Help understanding the usage of faster-whisper container

[ghost]

I've started a Docker container of faster-whisper, but I don't know how to use it. I've read faster-whispers documentation, which is its README.md.

What I've tried:

1. curl http:localhost:10300. I was waiting for some kind of a greeting or an error, but nothing.
2. POSTing a file using curl to the same address, still nothing.
3. POSTing a file to the /transcribe endpoint - same

Here is lsiosupport

What am I doing wrong?

[drizuid]

where are you running the curl from?
please provide the docker run command or your compose. if you're using a UI, note that we do not support any of them except unraid (which provides a docker run)
also please provide the output of docker version

[ghost]

I started a container in a terminal's tab, then opened another one tab, and ran curl

docker file

I don't use UI, just docker compose

docker version

[ghost]

tried to install it on WSL 2, just to try it out.

the container itself starts without any error:

faster-whisper  | INFO:__main__:Ready
faster-whisper  | Connection to localhost (127.0.0.1) 10300 port [tcp/*] succeeded!
faster-whisper  | [ls.io-init] done.

curl still doesn't print anything from the container, but shows that connection is successful:

* IPv6: ::1
* IPv4: 127.0.0.1
*   Trying [::1]:10300...
* connect to ::1 port 10300 from ::1 port 44302 failed: Connection refused
*   Trying 127.0.0.1:10300...
* Connected to localhost (127.0.0.1) port 10300
> GET / HTTP/1.1
> Host: localhost:10300
> User-Agent: curl/8.5.0
> Accept: */*

I've also tried to run a Python script, that I found in the faster-whisper's README:

from faster_whisper import WhisperModel

model_size = "large-v3-turbo"

# Run on GPU with FP16
model = WhisperModel(model_size, device="cuda", compute_type="float16")

# or run on GPU with INT8
# model = WhisperModel(model_size, device="cuda", compute_type="int8_float16")
# or run on CPU with INT8
# model = WhisperModel(model_size, device="cpu", compute_type="int8")

segments, info = model.transcribe("first.mp3", beam_size=5)

print("Detected language '%s' with probability %f" % (info.language, info.language_probability))

for segment in segments:
    print("[%.2fs -> %.2fs] %s" % (segment.start, segment.end, segment.text))

it exits with an error:

Unable to load any of {libcudnn_ops.so.9.1.0, libcudnn_ops.so.9.1, libcudnn_ops.so.9, libcudnn_ops.so}
Invalid handle. Cannot load symbol cudnnCreateTensorDescriptor
Aborted (core dumped)

it seems to be fixed in #42

[ghost]

oh, forgot to say, I've changed the compose.yaml:

---
services:
  faster-whisper:
    image: lscr.io/linuxserver/faster-whisper:gpu
    container_name: faster-whisper
    deploy:
      resources:
        reservations:
          devices:
            - driver: "nvidia"
              count: all
              capabilities: [gpu]
    environment:
      - PUID=1000
      - PGID=1000
      - TZ=Europe/Moscow
      - WHISPER_MODEL=large-v3-turbo
      - WHISPER_BEAM=1
      - WHISPER_LANG=ru
    volumes:
      - /root/containers/faster-whisper:/config
    ports:
      - 10300:10300
    restart: unless-stopped