Add Gemma pipeline to benchmark_large #17749
Conversation
mariecwhite
force-pushed
the
add-gemma
branch
4 times, most recently
from
3a56262
to
f30269d
Compare
| @@ -23,6 +23,7 @@ on: | |||
| # Please keep this default value in sync with the jobs.build_e2e_test_artifacts.with.shard-count field below | |||
| default: a2-highgpu-1g=1,c2-standard-60=2,default=1 | |||
| type: string | |||
| pull_request: | |||
There was a problem hiding this comment.
revert before merge
also, you can trigger with workflow_dispatch if you push to a branch in this repo (not a fork) and use the "run workflow" button on https://github.com/iree-org/iree/actions/workflows/benchmark_large.yml
| source_type=common_definitions.ModelSourceType.EXPORTED_STABLEHLO_MLIR, | ||
| source_url="https://storage.googleapis.com/iree-model-artifacts/jax/jax_models_0.4.25_1709787220/GEMMA2BIT_GREEDY_FP32_JAX_1X1024XI32_256XI32/stablehlo.mlirbc", | ||
| entry_function="main", | ||
| input_types=["1x1024xi32"], |
There was a problem hiding this comment.
Looks like this failed with
cmd: [PosixPath('/work/linux-x86_64-benchmark-tools-dir/build/tools/iree-benchmark-module'), '--module=/tmp/iree-benchmarks/97adcc6379d65b94ca854da90290e4a38f62ca40/iree_module_Gemma2bit_fp32_stablehlo___x86_64-cascadelake-linux_gnu-llvm_cpu__default-flags_dt-uk_/module.vmfb', '--function=main', '--device_allocator=caching', '--task_topology_max_group_count=30', '--device=local-task', '--input=1x1024xi32=0', '--time_unit=ns', '--benchmark_format=json', '--benchmark_out_format=json', '--print_statistics=true', '--benchmark_repetitions=10']
work/runtime/src/iree/tooling/function_io.c:387: UNIMPLEMENTED; unimplemented cconv type `v`; parsing input `1x1024xi32=0
https://github.com/iree-org/iree/actions/runs/9691839363/job/26744473157?pr=17749#step:9:46
There was a problem hiding this comment.
Fixed that error but now running into a new one: #17774
Signed-off-by: mariecwhite <[email protected]>
No description provided.