A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 16, 07:00 UTC
[Misc] Change RedundantReshapesPass and FusionPass logging from info to debug (#10308)
Signed-off-by: Tyler Michael Smith <tyler@neuralmagic.com>
Tyler Michael Smith committed
Nov 14, 2024 at 21:44 UTC
2885ba0e24e536d0a5b2439be5e96aef504a2e7f
2 files changed
+3
-3
vllm/compilation/fusion.py
+2
-2
@@ -281,11 +281,11 @@ class FusionPass(InductorPass):
281
self.dump_graph(graph, "before_fusion")
282
283
count = self.patterns.apply(graph)
284
- logger.info("Replaced %s patterns", count)
284
+ logger.debug("Replaced %s patterns", count)
285
self.dump_graph(graph, "after_pattern_match")
286
287
# Manually process multi-output matches (and run DCE)
288
self.process_matches(graph)
289
- logger.info("Post-processed %s matches", len(self.matches))
289
+ logger.debug("Post-processed %s matches", len(self.matches))
290
self.dump_graph(graph, "after_fusion")
291
self.matches.clear()
vllm/compilation/reshapes.py
+1
-1
@@ -53,7 +53,7 @@ class RedundantReshapesPass(InductorPass):
53
graph.erase_node(node)
54
count += 1
55
56
- logger.info("Removed %s no-op reshapes", count)
56
+ logger.debug("Removed %s no-op reshapes", count)
57
58
self.dump_graph(graph, "after_reshapes")
59