sunzeyeah/RLHF

Issues

• 基于ChatGLM2的RLHF训练问题

  #23 opened a year ago by UltraZeroyH
  2

• Pangu 2.6b 启动失败。

  #25 opened a year ago by Liufeiran123
  3

• 请问为什么我在基于Chatglm-6b模型执行train_sft.sh训练时，启用deepspeed没有效果

  #20 opened 10 months ago by IconShan
  2

• pretrain_data_v1.jsonl 文件在哪里

  #24 opened a year ago by Liufeiran123
  5

• 后续会考虑RLHF平替方案的集成么

  #21 opened a year ago by skykiseki
  2

• 有这个推理的代码吗

  #22 opened a year ago by ww0o0
  0

• No module named 'transformers_modules.sunzeyeah.pangu-2'

  #8 opened 2 years ago by MRKINKI
  4

• rlhf deepspeed和trlx能否支持 sft chatglm 6b

  #13 opened 2 years ago by GUORUIWANG
  13

• reward推理问题

  #16 opened 2 years ago by ItGirls
  5

• chatglm+RLHF

  #18 opened 2 years ago by MAJIN123
  1

• 请问支持lora方式吗

  #19 opened 2 years ago by 70557dzqc
  1

• train_rlhf-trlx.py代码问题

  #15 opened 2 years ago by taofennanhai
  3

• 关于取最后一个token作为reward分数的方式

  #17 opened 2 years ago by Bo396543018
  2

• 用chatGLM-6B训RW的时候loss不收敛

  #10 opened 2 years ago by GUORUIWANG
  7

• 请教一下有考虑加IDEA-CCNL/Wenzhong2.0-GPT2-3.5B-chinese这个模型做评测比较吗

  #7 opened 2 years ago by MRKINKI
  1

• 有对比不加RLHF和加入RLHF的效果吗

  #4 opened 2 years ago by macheng6
  1

• reward model的实现问题

  #12 opened 2 years ago by DamonYangyang
  1

• RLHF第三步进行训练时，在modeling_glm.py如下代码位置会显示“size mismatch, got 1028, 1028x1024,0”

  #14 opened 2 years ago by IconShan
  4

• 使用LoRA的GLM-10B-chinese模型是如何保存的

  #9 opened 2 years ago by taofennanhai
  3

• 用GLM-10B-chinese训练RLHF过程，有没有模型并行的方式？

  #11 opened 2 years ago by taofennanhai
  4

• RLHF相关问题

  #5 opened 2 years ago by taofennanhai
  4

• deepspeed速度

  #6 opened 2 years ago by superqing001
  0

• 训练相关

  #3 opened 2 years ago by macheng6
  2

• 为什么训练的时候要加入<sep> token？

  #2 opened 2 years ago by Nipi64310
  3

• 加入RW后模型的效果

  #1 opened 2 years ago by yxk9810
  2