Repository Issue Activity (beta)

#996 Compute_rewards in PPO:rewards[j, start:ends[j]][-1] += reward_clip[j] is wrong

#995 Step1 failed when I run "bash training_scripts/opt/single_gpu/run_1.3b.sh"

#524 RuntimeError: Error building extension 'transformer_inference'

#989 One example, multiple config files

#988 Gradient overflow issue while using deepspeed

#703 how to understand the code for calculating rewards

#986 Is there a micro-benchmark that can test the last two optimizations mentioned in the FastPersist paper, "Parallelizing checkpoint writes across DP ranks and Pipelining checkpoint writes"?

#984 moe example 404

#186 OOM despite ZeRO stage 3

#977 FastPersist micro-benchmarks test results are inconsistent with expectations

#979 Possible to include an example of DeepNVMe + state dict

#943 KV_cache offload

#941 A bug in argument parser.

#172 My deepspeed code is very slow

#956 Why Does vf_loss Take the Maximum Value, Rendering Clamp Meaningless?

#969 Apply Zero-3 and LoRA appears empty lora weight [0]

#845 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

#972 DeepSpeed-Chat step-1 take a long time and no response

#960 DeepSpeed-FastGen support ascend npu?

#892 Does Zero-Inference support TP?

#831 [Discussion] Can anyone show the performance on every step with any dataset

#946 Assertion `srcIndex < srcSelectDimSize` failed

#456 enable_hybrid_engine issue

#951 Is there any example about DeepSpeed Zero with Ulysses/Ulysses-offload

#950 Domino + PP