[WIP] EasyMP vllm-omni model definition#15741

Open

vklimkov-nvidia wants to merge 31 commits into

NVIDIA-NeMo:easymp_voiceagentfrom

vklimkov-nvidia:easymp_vllm_omni

Commits on Jun 2, 2026

nemo/collections/tts/models/easy_magpietts_inference.py: remove duplicate speaker encoder application
vklimkov-nvidia
committed
examples/tts/easymagpie_vllm_omni: initial commit for vllm_omni definition of Easy Magpie
vklimkov-nvidia
committed
examples/tts/easymagpie_vllm_omni: switch to actual configuration
vklimkov-nvidia
committed
examples/tts/easymagpie_vllm_omni: make sure model runs with cuda graphs
vklimkov-nvidia
committed
examples/tts/easymagpie_vllm_omni: extend preprocess to take speaker embeddings and prepare prefill embeddings
vklimkov-nvidia
committed
examples/tts/easymagpie_vllm_omni: introduce script to convert the checkpoint to vllm omni one
vklimkov-nvidia
committed
examples/tts/easymagpie_vllm_omni: clean up, add readme
vklimkov-nvidia
committed
examples/tts/easymagpie_vllm_omni: implement delay and proper phoneme prediction processing
vklimkov-nvidia
committed

Commits on Jun 3, 2026

Commits on Jun 4, 2026

Commits on Jun 5, 2026

Commits on Jun 8, 2026

examples/tts/easymagpie_vllm_omni: log fails, adjust numpy installation
vklimkov-nvidia
committed

Commits on Jun 9, 2026