This Space runs an Ark-ASR 0.6B demo with the Transformers backend.

Default runtime limits are set for a single-GPU Space with GPU memory usage targeted below 15 GB:

• ARK_ASR_MAX_AUDIO_SECONDS=30
• ARK_ASR_DTYPE=float16
• ARK_ASR_ATTN_IMPL=sdpa
• Gradio queue concurrency is limited to one request.

For local testing on this machine, use the local checkpoint instead of pulling from the Hub:

scripts/run_local_gradio.sh