1
0
mirror of https://github.com/osmarks/nanogpt-experiments.git synced 2024-12-18 14:10:28 +00:00

small usability tweaks to bench

This commit is contained in:
Andrej Karpathy 2023-02-02 17:23:46 +00:00
parent d995c22128
commit d01863ef01

View File

@ -9,13 +9,15 @@ import torch
from model import GPTConfig, GPT
# -----------------------------------------------------------------------------
batch_size = 8
batch_size = 12
block_size = 1024
bias = True
bias = False
real_data = True
seed = 1337
device = 'cuda' # examples: 'cpu', 'cuda', 'cuda:0', 'cuda:1', etc.
dtype = 'bfloat16' # 'float32' or 'bfloat16' or 'float16'
compile = True # use PyTorch 2.0 to compile the model to be faster
profile = False # use pytorch profiler, or just simple benchmarking?
exec(open('configurator.py').read()) # overrides from command line or config file
# -----------------------------------------------------------------------------
@ -28,7 +30,6 @@ ptdtype = {'float32': torch.float32, 'bfloat16': torch.bfloat16, 'float16': torc
ctx = nullcontext() if device_type == 'cpu' else torch.amp.autocast(device_type=device_type, dtype=ptdtype)
# data loading init
real_data = True
if real_data:
dataset = 'openwebtext'
data_dir = os.path.join('data', dataset)
@ -62,7 +63,6 @@ if compile:
print("Compiling model...")
model = torch.compile(model) # pytorch 2.0
profile = False # use pytorch profiler, or just simple benchmarking?
if profile:
# useful docs on pytorch profiler:
# - tutorial https://pytorch.org/tutorials/intermediate/tensorboard_profiler_tutorial.html
@ -73,10 +73,10 @@ if profile:
activities=[torch.profiler.ProfilerActivity.CPU, torch.profiler.ProfilerActivity.CUDA],
schedule=torch.profiler.schedule(wait=wait, warmup=warmup, active=active, repeat=1),
on_trace_ready=torch.profiler.tensorboard_trace_handler('./bench_log'),
record_shapes=True,
profile_memory=True,
with_stack=True, # incurs an additional overhead, disable if not needed
with_flops=True,
record_shapes=False,
profile_memory=False,
with_stack=False, # incurs an additional overhead, disable if not needed
with_flops=False,
with_modules=False, # only for torchscript models atm
) as prof: