Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

[train_engine] support fsdp #2412

Merged
merged 26 commits into from
Apr 7, 2024
Merged

[train_engine] support fsdp #2412

merged 26 commits into from
Apr 7, 2024

Commits on Mar 15, 2024

  1. [train_engine] support fsdp

    @Mddct
    Mddct committed Mar 15, 2024
    Configuration menu
    Copy the full SHA
    9406947 View commit details
    Browse the repository at this point in the history

  2. [train_engine] support fsdp

    @Mddct
    Mddct committed Mar 15, 2024
    Configuration menu
    Copy the full SHA
    67c1451 View commit details
    Browse the repository at this point in the history

  3. unify scaler and amp

    @Mddct
    Mddct committed Mar 15, 2024
    Configuration menu
    Copy the full SHA
    eb5c720 View commit details
    Browse the repository at this point in the history

  4. fp32&&fp16 works in fsdp env

    @Mddct
    Mddct committed Mar 15, 2024
    Configuration menu
    Copy the full SHA
    62e018d View commit details
    Browse the repository at this point in the history

  5. Merge branch 'main' into Mddct-fsdp

    @Mddct
    Mddct committed Mar 15, 2024
    Configuration menu
    Copy the full SHA
    3475bff View commit details
    Browse the repository at this point in the history

  6. fix fsdp in cv auto cast

    @Mddct
    Mddct committed Mar 15, 2024
    Configuration menu
    Copy the full SHA
    d13de47 View commit details
    Browse the repository at this point in the history

  7. try to fix wenet.join fsdp

    @Mddct
    Mddct committed Mar 15, 2024
    Configuration menu
    Copy the full SHA
    05ccbf3 View commit details
    Browse the repository at this point in the history

Commits on Mar 17, 2024

  1. implementing zero1 under fsdp is almost equivalent to deepspeed's zero1

    @Mddct
    Mddct committed Mar 17, 2024
    Configuration menu
    Copy the full SHA
    caa97c6 View commit details
    Browse the repository at this point in the history

Commits on Mar 18, 2024

  1. fix clip_and_grad_

    @Mddct
    Mddct committed Mar 18, 2024
    Configuration menu
    Copy the full SHA
    24639e4 View commit details
    Browse the repository at this point in the history

  2. fix train summary

    @Mddct
    Mddct committed Mar 18, 2024
    Configuration menu
    Copy the full SHA
    eda080d View commit details
    Browse the repository at this point in the history

  3. merge main

    @Mddct
    Mddct committed Mar 18, 2024
    Configuration menu
    Copy the full SHA
    c2a832a View commit details
    Browse the repository at this point in the history

Commits on Mar 22, 2024

  1. all wenet xxxformer works (-paraformer -transducer)

    @Mddct
    Mddct committed Mar 22, 2024
    Configuration menu
    Copy the full SHA
    0644294 View commit details
    Browse the repository at this point in the history

  2. Merge branch 'main' into Mddct-fsdp

    @Mddct
    Mddct committed Mar 22, 2024
    Configuration menu
    Copy the full SHA
    2b966f0 View commit details
    Browse the repository at this point in the history

Commits on Mar 23, 2024

  1. try to fix nan

    @Mddct
    Mddct committed Mar 23, 2024
    Configuration menu
    Copy the full SHA
    603c600 View commit details
    Browse the repository at this point in the history

Commits on Mar 24, 2024

  1. add barrier for cv

    @Mddct
    Mddct committed Mar 24, 2024
    Configuration menu
    Copy the full SHA
    979c3e1 View commit details
    Browse the repository at this point in the history

  2. add destroy group for end of all train

    @Mddct
    Mddct committed Mar 24, 2024
    Configuration menu
    Copy the full SHA
    5f5010a View commit details
    Browse the repository at this point in the history

Commits on Mar 26, 2024

  1. refactor wrap methods and ckpt works

    @Mddct
    Mddct committed Mar 26, 2024
    Configuration menu
    Copy the full SHA
    79448ff View commit details
    Browse the repository at this point in the history

  2. fix ckpt

    @Mddct
    Mddct committed Mar 26, 2024
    Configuration menu
    Copy the full SHA
    f7aea43 View commit details
    Browse the repository at this point in the history

Commits on Mar 27, 2024

  1. fix cv in dtype != float32

    @Mddct
    Mddct committed Mar 27, 2024
    Configuration menu
    Copy the full SHA
    1740e39 View commit details
    Browse the repository at this point in the history

Commits on Mar 30, 2024

  1. fix ckpt in model mode

    @Mddct
    Mddct committed Mar 30, 2024
    Configuration menu
    Copy the full SHA
    6008bab View commit details
    Browse the repository at this point in the history

Commits on Apr 5, 2024

  1. fix bf16 amp

    @Mddct
    Mddct committed Apr 5, 2024
    Configuration menu
    Copy the full SHA
    2f1f4df View commit details
    Browse the repository at this point in the history

Commits on Apr 6, 2024

  1. refactor scaler and autocast, fix fp32 fp16 bf16 for fsdp

    @Mddct
    Mddct committed Apr 6, 2024
    Configuration menu
    Copy the full SHA
    2d70039 View commit details
    Browse the repository at this point in the history

Commits on Apr 7, 2024

  1. fix fp32 nullcontext to nullcontext()

    @Mddct
    Mddct committed Apr 7, 2024
    Configuration menu
    Copy the full SHA
    767b0ed View commit details
    Browse the repository at this point in the history

  2. modify after review

    @Mddct
    Mddct committed Apr 7, 2024
    Configuration menu
    Copy the full SHA
    9d615e2 View commit details
    Browse the repository at this point in the history

  3. fix lint

    @Mddct
    Mddct committed Apr 7, 2024
    Configuration menu
    Copy the full SHA
    8716038 View commit details
    Browse the repository at this point in the history

  4. fix lint

    @Mddct
    Mddct committed Apr 7, 2024
    Configuration menu
    Copy the full SHA
    a92ee10 View commit details
    Browse the repository at this point in the history