Add seamless text-to-speech and text-to-text translation with updated…

… dependencies

- Create `seamless_v2.py` for new seamless text-to-speech and text-to-text translation tasks.
- Update Dockerfile to use CUDA 12.4.1 and Ubuntu 22.04.
- Upgrade dependencies in `requirements.txt`.
- Modify `model-values.yaml` to update image reference, GPU, and memory limits.
- Simplify `seamless_asr.py` for seamless ASR only (as huggingface expects a different model class for asr vs other tasks)
- Fix `run-dev.sh` to enable NVIDIA runtime and GPU usage.

chart/model-values.yaml

@@ -195,9 +195,10 @@ deployments:
         bark
 
   - name: "common-seamless"
-    image: "us-docker.pkg.dev/dara-c1b52/cloudbuild/gooey-gpu/common@sha256:5fb0ffa128cbdda86747fedf5ef68e9df8256735d8535149c6fffa41a3749883"
+    image: "crgooeyprodwestus1.azurecr.io/gooey-gpu-common:3"
     limits:
-      memory: "19Gi"
+      gpu: "10Gi"
+      memory: "28Gi"  # (220 / 80) * 10
       cpu: "1"
     env:
       IMPORTS: |-

common/Dockerfile

@@ -1,4 +1,4 @@
-FROM nvidia/cuda:12.2.0-devel-ubuntu20.04
+FROM nvidia/cuda:12.4.1-cudnn-devel-ubuntu22.04
 
 ARG DEBIAN_FRONTEND=noninteractive