GPU Speech to text container (Melia 1)

Learn about the Speechmatics Melia 1 GPU container system

Melia 1 is a multilingual, GPU-only model, available for Batch transcription only. For an overview of the model, see Models.

Prerequisites

• A license file or a license token
  • There is no specific license for the GPU Inference Container, it will run using an existing Speechmatics license for the Batch Container
• Access to our Docker repository

System requirements

The system must have:

• Nvidia GPU(s) with at least 16GB of GPU memory
• Nvidia drivers (see below for supported versions)
• CUDA compute capability of 7.5-12.1 inclusive, which corresponds to the Turing, Ampere, Lovelace, Hopper, Blackwell architectures. Cards with the Volta architecture or below are not able to run the models
• 24 GB RAM
• The nvidia-container-toolkit installed
• Docker version > 19.03

See Performance and cost for more information on the performance and cost of the container.

Nvidia drivers

• The GPU Inference Container is based on CUDA 13.0.1, which requires NVIDIA Driver release 580 or later.

Driver installation can be validated by running nvidia-smi. This command should return the Nvidia driver version and show additional information about the GPU(s).

Cloud instances

The GPU node can be provisioned in the cloud.

Running the image

Currently, each GPU Inference Container can only run on a single GPU. If a system has more than one GPU, the device must be specified using CUDA_VISIBLE_DEVICES or selecting the device using the --gpus argument. See Nvidia/CUDA documentation for details.

Pull the Melia 1 inference server image as described in Accessing images, then run it:

docker run --rm -it \
  -v $PWD/license.json:/license.json \
  --gpus '"device=0"' \
  -e CUDA_VISIBLE_DEVICES \
  -p 8001:8001 \
  speechmaticspublic.azurecr.io/sm-asr-inference-server-melia-1:1.3.0

When the Container starts you should see output similar to this, indicating that the server has started and is ready to serve requests.

I0705 08:12:55.419608 1 server.cc:709]
+--------------+---------+--------+
| Model        | Version | Status |
+--------------+---------+--------+
| aed          | 1       | READY  |
| body         | 1       | READY  |
| detokenizer  | 1       | READY  |
| dz           | 1       | READY  |
| ensemble     | 1       | READY  |
| preprocessor | 1       | READY  |
+--------------+---------+--------+
...
I0705 08:12:55.515473 1 grpc_server.cc:2579] "Started GRPCInferenceService at 0.0.0.0:8001"
I0705 08:12:55.515940 1 http_server.cc:4961] "Started HTTPService at 0.0.0.0:8000"
I0705 08:12:55.598672 1 http_server.cc:400] "Started Metrics Service at 0.0.0.0:8002"

Batch inference

The Melia 1 inference server currently runs in batch mode only, processing whole files and returning the transcript at the end.

Once the inference server is running, run the Melia 1 transcriber with the SM_INFERENCE_ENDPOINT environment variable set to the gRPC endpoint of the inference server. The transcriber offloads inference to the server, so it does not require a GPU:

docker run --rm -it \
  -e SM_INFERENCE_ENDPOINT=<server>:<port> \
  -v $PWD/license.json:/license.json \
  -v $PWD/example.wav:/input.audio \
  <speech_container_image_name>

To accept multiple jobs over an HTTP API without restarting the container between jobs, run the transcriber as a batch persistent worker.

Monitoring the server

The inference server is based on Nvidia's Triton architecture and as such can be monitored using Triton's inbuilt Prometheus metrics, or the GRPC/HTTP APIs. To expose these, configure an external mapping for port 8002(Prometheus) or 8000(HTTP).

Docker compose example

This Docker Compose file will create a Speechmatics Melia 1 GPU Inference Server:

(assumes your license.json file is in the current working directory)

docker-compose.yml

version: "3.8"

networks:
  transcriber:
    driver: bridge

services:
  triton:
    image: speechmaticspublic.azurecr.io/sm-asr-inference-server-melia-1:1.3.0
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              ### Limit to N GPUs
              # count: 1
              ### Pick specific GPUs by device ID
              # device_ids:
              #   - 0
              #   - 3
              capabilities:
                - gpu
    container_name: triton
    networks:
      - transcriber
    expose:
      - 8000/tcp
      - 8001/tcp
      - 8002/tcp
    environment:
      - NVIDIA_DRIVER_CAPABILITIES=all
      - NVIDIA_VISIBLE_DEVICES=all
      - CUDA_VISIBLE_DEVICES=0
    volumes:
      - $PWD/license.json:/license.json:ro