Higgs Audio V2 hits RunPod setup snag

This is the classic open-source audio-model reality check: the weights may be public, but the deployment recipe is still the hard part. Boson AI’s docs point users toward NVIDIA NGC PyTorch containers rather than a generic app image, so the base Docker choice matters a lot. The repo exposes both a native Python serving path and a vLLM backend, but GitHub issues show the vLLM image can mis-handle audio generation and truncate output. Boson AI says the 3B model wants roughly RTX 4090-class hardware for efficient inference, so underpowered RunPod pods can masquerade as software problems. The demand signal is real: Higgs Audio V2 is one of the stronger open-source TTS/audio models for expressive, multi-speaker generation, which makes the deployment friction worth solving.