volcengine/veScale

Issues

• [ENHANCEMENT] Extending to DeepSpeed

  #57 opened a month ago by moghadas76
  0

• [QUESTION]How is vescale zero2 implemented?

  #54 opened 2 months ago by starstream
  5

• [QUESTION]How to Use ndtimeline in a Multi-Machine Multi-GPU Environment

  #55 opened 2 months ago by zmtttt
  4

• [QUESTION] questions about Collective Communication Group Initialization Optimization in the paper

  #40 opened 2 months ago by siddharthaOnRoad
  2

• [QUESTION]How to use MQhandler for muti machines？

  #56 opened 2 months ago by zmtttt
  2

• > Using ndtimeline-tool to Monitor Megatron-GPT I want to use the ndtimeline-tool to monitor the computation and communication of each rank in Megatron-GPT. I have two concerns:

  #53 opened 2 months ago by zmtttt
  3

• [QUESTION]Using ndtimeline-tool to Monitor Megatron-GPT

  #51 opened 2 months ago by zmtttt
  1

• [RFC] Single-Device-Abstract DDP

  #52 opened 3 months ago by lllukehuang
  1

• The times for forward-compute and backward-compute captured by the ndtimeline-tool are inaccurate

  #47 opened 3 months ago by zmtttt
  10

• [QUESTION] implementation of `get_p2p_cuda_stream_id` and `get_coll_cuda_stream_id`

  #46 opened 3 months ago by nooblyh
  5

• [QUESTION] How flexible is the veScale checkpoint

  #43 opened 4 months ago by Dream-Seeker123
  3

• [QUESTION] how and where to use multi-node trace profiler in paper of megascale

  #37 opened 4 months ago by oliverYoung2001
  3

• [RFC] veScale: High-Level API for nD Parallel Training

  #39 opened 6 months ago by leonardo0lyj
  0

• [QUESTION]`vescale.dtensor` vs "PyTorch DTensor"

  #28 opened 8 months ago by GHGmc2
  3

• [QUESTION] Save checkpoint

  #26 opened 8 months ago by Ryanuppp
  1

• [QUESTION] will the patch code merged into upstream?

  #17 opened 8 months ago by ultranity
  2

• Code Example & Docs

  #14 opened 8 months ago by ultranity
  2