From 275dfcbdb478df876217bf58c2bb586f7a11c117 Mon Sep 17 00:00:00 2001 From: Dark Knight Date: Mon, 29 Apr 2024 11:53:23 -0700 Subject: [PATCH] Revert D56685840: Multisect successfully blamed "D56685840: [fbgemm] Change model transform fp8 linear op to fbgemm quantize ops" for one test failure Summary: This diff reverts D56685840 D56685840: [fbgemm] Change model transform fp8 linear op to fbgemm quantize ops by jianyuh causes the following test failure: Tests affected: - [cogwheel:cogwheel_gpu_ait_lowering_latency_regression_test#main](https://www.internalfb.com/intern/test/281475067301657/) Here's the Multisect link: https://www.internalfb.com/multisect/4966282 Here are the tasks that are relevant to this breakage: T174133180: 10+ tests unhealthy for oncall_model_processing_components_infra The backout may land if someone accepts it. If this diff has been generated in error, you can Commandeer and Abandon it. Reviewed By: jianyuh Differential Revision: D56714397 --- fbgemm_gpu/experimental/gen_ai/src/quantize/quantize.cpp | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/fbgemm_gpu/experimental/gen_ai/src/quantize/quantize.cpp b/fbgemm_gpu/experimental/gen_ai/src/quantize/quantize.cpp index a51674044..2a88ee10f 100644 --- a/fbgemm_gpu/experimental/gen_ai/src/quantize/quantize.cpp +++ b/fbgemm_gpu/experimental/gen_ai/src/quantize/quantize.cpp @@ -79,7 +79,7 @@ at::Tensor get_fp8_per_tensor_scale( c10::optional bs, c10::optional scale_ub); // scale upperbound -TORCH_LIBRARY_FRAGMENT(fbgemm, m) { +TORCH_LIBRARY(fbgemm, m) { #ifndef USE_ROCM // TODO: on AMD this throws "Undefined symbol" when loading // quantize_ops with