Tether AI, an efficient on-device AI entity, has recently released the exclusive version of its popular QVAC software development kit (SDK). The new rollout introduces several enhancements that focus on enhancing on-device AI performance as well as developer workflows. As per QVAC’s official X announcement, the latest SDK 0.15.0 supports prompt batching, improved mobile version processing, an extra local coding agent inclusion, and a local AMD GPU backend. Hence, the update underscores Tether AI’s consistent attention toward the provision of high-performance AI apps that run without depending on cloud infrastructure.
The CEO of Tether, Paolo Ardoino, retweeted this announcement by QVAC raising the importance of this update.
QVAC SDK 0.15.0 Goes Live with Prompt Batching and Local AMD GPU Support
The release of QVAC SDK 0.15.0 includes a key feature of prompt batching to support a large language model (LLM) add-on. The respective feature permits developers to merge diverse prompts into a single processing task while conducting their concurrent execution. Instead of waiting for all of the prompts to complete together, every generated response is reportedly returned following completion.
Apart from that, the release offers a native HIP/ROCm backend to facilitate AMD GPUs via the @qvac/vla-ggml stack. While using Linux x64 devices, the backend is automatically chosen over Vulkan at the time of ROCm’s availability. The implementation provides nearly 23% increased performance in comparison with Vulkan, as well as almost 14% better performance against PyTorch-ROCm.
Improving Developer Flexibility and Transparency
At the same time, another crucial improvement is the OpenClaw integration alongside OpenCode. As a result, developers are now permitted to select between the cloud-free coding workstreams, broadening flexibility for AI development. Overall, Tether AI’s QVAC SDK 0.15.0 launch adopts an inclusive code style, providing more ease, consistency, and codebase transparency to developers.
