[WIP] Before plan execute.

2026-01-19 03:30:44 +08:00
parent e6e0dc5d7d
commit 9e6fdc0650
5 changed files with 377 additions and 10 deletions
--- a/nanovllm/config.py
+++ b/nanovllm/config.py
@@ -7,8 +7,8 @@ import torch

 class SparsePolicyType(Enum):
    """Sparse attention policy types."""
-    FULL = auto()   # No sparse attention (load all blocks)
-    QUEST = auto()  # Query-aware Top-K block selection (decode only)
+    FULL = auto()       # No sparse attention (load all blocks)
+    QUEST = auto()      # Query-aware Top-K block selection (decode only)


@dataclass