osmarks/nanogpt-experiments
1
0
Fork 0
mirror of https://github.com/osmarks/nanogpt-experiments.git synced 2024-11-10 20:09:58 +00:00
Code Issues Releases Wiki Activity

correctly propagate the vocab_size from the rendered dataset into the model args

Browse Source
This commit is contained in:
Andrej Karpathy 2023-01-14 02:26:44 +00:00
parent 43b37fd568
commit 57735f532d
1 changed files with 14 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

16
train.py
View File

@ -12,6 +12,7 @@ $ torchrun --standalone --nproc_per_node=4 train.py
import os
import time
import math
import pickle
from contextlib import nullcontext
import numpy as np
@ -102,8 +103,19 @@ def get_batch(split):
iter_num = 0
best_val_loss = 1e9
# model init. TODO: fix bug we should also propagate the correct vocab_size to the model_args
model_args = dict(n_layer = n_layer, n_head = n_head, n_embd = n_embd, block_size = block_size, dropout = dropout)
# attempt to derive vocab_size from the dataset
meta_path = os.path.join(data_dir, 'meta.pkl')
if os.path.exists(meta_path):
with open(meta_path, 'rb') as f:
meta = pickle.load(f)
vocab_size = meta['vocab_size']
print(f"vocab_size = {vocab_size} (from {meta_path})")
else:
print(f"vocab_size not found in {meta_path}, using GPT-2 default of 50257")
vocab_size = 50257
# model init
model_args = dict(n_layer = n_layer, n_head = n_head, n_embd = n_embd, block_size = block_size, dropout = dropout, vocab_size = vocab_size)
if init_from == 'scratch':
# init a new model from scratch
print("Initializing a new model from scratch")