add pd

Signed-off-by: heheda <[email protected]>

Author: heheda12345

_posts/2025-09-11-qwen3-next.md

@@ -66,7 +66,7 @@ Another innovation in Qwen3-Next is **multi-token prediction**, which boosts bot
 Our Qwen3-Next integration is just the beginning. On the roadmap:
 
 * Further kernel optimizations for GatedDeltaNet layers.  
-* Better memory management and prefix caching for hybrid models.  
+* Better memory management, plus the support of automatic prefix caching and P/D disaggregation for hybrid models.  
 * Continuous throughput and CPU overhead reductions.
 
 ## **Acknowledgements**