janelu9
janelu9
How to improve the speed of lightgbm on spark?
if i con't run lighgbm on yarn. the original version rather than the spark version may be the best choice
然后acc就会远远低于之前训练好的模型
HTTPConnectionPool(host='127.0.0.1', port=8000): Max retries exceeded with url: /v1/auth/login/ (Caused by NewConnectionError(': Failed to establish a new connection: [Errno 111] Connection refused'))
deespeed==0.12.6 I do believe that the distribution of parameters is more reasonable, but why slower the training speed becomed?
model: deepseek-llm-67b-chat -awq-int4 tp = 2 gpu: A100-40G x2 v2.0.7: 16tokens/s v2.0.6: 21tokens/s