Issues
- 0
- 2
执行第二步出错
#32 opened - 3
- 1
【讨论】gpt2-ml,30G,22w步模型微调报错解决方案
#31 opened by NLPIG - 0
- 9
可以给一个输入数据的例子吗,就是 pre_data.py的输入文件格式
#10 opened by EiraZhang - 0
请问一下 batch_size=1的微调情况下,learning rate的选择?
#28 opened by huangdacheng - 9
自己的数据集需要多大效果会比较好
#7 opened by fred-github - 0
- 1
CUDA_VISIBLE_DEVICES=0 是做什么的呢?
#8 opened by lrz512699597 - 3
train/train_wc.py的输出路径是不是改为原模型地址
#20 opened by cncbec - 1
为啥会出现:is not in all_model_checkpoint_paths
#23 opened by cncbec - 0
finetuning的batchsize是1 就可以了?
#24 opened by huangdacheng - 3
- 0
22.json可以生成,train.tfrecord生成不了
#22 opened by cncbec - 0
loss下降到多少是差不多了
#21 opened by joytianya - 1
bs=1,训练有效果么
#16 opened by joytianya - 5
- 1
加载原gpt2-ml的大模型,9G的显存也会满了,报OOM,请问作者是使用多大显存?
#17 opened by huangdacheng - 1
生成tfrecord这一步时出错
#14 opened by dakkor - 2
- 0
finetune时,总是找不到预训练模型文件是怎么回事呢?
#9 opened by lrz512699597 - 3
pre_data.py
#6 opened by lrz512699597 - 1
- 0
"iterations_per_loop", 1600,
#5 opened by lrz512699597 - 0
能不能使用多机多卡训练,小作坊用的是4*4 1080ti,
#4 opened by SeekPoint