The v0.25.0rc3 release includes a bugfix addressing the handling of lookahead in PD async KV load for MTP speculative decoding.
- Resolves issues with PD async KV load lookahead handling specifically for MTP spec decoding.
The v0.25.0rc3 release includes a bugfix addressing the handling of lookahead in PD async KV load for MTP speculative decoding.