From ddbba17b04b3fc2929ce595a9b13055771214327 Mon Sep 17 00:00:00 2001 From: Priya Kasimbeg Date: Tue, 26 Sep 2023 00:58:55 +0000 Subject: [PATCH] change bsz back --- .../workloads/librispeech_conformer/workload.py | 2 +- .../target_setting_algorithms/get_batch_size.py | 2 +- submission_runner.py | 2 +- 3 files changed, 3 insertions(+), 3 deletions(-) diff --git a/algorithmic_efficiency/workloads/librispeech_conformer/workload.py b/algorithmic_efficiency/workloads/librispeech_conformer/workload.py index b19593644..dc7fb912b 100644 --- a/algorithmic_efficiency/workloads/librispeech_conformer/workload.py +++ b/algorithmic_efficiency/workloads/librispeech_conformer/workload.py @@ -55,7 +55,7 @@ def num_test_examples(self) -> int: @property def eval_batch_size(self) -> int: - return 32 + return 256 @property def train_mean(self): diff --git a/reference_algorithms/target_setting_algorithms/get_batch_size.py b/reference_algorithms/target_setting_algorithms/get_batch_size.py index 9f76d86d1..2351e8185 100644 --- a/reference_algorithms/target_setting_algorithms/get_batch_size.py +++ b/reference_algorithms/target_setting_algorithms/get_batch_size.py @@ -12,7 +12,7 @@ def get_batch_size(workload_name): elif workload_name == 'imagenet_vit': return 1024 elif workload_name == 'librispeech_conformer': - return 32 + return 256 elif workload_name == 'librispeech_deepspeech': return 256 elif workload_name == 'ogbg': diff --git a/submission_runner.py b/submission_runner.py index 5519399a8..71f7ec552 100644 --- a/submission_runner.py +++ b/submission_runner.py @@ -426,7 +426,7 @@ def train_once( train_state['last_step_end_time'] = get_time() if global_step == 2: s = torch.cuda.memory._snapshot() - with open(f"/experiment_runs/snapshot_{global_step}_full_bsz_adamw.pickle", "wb") as f: + with open(f"/experiment_runs/snapshot_{global_step}_full_bsz_adamw_reduced_alloc_size.pickle", "wb") as f: dump(s, f) metrics = {'eval_results': eval_results, 'global_step': global_step}