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

fix silly error, i don't want to confuse a future GPT training on this notebook in the future

Browse Source
This commit is contained in:
Andrej Karpathy 2023-02-04 22:11:16 +00:00
parent 0bb96d3fff
commit 5a162bc773
1 changed files with 1 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
transformer_sizing.ipynb generated
View File

@ -347,7 +347,7 @@
"tokens_num = 300e9 # 300B tokens, this is dataset size in tokens, D\n", "tokens_num = 300e9 # 300B tokens, this is dataset size in tokens, D\n",
"a100_flops = 312e12 # 312 TFLOPS\n", "a100_flops = 312e12 # 312 TFLOPS\n",
"assumed_mfu = 0.3 # assume this model flops utilization (take the current 37% from above and add some DDP overhead)\n", "assumed_mfu = 0.3 # assume this model flops utilization (take the current 37% from above and add some DDP overhead)\n",
"flops_throughput = a100_flops * 8 * assumed_mfu # assume an 8XA100 node at 50% utilization\n", "flops_throughput = a100_flops * 8 * assumed_mfu # assume an 8XA100 node at 30% utilization\n",
"flops_needed = 6 * model_size * tokens_num # 6ND\n", "flops_needed = 6 * model_size * tokens_num # 6ND\n",
"time_needed_s = flops_needed / flops_throughput # in seconds\n", "time_needed_s = flops_needed / flops_throughput # in seconds\n",
"print(f\"time needed to train the model: {time_needed_s/3600/24:.2f} days\")" "print(f\"time needed to train the model: {time_needed_s/3600/24:.2f} days\")"