hiyouga/ChatGLM-Efficient-Tuning

Issues

• 微调后不起效果，是我哪步有问题吗

  #406 opened a year ago by RemSynch
  1

• CUDA error: an illegal memory access was encountered

  #405 opened a year ago by suckseed5
  1

• epoch的选择&数据集构造

  #404 opened a year ago by 1028686314
  1

• step和epoch一样

  #401 opened a year ago by qingjiaozyn
  1

• 用默认参数微调chatglm2之后对话能力大幅度下降

  #402 opened a year ago by caoxueqian19901214
  1

• 仓库过大，git下载慢

  #398 opened a year ago by JackKuo666
  2

• 请问下在4-bit量化模式Lora微调最小的显存要求

  #394 opened a year ago by NJU201220184
  1

• 4*V100 32g配置，满足zero3全量微调吗？

  #392 opened a year ago by gongjunjin
  4

• ValueError: Cannot merge LORA layers when the model is loaded in 8-bit mode

  #390 opened a year ago by jeriff
  0

• 显存占用问题

  #389 opened a year ago by lileilai
  0

• ptuning后，模型参数不匹配

  #383 opened a year ago by HLF217
  0

• 为什么加载sft权重计算的指标和sft+ppo微调的权重计算的指标是一样的？

  #382 opened a year ago by JingyuHuAtVoxelCloud
  4

• [BUG] PPO训练报错 RuntimeError: Expected all tensors to be on the same device

  #381 opened a year ago by qinqinqaq
  1

• 为什么glm2训练需要较大的学习率才有效

  #379 opened a year ago by JingBob
  4

• lora后模型的泛化能力如何增强呢？

  #377 opened a year ago by mawenju203
  2

• 多卡微调错误：Exception: Could not find the transformer layer class to wrap in the model.

  #378 opened a year ago by lrx1213
  0

• 请问用Lora微调2万条医疗问答，为什么Loss只能降低到2.7，效果不佳

  #375 opened a year ago by tongjialiang
  5

• 使用train_sft_with_dev_set.sh进行 lora的微调 如何平衡 'loss' 和 'eval_loss'

  #369 opened a year ago by mawenju203
  2

• LoRA微调已经训练好了，关于Evaluate和Predict的几个疑问

  #368 opened a year ago by FreeYiran
  1

• 多机多卡训练是否有相关资料呢？

  #367 opened a year ago by yifan-chen-2020
  2

• 模型训练的问题？ 模型出现过拟合

  #360 opened a year ago by mawenju203
  2

• 训练成功 启动测试的时候报错了

  #356 opened a year ago by bubblebyfly
  4

• 关于sft之后，奖励模型和rlhf

  #372 opened a year ago by lzh1998-jansen
  3

• 数据集oaast_rm_zh问题

  #403 opened a year ago by PluginsKers
  0

• 有关ChatGLM-Efficient-Tuning + lora微调出现过拟合的讨论

  #365 opened a year ago by mawenju203
  3

• 请问lora训练是只会训练注意力层，在注意力层加适配器吗？而不训练前馈层。然后freeze是训练前馈层

  #387 opened a year ago by DRZJ1
  1

• lora微调之后导出的完整模型文件应该要如何调用 我使用transformers的接口加载模型报错

  #388 opened a year ago by ecent2000
  2

• 请问如何制作自己的数据集？

  #376 opened a year ago by andyzhou1982
  4

• lora_target 的可用值有哪些？

  #400 opened a year ago by godcrying
  2

• when `per_device_eval_batch_size` > 1 and launch by deepspeed, RuntimeError: Tensors must be contiguous

  #385 opened a year ago by jiahuanluo
  5

• 不理解为什么这个模型是paddingleft. 编码之后开始全是一堆-100的token, 感觉很难llm收敛.

  #399 opened a year ago by zhangbo2008
  0

• 跑sft阶段出现这个问题，环境应该是ok的

  #397 opened a year ago by wangweihua11
  2

• 2400条数据,10个epoch,pre_seq_len=128,lora训练,为什么推理时显示trainable params: 0 || all params: 6243584000 || trainable%: 0.0000,是数据量不够吗,下边贴出训练参数

  #396 opened a year ago by zeze813
  1

• Learning Scheduler Issue

  #393 opened a year ago by XingzhiZhou
  2

• collator.py的第126行与preprocess.py的preprocess_supervised_dataset是不是存在冲突？input_ids拼接了2次labels

  #391 opened a year ago by jennfer0808
  2

• 如何设置输入长度

  #363 opened a year ago by OpenHuShen
  5

• chatglm2进行lora微调，怎么执行断点续训呢？

  #384 opened a year ago by Qiang-HU
  2

• 单机4卡3090训练lora报错

  #371 opened a year ago by 1006076811
  3

• 请问用example/里的quantized_finetune_with_local_model微调后怎么正确加载output并启动demo

  #364 opened a year ago by R0nnyz
  1

• RuntimeError: expected scalar type Half but found Float

  #370 opened a year ago by jazzlee008
  3

• 关于PPO中ref_model的问题

  #380 opened a year ago by JoyYChen
  9

• ModuleNotFoundError: No module named 'transformers_modules.' 遇到这个bug

  #355 opened a year ago by bubblebyfly
  2

• 在服务器上执行 src/train_web.py 报错

  #361 opened a year ago by FreeYiran
  4

• QLoRA报错expected scalar type Half but found Float

  #357 opened a year ago by wangweikang1229
  1

• datasets.builder.DatasetGenerationError: An error occurred while generating the dataset

  #374 opened a year ago by ArtificialZeng
  0

• 建议对 data/*.json 文件使用 lfs 管理

  #366 opened a year ago by Haskely
  3

• 请问int8版本的ppo怎么运行

  #373 opened a year ago by couldn
  2

• 训练不小心中断后，如何设置进行断点续训？

  #362 opened a year ago by FreeYiran
  9

• 微信群满啦，麻烦更新下二维码

  #359 opened a year ago by hongshuo-wang
  1

• 在web_demo_finetuned.py运行ChatGLM（不是ChatGLM2）时报ValueError: not enough values to unpack (expected 3, got 2)

  #358 opened a year ago by yug030
  0