Replies: 2 comments
-
从提供的日志和问题来看, 1. 是否因为高度为 64 导致爆显存?是的,高度设置为 64 的确可能导致显存过高甚至爆显存。 原因:
日志分析:
虽然日志中的显存差别并不是特别明显,但在更大批量( 2. 将高度改为 24 是否可行?有什么影响?将高度改为较小的值(如 24)是可行的,但需要权衡以下影响: 优点:
缺点:
日志分析:
3. 优化建议
结论:
Response generated by feifei-bot | chatgpt-4o-latest |
Beta Was this translation helpful? Give feedback.
0 replies
-
[[320, 32], [320, 48], [320, 64]] 原先默认的32,48高度没变,先继续用改后的尺寸试试,看会不会爆 |
Beta Was this translation helpful? Give feedback.
0 replies
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
-
sampler: name: MultiScaleSampler scales: [[320, 32], [320, 48], [320, 64]] 这个64高度会不会是造成爆显存的原因?64改成24之类的高度有什么问题吗?
64高度
[2024/12/29 21:06:27] ppocr INFO: epoch: [1/5], global_step: 26600, lr: 0.000002, CTCLoss: 12.595758, NRTRLoss: 2.748089, loss: 15.519665, avg_reader_cost: 0.00189 s, avg_batch_cost: 0.14987 s, avg_samples: 1.32, ips: 8.80787 samples/s, eta: 29 days, 14:40:47, max_mem_reserved: 1952 MB, max_mem_allocated: 1756 MB
64换成24高度
[2025/01/06 08:02:12] ppocr INFO: epoch: [1/50], global_step: 600, lr: 0.000000, CTCLoss: 33.865753, NRTRLoss: 4.422851, loss: 38.640900, avg_reader_cost: 0.00187 s, avg_batch_cost: 0.18453 s, avg_samples: 2.43, ips: 13.16867 samples/s, eta: 204 days, 1:23:32, max_mem_reserved: 2065 MB, max_mem_allocated: 1814 MB
Beta Was this translation helpful? Give feedback.
All reactions