Paddle vl benchmark runner saves off data

This commit is contained in:
Jake Poznanski 2025-10-21 20:09:39 +00:00
parent 0c3d2d2e16
commit 5c16f52d3b

View File

@ -85,6 +85,8 @@ commands.extend([
"python -m pip install https://paddle-whl.bj.bcebos.com/nightly/cu126/safetensors/safetensors-0.6.2.dev0-cp38-abi3-linux_x86_64.whl",
"python -m olmocr.bench.convert paddlevl --dir ./olmOCR-bench/bench_data --parallel 0",
"pip install s5cmd",
"s5cmd cp olmOCR-bench/bench_data/ s3://ai2-oe-data/jakep/olmocr-bench-runs/$BEAKER_WORKLOAD_ID/",
"python -m olmocr.bench.benchmark --dir ./olmOCR-bench/bench_data"
])
@ -97,11 +99,11 @@ task_spec_args = {
" && ".join(commands)
],
"context": TaskContext(
priority=Priority.normal,
preemptible=True,
priority=Priority.urgent,
preemptible=False,
),
"resources": TaskResources(gpu_count=1),
"constraints": Constraints(cluster=["ai2/neptune-cirrascale"]),
"constraints": Constraints(cluster=["ai2/jupiter"]),
"result": ResultSpec(path="/noop-results"),
}