microsoft/DeepSpeedExamples

Issues

• Failed to run Domino example

  #940 opened a month ago by lucifer1004
  0

• How can I change the master_port when using deepspeed for multi-GPU on single node, i.e. localhost

  #936 opened a month ago by lovedoubledan
  4

• RuntimeError: CUDA error: no kernel image is available for execution on the device

  #935 opened 2 months ago by mrpeerat
  1

• FileNotFoundError: [Errno 2] No such file or directory: 'numactl'

  #920 opened 2 months ago by zhiwentian
  6

• AttributeError： 'DeepSpeedEngine' object has no attribute 'model'，

  #924 opened 2 months ago by lovychen
  1

• No module named 'transformers.deepspeed'

  #934 opened 2 months ago by TianyuJIAA
  2

• Does DeepSpeed's Pipeline-Parallelism optimizer supports skip connections?

  #932 opened 2 months ago by RoyMahlab
  0

• After using steps 1, 2, and 3, the test reply content only replies Assistant: </s>。

  #928 opened 3 months ago by jianmomo
  0

• How to calculate training efficiency ,i.e tokens/sec of step 1 fine tuning of llama2 model ?

  #923 opened 4 months ago by sowmya04101998
  0

• Actor loss nan and Resizing model embedding

  #922 opened 4 months ago by ouyanmei
  1

• 不同机器上python环境变量路径不同，deepspeed启动后发现找不到其他机器的python环境，如何解决

  #889 opened 4 months ago by liqwertyu
  0

• How to start deepspeed automatically?

  #910 opened 4 months ago by qwerfdsadad
  2

• zero3 and enable hybrid engine are not suitable for llama2, how to solve it?

  #864 opened 10 months ago by terence1023
  3

• The actor constantly generates ['</s>'] or ['<|endoftext|></s>'] after 200 steps in RLHF with hybrid engine disabled

  #887 opened 9 months ago by mousewu
  1

• step2 without any response for a long time

  #915 opened 4 months ago by asfadfaf
  0

• Consult the first phase.

  #909 opened 4 months ago by csxrzhang
  2

• 单机多卡进行RLHF在第三步中使用Qwen模型作Actor Model报错

  #907 opened 6 months ago by Dakai798
  1

• run-example.sh fails with urllib3.exceptions.ProtocolError: Response ended prematurely

  #896 opened 5 months ago by awan-10
  11

• Different zero stage the training memory compute

  #912 opened 5 months ago by Arcmoon-Hu
  0

• nvcc fatal : Unsupported gpu architecture 'compute_86' and nvcc fatal : Value 'c++17' is not defined for option 'std'

  #911 opened 5 months ago by Xccanxin
  1

• an error with gradient checkpointing in DeepspeedChat step 3

  #908 opened 6 months ago by wangyuwen1999
  0

• DeepSpeed-Chat step-1 hanging for a long time

  #906 opened 6 months ago by lemon-little
  0

• CPU OOM when inferencing Llama3-70B-Chinese-Chat

  #904 opened 7 months ago by GORGEOUSLCX
  0

• Confusion about Deepspeed Inference

  #879 opened 9 months ago by ZekaiGalaxy
  1

• cannot pickle 'Stream' object

  #903 opened 7 months ago by teis-e
  0

• can not run the test-gpt.sh because of assertionError

  #902 opened 7 months ago by leachee99
  0

• 请问fastgen 是否支持长文本和序列并行推理

  #901 opened 7 months ago by AceCoder0
  0

• [Error] AutoTune: `connect to host localhost port 22: Connection refused`

  #894 opened 8 months ago by wqw547243068
  0

• How to use deepspeed for multi-node and multi-card task in slurm cluster

  #893 opened 8 months ago by dshwei
  0

• Does Zero-Inference support TP?

  #892 opened 8 months ago by preminstrel
  11

• Deepspeed support finetune extra model with lora ?

  #890 opened 8 months ago by wanghongqu
  1

• when calculating actor loss, why the mask is "action_mask[:, start: ] "

  #888 opened 8 months ago by fancghit
  0

• About multiple-thread attention computation on CPU using zero-inference example.

  #886 opened 9 months ago by luckyq
  0

• Suggested GPU to run the demo code of step2_reward_model_finetuning (DeepSpeed-Chat)

  #885 opened 9 months ago by wenbozhangjs
  0

• [REQUEST] More fine-grained distributed strategies for RLHF training

  #884 opened 9 months ago by krisx0101
  0

• RLHF problems when using Qwen model

  #861 opened 10 months ago by 128Ghe980
  1

• The reward value did not increase.

  #883 opened 9 months ago by Sun-Shiqi
  1

• `AttributeError: readonly attribute` while trying to run training/HelloDeepSpeed

  #878 opened 9 months ago by htjain
  0

• [BUG in Stable Diffusion inference] There's an error on CUDAGraph when using deepspeed inference. How to fix it?

  #866 opened 10 months ago by foin6
  2

• [Bug] DeepSpeed Inference Does not Work with LLaMA (Latest verison)

  #867 opened 10 months ago by allanj
  3

• Codellama finetune

  #860 opened 10 months ago by nani1149
  0

• Throughput should be `num_queries/latency` as opposed to `num_clients/latency`?

  #858 opened a year ago by goelayu
  0

• The inaccurate flop results after several rounds

  #855 opened a year ago by BitCalSaul
  1

• How to resume Deepspeed-Chat RLHF step-3 training?

  #850 opened a year ago by DespairL
  0

• remove redundant code

  #852 opened a year ago by ilml
  0

• Question: Why not padding to the same sequence length within the batch during the sft training phase?

  #849 opened a year ago by LKLKyy
  0

• running gpt2-xl/test_tune.sh fails - ParquetConfig.__init__() got an unexpected keyword argument 'token'

  #847 opened a year ago by ccruttjr
  0

• torch.distributed.DistBackendError: NCCL error in: ../torch/csrc/distributed/c10d/ProcessGroupNCCL.cpp:1333, remote process exited or there was a network error, NCCL version 2.18.6

  #845 opened a year ago by Rainbowman0
  3

• Step3 hanging for a long time

  #842 opened a year ago by Jeayea
  1

• Invalidate trace cache @ step 0: expected module 0, but got module 6

  #840 opened a year ago by boundles
  0