Methods and devices for multiple model single deployment engine implementation
Methods and apparatuses are provided for serving multiple models with a single engine. The method includes: providing a single engine for a plurality of models on a server, where the server includes at least one graphics processing unit (GPU) and memory coupled with the at least one GPU; loading the plurality of models onto the memory of the server at once; and serving, by the single engine, the plurality of models, where the single engine accommodates all structures and weights of the plurality of models with a shared input and output of the memory.
1. A single-engine serving method, comprising:
providing a single engine for a plurality of models on a server, wherein the server comprises at least one graphics processing unit (GPU) and memory coupled with the at least one GPU;
loading the plurality of models onto the memory of the server at once; and
serving, by the single engine, the plurality of models, wherein the single engine accommodates all structures and weights of the plurality of models with a shared input and output of the memory.
2. The single-engine serving method of claim 1 , further comprising:
maintaining, by the single engine, a set of queues for each of the plurality of models; and
sending, by the single engine, a head job from each queue into an execution engine in a batched manner.
3. The single-engine serving method of claim 1 , further comprising:
batch processing, by the single engine, all the inputs of the plurality of models.
4. The single-engine serving method of claim 1 , further comprising:
deploying the plurality of models on one single graphics processing unit (GPU).
5. The single-engine serving method of claim 4 , further comprising:
receiving a decoded video uploaded by a user; and
processing the decoded video with the plurality of models, wherein all the plurality of models are deployed on one single GPU.
6. The single-engine serving method of claim 5 , wherein each of the plurality of models has its own engine and context, and is preloaded on the one single GPU for processing new input.
7. The single-engine serving method of claim 4 , further comprising:
receiving a decoded video uploaded by a user; and
processing the decoded video with a set of detection models and a set of enhancement models, wherein all models in the set of detection models are deployed on one single GPU, and all models in the set of enhancement models are deployed on one single GPU.
8. The single-engine serving method of claim 7 , wherein each of the detection models and the enhancement models has its own engine and context, and is preloaded on the one single GPU for processing new input.
9. An apparatus for single-engine serving, comprising:
one or more processors; and
a memory configured to store instructions executable by the one or more processors;
wherein the one or more processors, upon execution of the instructions, are configured to:
provide a single engine for a plurality of models on a server, wherein the server comprises at least one graphics processing unit (GPU) and memory coupled with the at least one GPU;
load the plurality of models onto the memory of the server at once; and
serve the plurality of models, wherein the single engine accommodates all structures and weights of the plurality of models with a shared input and output of the memory.
10. The apparatus for single-engine serving of claim 9 , wherein the one or more processors are further configured to:
maintain a set of queues for each of the plurality of models; and
send a head job from each queue into an execution engine in a batched manner.
11. The apparatus for single-engine serving of claim 9 , wherein the one or more processors are further configured to:
batch process all the inputs of the plurality of models.
12. The apparatus for single-engine serving of claim 9 , wherein the one or more processors are further configured to:
deploy the plurality of models on one single graphics processing unit (GPU).
13. The apparatus for single-engine serving of claim 12 , wherein the one or more processors are further configured to:
receive a decoded video uploaded by a user; and
process the decoded video with the plurality of models, wherein all the plurality of models are deployed on one single GPU.
14. The apparatus for single-engine serving of claim 13 , wherein each of the plurality of models has its own engine and context, and is preloaded on the one single GPU for processing new input.
15. The apparatus for single-engine serving of claim 12 , wherein the one or more processors are further configured to:
receive a decoded video uploaded by a user; and
process the decoded video with a set of detection models and a set of enhancement models, wherein all models in the set of detection models are deployed on one single GPU, and all models in the set of enhancement models are deployed on one single GPU.
16. The apparatus for single-engine serving of claim 15 , wherein each of the detection models and the enhancement models has its own engine and context, and is preloaded on the one single GPU for processing new input.
17. A non-transitory computer readable storage medium, comprising instructions stored therein to implement a single-engine serving, wherein, upon execution of the instructions by one or more processors, the instructions cause the one or more processors to perform acts comprising:
providing a single engine for a plurality of models on a server, wherein the server comprises at least one graphics processing unit (GPU) and memory coupled with the at least one GPU;
loading the plurality of models onto the memory of the server at once; and
serving the plurality of models, wherein the single engine accommodates all structures and weights of the plurality of models with a shared input and output of the memory.
18. The non-transitory computer readable storage medium of claim 17 , wherein the instructions cause the one or more processors to perform acts further comprising:
maintaining a set of queues for each of the plurality of models; and
sending a head job from each queue into an execution engine in a batched manner.
19. The non-transitory computer readable storage medium of claim 17 , wherein the instructions cause the one or more processors to perform acts further comprising:
deploying the plurality of models on one single graphics processing unit (GPU).
20. The non-transitory computer readable storage medium of claim 19 , wherein the instructions cause the one or more processors to perform acts further comprising:
receiving a decoded video uploaded by a user; and
processing the decoded video with the plurality of models, wherein all the plurality of models are deployed on one single GPU.