rkimball commented on a change in pull request #9248: URL: https://github.com/apache/tvm/pull/9248#discussion_r726566056
########## File path: python/tvm/autotvm/graph_tuner/utils/traverse_graph.py ########## @@ -130,8 +130,11 @@ def _traverse_expr(node): call = relay.Call(node.op, params, node.attrs) mod = tvm.IRModule.from_expr(relay.Function(params, call)) relay.backend.compile_engine.get().clear() + target_string = ( + tvm_target if " -device" in tvm_target else f"{tvm_target} -device=tracing" + ) build_thread = threading.Thread( - target=relay.build, args=(mod, f"{tvm_target} -device=tracing", None, None) + target=relay.build, args=(mod, target_string, None, None) Review comment: I found this in traverse_graph.py. tracing is only used trace the calls to topi, not for building anything so replacing the device with `tracing` here should be fine. I did test this with cuda and it no longer crashes and completes successfully and the model runs. ``` # Utilize tracing target to fetch workload with topo-order. # Since we only need workload, dummy target can be used to # create task. ``` -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: commits-unsubscr...@tvm.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org