version of recognize are you using?

6.11

Enabled Modes

Object recognition, Face recognition, Video recognition, Music recognition

TensorFlow mode

GPU mode

Downstream App

Memories App

Which Nextcloud version do you have installed?

28.0.4.1

Which Operating system do you have installed?

Ubuntu 20.0.4.4

Which database are you running Nextcloud on?

Postgres

Which Docker container are you using to run Nextcloud? (if applicable)

28.0.4.1

How much RAM does your server have?

32

What processor Architecture does your CPU have?

x86_64

Describe the Bug

Seems like after upgrade to NC 28.0.4.1 & Recognize 6.1.1 it started to report below.
It seems like it launches process, based on looking at nvidia-smi and it stays there though doesn't seem like it puts a load on GPU.

Classifier process output: Error: Session fail to run with error: 2 root error(s) found.
  (0) NOT_FOUND: could not find registered platform with id: 0x7fd379c7fae4
\t [[{{node movinet_classifier/movinet/stem/stem/conv3d/StatefulPartitionedCall}}]]
\t [[StatefulPartitionedCall/_1555]]
  (1) NOT_FOUND: could not find registered platform with id: 0x7fd379c7fae4
\t [[{{node movinet_classifier/movinet/stem/stem/conv3d/StatefulPartitionedCall}}]]
0 successful operations.
0 derived errors ignored.
    at NodeJSKernelBackend.runSavedModel (/var/www/html/custom_apps/recognize/node_modules/@tensorflow/tfjs-node-gpu/dist/nodejs_kernel_backend.js:461:43)
    at TFSavedModel.predict (/var/www/html/custom_apps/recognize/node_modules/@tensorflow/tfjs-node-gpu/dist/saved_model.js:341:43)
    at MovinetModel.predict (/var/www/html/custom_apps/recognize/src/movinet/MovinetModel.js:46:21)
    at /var/www/html/custom_apps/recognize/src/movinet/MovinetModel.js:95:24
    at /var/www/html/custom_apps/recognize/node_modules/@tensorflow/tfjs-core/dist/tf-core.node.js:4559:22
    at Engine.scopedRun (/var/www/html/custom_apps/recognize/node_modules/@tensorflow/tfjs-core/dist/tf-core.node.js:4569:23)
    at Engine.tidy (/var/www/html/custom_apps/recognize/node_modules/@tensorflow/tfjs-core/dist/tf-core.node.js:4558:21)
    at Object.tidy (/var/www/html/custom_apps/recognize/node_modules/@tensorflow/tfjs-core/dist/tf-core.node.js:8291:19)
    at MovinetModel.inference (/var/www/html/custom_apps/recognize/src/movinet/MovinetModel.js:92:21)
    at runMicrotasks (<anonymous>)

At the same time it seems to have all what's needed (btw. it didn't raise this error prior upgrades):

./bin/node src/test_gputensorflow.js 
2024-04-07 21:37:06.584377: I tensorflow/core/platform/cpu_feature_guard.cc:193] This TensorFlow binary is optimized with oneAPI Deep Neural Network Library (oneDNN) to use the following CPU instructions in performance-critical operations:  AVX2 FMA
To enable them in other operations, rebuild TensorFlow with the appropriate compiler flags.
2024-04-07 21:37:06.593729: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:975] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2024-04-07 21:37:06.636405: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:975] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2024-04-07 21:37:06.636674: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:975] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2024-04-07 21:37:07.109813: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:975] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2024-04-07 21:37:07.110064: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:975] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2024-04-07 21:37:07.110196: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:975] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2024-04-07 21:37:07.110354: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1532] Created device /job:localhost/replica:0/task:0/device:GPU:0 with 3393 MB memory:  -> device: 0, name: Quadro M1200, pci bus id: 0000:01:00.0, compute capability: 5.0

And models seem to be there too:

du -shx ./models/*
50M	./models/efficientnet_lite4
794M	./models/efficientnetv2
22M	./models/landmarks_africa
41M	./models/landmarks_asia
41M	./models/landmarks_europe
41M	./models/landmarks_north_america
31M	./models/landmarks_oceania
41M	./models/landmarks_south_america
47M	./models/movinet-a3
31M	./models/musicnn

Unsure where to look for an issue.
It is after ffmpeg finishes its job.

Thanks!

Expected Behavior

Would just proceed to classify.

To Reproduce

Unsure.

Debug log

No response