Docker compose file for V2

[hightide82]

Hi,

Great program and I'm impressed with and grateful for your work.

I am running V1 in Docker Desktop for Windows under WSL2 with Ubuntu. I modified the included compose file to run tailscale so I can access the service remotely from within my tailnet.

I have no earthly idea how I would go about doing this with V2 and once we start talking about modifying dockerfiles and building images and so on, I'm lost. If anyone has the ability to write such a compose file for V2 it would be appreciated. If that is a monumental undertaking then I am ok with V1. :)

[Paladinium]

I cannot help you with Docker compose. But what I can say is that there are Docker images published to Docker hub. No need to build them yourself.

See #378 (comment)

[RenNagasaki]

Hi,

Great program and I'm impressed with and grateful for your work.

I am running V1 in Docker Desktop for Windows under WSL2 with Ubuntu. I modified the included compose file to run tailscale so I can access the service remotely from within my tailnet.

I have no earthly idea how I would go about doing this with V2 and once we start talking about modifying dockerfiles and building images and so on, I'm lost. If anyone has the ability to write such a compose file for V2 it would be appreciated. If that is a monumental undertaking then I am ok with V1. :)

If you could explain in more detail what exactly you did to V1 to run it with Tailscale, we might be able to help.
AFAIK docker compose using the above mentioned Docker Image (erew123/alltalk_tts) should be enough to combine with tailscale.

[hightide82]

Thanks. Here's the docker compose file that got V1 working with Tailscale:

version: '3.8'
services:

  tailscale: 
    image: tailscale/tailscale:latest
    container_name: alltalk-tailscale
    restart: unless-stopped
    environment:
      TS_AUTHKEY: ${TS_AUTHKEY:?err}
      TS_HOSTNAME: ${TS_HOSTNAME:-ts-docker}
      TS_STATE_DIR: /var/lib/tailscale/
    volumes:
      - ./ts_data:/var/lib/tailscale/

  alltalk-tts:
    image: erew123/alltalk_tts:cuda
    restart: unless-stopped
# don't publish ports
#    ports:
#      - "7851:7851"
# add tailscale network
    network_mode: service:tailscale 

    volumes:
      - ./outputs:/app/outputs/
      - ./models:/app/models/
      - ./voices:/app/voices/
      - ./finetune/put-voice-samples-in-here:/app/finetune/put-voice-samples-in-here
      # use the V1 confignew.json
      - ./confignew.json:/app/confignew.json
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: all
              capabilities: [ gpu ]

[hightide82]

I changed the IP address in confignew.json to: "ip_address": "alltalk.MY-TAILNET.ts.net"
Then I went to the exec tab in my tailscale container and ran "tailscale serve --bg 7851"

[RenNagasaki]

AFAIK this should still work with v2. Don't know a reason why it shouldnt.

[RenNagasaki]

@hightide82 the only difference is, you call the new UI via port 7852 (default), the old ui is still accessible via 7851 though.

[hightide82]

I made sure Docker Desktop is using WSL2 backend / Ubuntu 20.04 / CUDA toolkit 12.4 per the Docker Support (Beta) Wiki page.
Then, I replaced the V1 config file with V2, and pulled the same image using the same compose file above.

I get three errors resulting in a startup loop. If any of these errors mean anything to you, I'd appreciate your comments.

File "/app/tts_server.py", line 58, in
params["low_vram"] = "false" if not torch.cuda.is_available() else params["low_vram"]
KeyError: 'low_vram'

File "/app/script.py", line 70, in
from modules import chat, shared, ui_chat
During handling of the above exception, another exception occurred:

File "/app/script.py", line 79, in
output_folder_wav = params["output_folder_wav_standalone"]
KeyError: 'output_folder_wav_standalone'

[RenNagasaki]

If I may ask why don't you just use alltalks windows version instead of running via wsl2 on docker?
But the solution should be pretty simple: switch the :cuda tag to one of the active ones. (XTTS, PIPER, and so on, depenging on which engine you want to use)

[RenNagasaki]

cuda, latest and some others are over a year old and v1.
You can check here which tags are active: https://hub.docker.com/r/erew123/alltalk_tts/tags

[Paladinium]

Note that the Docker documentation in the wiki is actually out of date. The readme however is up-to-date: https://github.com/erew123/alltalk_tts/blob/alltalkbeta/DOCKER_README.md

I mentioned it because CUDA was bumped to 12.6.

[RenNagasaki]

AFAIK cuda's downward compatible, isn't it?
So him using 12.4 shouldn't hurt.

[hightide82]

Confirmed that the default Docker Desktop backend works without any special preparation.

[hightide82]

Hahaha! Duh. That indeed fixed everything.

I have a number of other services running in Docker, and attempting to connect them to the host machine would have been problematic.

Thank you so much!

[hightide82]

So sorry to bug you again. This works great but I'd really like to use my host for these directories:

volumes:
  - ./outputs:/app/outputs
  - ./models:/app/models
  - ./voices:/app/voices
  - ./finetune/put-voice-samples-in-here:/app/finetune/put-voice-samples-in-here
  - ./confignew.json:/app/confignew.json

These directories seem to be mounted to /home/alltalk despite this argument in my compose file. If there is some simple change to this syntax I would appreciate the input. I understand this is not a Docker 101 forum. :)

[hightide82]

I answered my own question. Change /app/ to /home/alltalk/.
I thought it wasn't working because I had an mp3 file in there.

Anyway I've successfully spun up alltalk v2 using a single docker compose file and the JSON config file. Now that I have it figured out, it was pretty simple. Thanks again for the help.