# EXP = merge.Merge_BertBase_SquadDonor_4096
# EXP = merge.Merge_BertBase_HighResource_SquadDonor_4096
execution_items = EXP.create_all_execution_items()
print(f"Number of execution items to process: {len(execution_items)}")
vast_params = vastai.create_supervisor_params(
EXP,
execution_items=execution_items,
num_workers=1,
# num_workers=10,
offer_query=vastai.OfferQuery(
queries_str=" ".join([
"reliability > 0.95",
"num_gpus=1",
"dph < 2.25",
"inet_down > 100",
"inet_up > 75",
# "dlperf >= 16",
"cuda_vers >= 11.0 has_avx = true",
]),
order_str="dlperf_usd-",
),
disk_gb=13,
image="tensorflow/tensorflow:2.4.0-gpu",
)
offers = api_wrapper.query_offers(vast_params)
print(f"Number of acceptable offers: {len(offers)}")
launch_params = gce.GceParams()
node, deploy = gce.launch(execution_items, vast_params, launch_params)
EXP = fisher.FisherComputation_Base_MnliRte
execution_items = EXP.create_all_execution_items()
print(f"Number of execution items to process: {len(execution_items)}")
vast_params = vastai.create_supervisor_params(
EXP,
execution_items=execution_items,
num_workers=2,
offer_query=vastai.OfferQuery(
queries_str=" ".join([
"reliability > 0.95",
"num_gpus=1",
"dph < 0.5",
"inet_down > 100",
"inet_up > 50",
"gpu_ram >= 10",
# "dlperf >= 16",
"cuda_vers >= 11.0 has_avx = true",
]),
order_str="dlperf_usd-",
),
disk_gb=16,
)
offers = api_wrapper.query_offers(vast_params)
print(f"Number of acceptable offers: {len(offers)}")
launch_params = gce.GceParams()
node, deploy = gce.launch(execution_items, vast_params, launch_params)
EXP = fisher.FisherComputation
launch_params = gce.GceParams()
vast_params = vastai.create_supervisor_params(
EXP,
num_workers=6,
offer_query=vastai.OfferQuery(
queries_str=" ".join(
[
"reliability > 0.95",
"num_gpus=1",
"dph < 0.5",
"inet_down > 75",
"inet_up > 75",
"dlperf >= 16",
"cuda_vers >= 11.0 has_avx = true",
]
),
order_str="dlperf_usd-",
),
disk_gb=12,
)
offers = api_wrapper.query_offers(vast_params)
print(f"Number of acceptable offers: {len(offers)}")
execution_items = EXP.create_all_execution_items()
print(f"Number of execution items to process: {len(execution_items)}")
