@voltagent/mcp-server@2.1.0 v0.27.0
DavidHDev/react-bitsAug 3, 2026 → Aug 10, 20267 days between releases
What's New
Major release adds Kimi K3 support and PyTorch 2.13.0 upgrade
Breaking Changes
- PyTorch 2.13.0 upgrade along with torchvision 0.28.0 and Triton 3.7.1
- XPU and CPU backends upgraded to PyTorch 2.13
New Features
- Full Kimi K3 support with complete stack landing
- Model Runner V2 expansion to encoder-only workloads
- JIT warmup infrastructure to remove first-request stalls
- Generic P2P secondary tier for KV offloading
- New models including Qwen3.5, K-EXAONE, VaultGemma, and jina-embeddings-v5-text-nano
Improvements
- FlashAttention 4 integration with FP8 KV cache support
- DeepSeek-V4 performance improvements via sequence parallelism
- Resilient large-scale serving framework and disaggregation
- Adaptive top-k width for improved inference
Bug Fixes
- Removal of first-request compilation stalls via JIT warmup
- Removal of redundant full kernel for 1.88x kernel improvement
Upgrade Notes
Requires upgrading the environment to PyTorch 2.13.0, torchvision 0.28.0, and Triton 3.7.1, which is described as a breaking change. XPU and CPU backends are also upgraded. Expect potential compatibility issues with existing models or configurations due to the environment shift.
The Two Releases
@voltagent/mcp-server@2.1.0
Aug 3, 2026· 3 months ago
This minor version update upgrades the Model Context Protocol SDK to version 1.30.0 and introduces new HTTP streaming capabilities.
Full release notesv0.27.0
Aug 10, 2026· 2 months ago
This release marked the full landing of Kimi K3 support and expanded Model Runner V2 to non-generative workloads like pooling and classification. It also included early next-gen hardware enablement for SM107 and significant DeepSeek-V4 performance improvements.
Full release notes