[CI][Torch] Enable split reduction and O3 for llama_8b_fp16 gfx942 config (#23231)
This PR enables `-opt-level=O3` for the llama 8b fp16 quality test.
"--iree-dispatch-creation-enable-split-reduction" will be removed when
it is set to true by default.
ci-extra: test_torch
Signed-off-by: Bangtian Liu <liubangtian@gmail.com>
diff --git a/tests/external/iree-test-suites/torch_models/llama_8b_fp16/modules/scheduler_gfx942.json b/tests/external/iree-test-suites/torch_models/llama_8b_fp16/modules/scheduler_gfx942.json
index 2fb2c5b..22fdffe 100644
--- a/tests/external/iree-test-suites/torch_models/llama_8b_fp16/modules/scheduler_gfx942.json
+++ b/tests/external/iree-test-suites/torch_models/llama_8b_fp16/modules/scheduler_gfx942.json
@@ -3,9 +3,10 @@
"compiler_flags": [
"--iree-hal-target-device=hip",
"--iree-hip-target=gfx942",
- "--iree-opt-level=O1",
+ "--iree-opt-level=O3",
"--iree-stream-resource-memory-model=discrete",
"--iree-hal-indirect-command-buffers=true",
- "--iree-hal-memoization=true"
+ "--iree-hal-memoization=true",
+ "--iree-dispatch-creation-enable-split-reduction"
]
}