diff --git a/vllm/compilation/fusion.py b/vllm/compilation/fusion.py index 2a0cf000..eb43604b 100644 --- a/vllm/compilation/fusion.py +++ b/vllm/compilation/fusion.py @@ -281,11 +281,11 @@ class FusionPass(InductorPass): self.dump_graph(graph, "before_fusion") count = self.patterns.apply(graph) - logger.info("Replaced %s patterns", count) + logger.debug("Replaced %s patterns", count) self.dump_graph(graph, "after_pattern_match") # Manually process multi-output matches (and run DCE) self.process_matches(graph) - logger.info("Post-processed %s matches", len(self.matches)) + logger.debug("Post-processed %s matches", len(self.matches)) self.dump_graph(graph, "after_fusion") self.matches.clear() diff --git a/vllm/compilation/reshapes.py b/vllm/compilation/reshapes.py index 0d284246..36597e11 100644 --- a/vllm/compilation/reshapes.py +++ b/vllm/compilation/reshapes.py @@ -53,7 +53,7 @@ class RedundantReshapesPass(InductorPass): graph.erase_node(node) count += 1 - logger.info("Removed %s no-op reshapes", count) + logger.debug("Removed %s no-op reshapes", count) self.dump_graph(graph, "after_reshapes")