diff --git a/bench.py b/bench.py index 60db2bb..840dcc5 100644 --- a/bench.py +++ b/bench.py @@ -15,7 +15,7 @@ torch.manual_seed(1337) batch_size = 8 block_size = 1024 dtype = torch.bfloat16 -compile_model = True +compile = True # data loading init real_data = True @@ -47,7 +47,7 @@ model.to(device) optimizer = model.configure_optimizers(weight_decay=1e-2, learning_rate=1e-4, betas=(0.9, 0.95)) -if compile_model: +if compile: print("Compiling model...") model = torch.compile(model) # pytorch 2.0 diff --git a/config/finetune_shakespeare.py b/config/finetune_shakespeare.py index 69d49a6..eb6545e 100644 --- a/config/finetune_shakespeare.py +++ b/config/finetune_shakespeare.py @@ -5,7 +5,7 @@ eval_interval = 200 wandb_log = False # feel free to turn on wandb_project = 'shakespeare' wandb_run_name = 'ft-' + str(time.time()) -compile_model = False # takes too little time to finetune, not worth it +compile = False # takes too little time to finetune, not worth it # save a nice and overfit checkpoint that # will only speak Shakespeare and forgets