osmarks/nanogpt-experiments
1
0
Fork 0
mirror of https://github.com/osmarks/nanogpt-experiments.git synced 2025-10-31 07:13:01 +00:00
Code Issues Releases Wiki Activity

small tweaks to docs and variable names stylistically

Browse Source
This commit is contained in:
Andrej Karpathy
2023-01-16 16:56:05 +00:00
parent 684800dd87
commit 46ce9971df
1 changed files with 8 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files

15
train.py
View File

@@ -9,10 +9,11 @@ To run with DDP on 4 gpus on 1 node, example:
$ torchrun --standalone --nproc_per_node=4 train.py $ torchrun --standalone --nproc_per_node=4 train.py
To run with DDP on 4 gpus across 2 nodes, example: To run with DDP on 4 gpus across 2 nodes, example:
- Run on the first (master) node: - Run on the first (master) node with example IP 123.456.123.456:
$ torchrun --nproc_per_node=8 --nnodes=2 --node_rank=0 --master_addr=123.456.123.456 --master_port=1234 train.py $ torchrun --nproc_per_node=8 --nnodes=2 --node_rank=0 --master_addr=123.456.123.456 --master_port=1234 train.py
- Run on the worker node: - Run on the worker node:
$ torchrun --nproc_per_node=8 --nnodes=2 --node_rank=1 --master_addr=123.456.123.456 --master_port=1234 train.py $ torchrun --nproc_per_node=8 --nnodes=2 --node_rank=1 --master_addr=123.456.123.456 --master_port=1234 train.py
(If your cluster does not have Infiniband interconnect prepend NCCL_IB_DISABLE=1)
""" """
import os import os
@@ -79,11 +80,11 @@ config = {k: globals()[k] for k in config_keys} # will be useful for logging
ddp = int(os.environ.get('RANK', -1)) != -1 # is this a ddp run? ddp = int(os.environ.get('RANK', -1)) != -1 # is this a ddp run?
if ddp: if ddp:
init_process_group(backend=backend) init_process_group(backend=backend)
DDP_RANK = int(os.environ['RANK']) ddp_rank = int(os.environ['RANK'])
DDP_LOCAL_RANK = int(os.environ['LOCAL_RANK']) ddp_local_rank = int(os.environ['LOCAL_RANK'])
device = f'cuda:{DDP_LOCAL_RANK}' device = f'cuda:{ddp_local_rank}'
master_process = DDP_RANK == 0 # this process will do logging, checkpointing etc. master_process = ddp_rank == 0 # this process will do logging, checkpointing etc.
seed_offset = DDP_RANK # each process gets a different seed seed_offset = ddp_rank # each process gets a different seed
else: else:
# if not ddp, we are running on a single gpu, and one process # if not ddp, we are running on a single gpu, and one process
master_process = True master_process = True
@@ -181,7 +182,7 @@ if compile:
# wrap model into DDP container # wrap model into DDP container
if ddp: if ddp:
model = DDP(model, device_ids=[DDP_LOCAL_RANK]) model = DDP(model, device_ids=[ddp_local_rank])
@torch.no_grad() @torch.no_grad()
def estimate_loss(): def estimate_loss():