[Bugfix] Zero new KV blocks for quantized + sliding-window hybrid caches

vllm/v1/kv_cache_interface.py

@@ -939,6 +939,38 @@ class KVCacheConfig:
     def has_mamba_layers(self) -> bool:
         return any(isinstance(g.kv_cache_spec, MambaSpec) for g in self.kv_cache_groups)
 
+    @property
+    def has_block_dropping_layers(self) -> bool:
+        """Any group uses an attention type that frees KV blocks mid-request
+        (sliding-window or chunked-local)."""
+        return any(
+            isinstance(g.kv_cache_spec, (SlidingWindowSpec, ChunkedLocalAttentionSpec))
+            for g in self.kv_cache_groups
+        )
+
+    @property
+    def has_quantized_kv_cache(self) -> bool:
+        """Any group stores its KV cache in a quantized dtype (FP8/NVFP4)."""
+        return any(
+            isinstance(g.kv_cache_spec, AttentionSpec)
+            and g.kv_cache_spec.kv_quant_mode != KVQuantMode.NONE
+            for g in self.kv_cache_groups
+        )
+
     @property
     def needs_kv_cache_zeroing(self) -> bool:
-        return self.has_mamba_layers
+        """Whether newly allocated KV cache blocks must be zeroed before use.
+
+        Required in two cases:
+        - Mamba layers, whose state is read before being fully written.
+        - Hybrid configs mixing a block-dropping attention type with a quantized
+          KV cache. Groups share the same tensors, so a page freed mid-request by
+          a block-dropping group can be reallocated and read before every slot is
+          written this step (partial-block tail, alignment padding). Stale
+          quantized bytes decode to NaN/Inf and poison attention; zeroing makes
+          such slots read as 0. Uniform/unquantized configs are unaffected.
+        """
+        needs_quantized_reuse_zeroing = (
+            self.has_block_dropping_layers and self.has_quantized_kv_cache
+        )
+        return self.has_mamba_layers or needs_quantized_reuse_zeroing