Use rank info to save trace files (#2504)

mreso · web-flow · commit 57c8d6b50d14 · 2025-03-21T07:57:15.000-07:00
diff --git a/torchtune/training/_profiler.py b/torchtune/training/_profiler.py
@@ -103,7 +103,7 @@ def trace_handler(
 
     exporter = tensorboard_trace_handler(
         curr_trace_dir,
-        worker_name=f"r0-{now.year}-{now.month}-{now.day}-{now.hour}-{now.minute}",
+        worker_name=f"r{rank}-{now.year}-{now.month}-{now.day}-{now.hour}-{now.minute}",
         use_gzip=True,
     )
     exporter(prof)

Original file line number	Diff line number	Diff line change
`@@ -103,7 +103,7 @@ def trace_handler(`
`103`	`103`
`104`	`104`	`exporter = tensorboard_trace_handler(`
`105`	`105`	`curr_trace_dir,`
`106`		`- worker_name=f"r0-{now.year}-{now.month}-{now.day}-{now.hour}-{now.minute}",`
	`106`	`+ worker_name=f"r{rank}-{now.year}-{now.month}-{now.day}-{now.hour}-{now.minute}",`
`107`	`107`	`use_gzip=True,`
`108`	`108`	`)`
`109`	`109`	`exporter(prof)`