Running multiple ML pipelines with almost the same code base

I have the task of running multiple different Python ML models/pipelines(a/b test with small changes in data processing stage ). Pipelines are with almost the same code base(data processing stage)

The architecture of my application is as follows:

1. pub/sub model with RabbitMQ
2. 1 publisher push messages to multiply queues
3. multiple subscribers consume data from queues
   I have the following solutions in mind:

• repeat/rewrite code for each ML pipeline
• write and deploy subscribers from different git branches
• some any solutions?

Which solution is optimal in terms of support and code duplication?