FlagOpen
finetuning loss不降低
第一epoch里面loss很快不变了，还有继续训练的必要吗？
{'loss': 1.043, 'learning_rate': 1e-05, 'epoch': 0.0}
{'loss': 1.0732, 'learning_rate': 9.997437868306432e-06, 'epoch': 0.01}
{'loss': 0.8342, 'learning_rate': 9.992313604919294e-06, 'epoch': 0.01}
{'loss': 0.7637, 'learning_rate': 9.987189341532154e-06, 'epoch': 0.01}
{'loss': 0.7827, 'learning_rate': 9.982065078145018e-06, 'epoch': 0.01}
{'loss': 0.7332, 'learning_rate': 9.976940814757879e-06, 'epoch': 0.02}
{'loss': 0.8379, 'learning_rate': 9.97181655137074e-06, 'epoch': 0.02}
{'loss': 0.7158, 'learning_rate': 9.966692287983603e-06, 'epoch': 0.02}
{'loss': 0.8149, 'learning_rate': 9.961568024596465e-06, 'epoch': 0.02}
{'loss': 0.7891, 'learning_rate': 9.956443761209327e-06, 'epoch': 0.03}
{'loss': 0.8286, 'learning_rate': 9.95131949782219e-06, 'epoch': 0.03}
{'loss': 0.7937, 'learning_rate': 9.946195234435051e-06, 'epoch': 0.03}
{'loss': 0.7175, 'learning_rate': 9.941070971047912e-06, 'epoch': 0.03}
{'loss': 0.7234, 'learning_rate': 9.935946707660774e-06, 'epoch': 0.04}
{'loss': 0.7764, 'learning_rate': 9.930822444273636e-06, 'epoch': 0.04}
{'loss': 0.6897, 'learning_rate': 9.925698180886498e-06, 'epoch': 0.04}
{'loss': 0.7483, 'learning_rate': 9.92057391749936e-06, 'epoch': 0.04}
{'loss': 0.697, 'learning_rate': 9.915449654112222e-06, 'epoch': 0.05}
{'loss': 0.7219, 'learning_rate': 9.910325390725085e-06, 'epoch': 0.05}
{'loss': 0.7715, 'learning_rate': 9.905201127337945e-06, 'epoch': 0.05}
{'loss': 0.7017, 'learning_rate': 9.900076863950809e-06, 'epoch': 0.05}
{'loss': 0.7283, 'learning_rate': 9.89495260056367e-06, 'epoch': 0.06}
{'loss': 0.7576, 'learning_rate': 9.889828337176531e-06, 'epoch': 0.06}
{'loss': 0.7246, 'learning_rate': 9.884704073789393e-06, 'epoch': 0.06}
{'loss': 0.8037, 'learning_rate': 9.879579810402256e-06, 'epoch': 0.06}
{'loss': 0.7993, 'learning_rate': 9.874455547015118e-06, 'epoch': 0.07}
{'loss': 0.7236, 'learning_rate': 9.86933128362798e-06, 'epoch': 0.07}
{'loss': 0.7871, 'learning_rate': 9.864207020240842e-06, 'epoch': 0.07}
{'loss': 0.7769, 'learning_rate': 9.859082756853702e-06, 'epoch': 0.07}
{'loss': 0.7354, 'learning_rate': 9.853958493466566e-06, 'epoch': 0.08}
{'loss': 0.76, 'learning_rate': 9.848834230079427e-06, 'epoch': 0.08}
loss很小，建议挖掘难负样本，提高训练难度：https://github.com/FlagOpen/FlagEmbedding/tree/master/examples/finetune#hard-negatives
