Training hangs and almost never reaches 100% (reached one or twice)

[boyoma]

Describe the bug My installation looks, fine (web embedding on another domain too) but every time I press train chabot percentage will raise slowly and eventually it will stop before reaching 100%, and I will need to press the button again. It got completed maybe once out of 100 times. More often than not it stuck at 0%.

The exact same bot was first produced in localhost and train there and it was working very fine. slow but it completes.

I'm using a 4 GB Memory / 80 GB Disk / Ubuntu 20.04 (LTS) x64

To Reproduce Steps to reproduce the behavior:

1. Go to 'a bot'
2. Click on 'train chatbot'
3. See error 'almost never reach 100%'

Expected behavior Being slow it is ok but at least it should complete

Environment (please complete the following information):

• OS: linux
• Browser chrome
• Browser Version 111.0.5563.64
• Botpress Version 12.30.7

[cccaballero]

Seems like I have the same problem, I installed on my local PC using the following docker-comopse.yml to test:

version: '3'

services:
  botpress:
    image: botpress/server
    expose:
      - 3000
    ports:
      - "3000:3000"
    environment:
      - DATABASE_URL=postgres://postgres:secretpw@postgres:5435/botpress_db
    depends_on:
      - postgres
    volumes:
      - ./build/botpress/data:/botpress/data

  postgres:
    image: postgres:11.2-alpine
    expose:
      - 5435
    environment:
      PGPORT: 5435
      POSTGRES_DB: botpress_db
      POSTGRES_PASSWORD: secretpw
      POSTGRES_USER: postgres
    volumes:
      - pgdata:/var/lib/postgresql/data

volumes:
  pgdata:

This is all I can see in the logs:

botpress_1 | 04/17/2023 00:17:21.386 [NLU] training-queue [test01/87443a324c26b933.b7f4a95061d75566.3265.en] Training Queued.
botpress_1 | 04/17/2023 00:17:21.692 [NLU] Engine:training Training worker successfully started on process with pid 180.

I am using to test a new bot called test01 from the Small Talk template and without making changes.

I have not been able to complete any training, the maximum that I have been able to reach is 80%

[sebburon]

How much memory is accessible to your Botpress containers? Usually, when the training stops between 80 and 99% it's because the training process was killed by the OS because it was using too much memory.

Make sure your Botpress node has access to at least 3GB of ram.

Thanks,

[cccaballero]

@sebburon I don't think it's a memory problem, I don't have any limits defined for the docker container, and I have plenty of ram. This is what docker stats tells me:

MEM USAGE / LIMIT
573.6MiB / 38.88GiB