Dynamo v1.3.0-kimi-k2.6-dev.1
Pre-release
Release Notes
Dynamo v1.3.0-kimi-k2.6-dev.1 is an experimental dev build giving an early look at Moonshot AI Kimi K2.6 support on the Dynamo vLLM backend. It is not recommended for production — features may be incomplete and APIs, behaviors, and defaults may change before the stable release. Use it for evaluation, testing, and early feedback only.
Summary
Dynamo v1.3.0-kimi-k2.6-dev.1 adds Moonshot AI Kimi K2.6 (moonshotai/Kimi-K2.6) serving on the Dynamo vLLM backend, with aggregated deployment recipes for B200 (NVFP4) and H200 (INT4) across chat and agentic profiles (#10187). It builds from a recent main commit (d1d794c2) on a vLLM v0.21 runtime base. Only the vLLM container is published.
Branch: release/1.3.0-kimi-k2.6-dev.1, built from commit d1d794c2
vLLM: v0.21 runtime base
Container Images
| Backend | Arch | Image |
|---|---|---|
| vLLM (CUDA 13) | multi-arch (amd64 + arm64) |
nvcr.io/nvidia/ai-dynamo/vllm-runtime:1.3.0-kimi-k2.6-dev.1 |
Backend Versions
| Backend | Source | CUDA | Python | Notes |
|---|---|---|---|---|
| vLLM | Dynamo-built runtime on a vLLM v0.21 base | 13.0 | 3.12 | — |
Models
| Model | Precision | GPU | Notes |
|---|---|---|---|
nvidia/Kimi-K2.6-NVFP4 |
NVFP4 | B200 | served as moonshotai/Kimi-K2.6; aggregated; chat + agentic; eagle3-mla speculative decoding (lightseekorg/kimi-k2.6-eagle3-mla) |
moonshotai/Kimi-K2.6 |
INT4 | H200 | aggregated; chat + agentic; eagle3-mla speculative decoding |
Full Changelog
Recipes
- Kimi K2.6 Deployment Recipes: Added Kimi K2.6 serving recipes and Kubernetes deployment configurations for B200 (NVFP4) and H200 (INT4) across chat and agentic workloads, with model-cache download/PVC manifests and a trace-replay performance job (recipe folder, README, #10187).