[WIP] fixing attention compute error.

2025-12-30 00:31:48 +08:00
parent bf4c63c7ec
commit 89f8020d38
12 changed files with 2175 additions and 103 deletions
--- a/nanovllm/kvcache/hybrid_manager.py
+++ b/nanovllm/kvcache/hybrid_manager.py
@@ -337,10 +337,10 @@ class HybridKVCacheManager(KVCacheManager):
                block = self.logical_blocks[logical_id]
                if block.location == BlockLocation.CPU:
                    cpu_blocks.append(block.cpu_block_id)
-        logger.debug(
-            f"get_prefilled_cpu_blocks: prefilled_blocks={list(self.prefilled_blocks)}, "
-            f"returned cpu_blocks={cpu_blocks}"
-        )
+        # logger.debug(
+        #     f"get_prefilled_cpu_blocks: prefilled_blocks={list(self.prefilled_blocks)}, "
+        #     f"returned cpu_blocks={cpu_blocks}"
+        # )
        return cpu_blocks

    # ========== Ring Buffer CPU-primary support ==========