Tokkio 5.0 GA#
Tokkio 5.0 GA builds on the 5.0-Beta release with the new architecture and reference workflow that enables seamless integration of digital humans with a LLM, RAG or an agentic application, supporting 1-6 concurrent users per deployment based on available GPU resources.
Notable Updates#
Introduces a Lite version of the one-click deployment scripts for AWS
Supports integration for Mistral and Deepseek NIMs as LLM sources
The Tokkio Ingress, Tokkio UI and Redis run as standalone charts so that the main Tokkio application chart can be scaled independently
Uses the latest pipecat version 0.0.68
Update the Aki character asset with a default smile and additional look and feel improvements
Known Issues#
The Avatar’s mouth occasionally remains open on refreshing the stream while the avatar is responding. The issue resolves after a few seconds without any user intervention.
After several hours of deployment, the users might observe that the Avatar is not visible on UI. When this happens, the nvidia-smi command run inside the VST pod will give error “Failed to initialize NVML: Unknown Error” . To overcome this issue, the user can restart the VST pod and wait for sometime for the issue to get resolved.
The Avatar’s lips move a little faster at times due to frame duplication. This does not impact the TTS audio though.
Initial barge-in ASR is not getting detected - It is observed that interacting with the Avatar and interrupting it at the start results in no ASR detection.
It is observed that modifying the ace controller file and uploading the change using the ace configurator results in non-working app intermittently and starts to work after some time after upload. This issue is observed both on CSPs and on bare metal.
The avatar does not animate any lip movement and no TTS is generated for any queries. The transcript is generated correctly.
Mic mute/unmute issues observed when enable_start_page set to true for Tokkio UI - Mic mute/unmute behavior was reversed when using Tokkio UI with enable start page.
Duplicated frames observed in avatar video stream resulting in choppiness.
Nvidia-GPU-operator pods fail to run on running the infra installation with CNS 14.0 and master branch as git_ref using AWS one click scripts.
Uploading the ace controller deployment change through ace configurator results in the app not working intermittently.