[zero] revert PR #3166, it disabled grad clip for bf16 #3790
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
* fix conv_flops_compute when padding is a str when stride=1 * fix error * change type of paddings to tuple * fix padding calculation * apply formatting check --------- Co-authored-by: Cheng Li <[email protected]> Co-authored-by: Olatunji Ruwase <[email protected]>
* Update profiler.py * pre-commit run --all-files * Delete .DS_Store * Delete .DS_Store * Delete .DS_Store --------- Co-authored-by: Jeff Rasley <[email protected]> Co-authored-by: Cheng Li <[email protected]>
* zeropp chinese blog * try better quality images * make title larger * even larger... * various fix * center captions * more fixes * fix format
Co-authored-by: Stephen Youn <[email protected]> Co-authored-by: Arash Bakhtiari <[email protected]> Co-authored-by: Cheng Li <[email protected]> Co-authored-by: Ethan Doe <[email protected]> Co-authored-by: yidoe <[email protected]> Co-authored-by: Jeff Rasley <[email protected]>
Co-authored-by: HeyangQin <[email protected]> Co-authored-by: GuanhuaWang <[email protected]> Co-authored-by: cmikeh2 <[email protected]> Co-authored-by: Ammar Ahmad Awan <[email protected]> Co-authored-by: Jeff Rasley <[email protected]> Co-authored-by: Michael Wyatt <[email protected]> Co-authored-by: Olatunji Ruwase <[email protected]> Co-authored-by: Reza Yazdani <[email protected]>
* zeropp chinese blog * try better quality images * make title larger * even larger... * various fix * center captions * more fixes * fix format * add ZeRO++ Japanese blog * add links --------- Co-authored-by: HeyangQin <[email protected]> Co-authored-by: Conglong Li <[email protected]>
* fix autotuner when backward is not called * fix format --------- Co-authored-by: Olatunji Ruwase <[email protected]>
Co-authored-by: Ammar Ahmad Awan <[email protected]> Co-authored-by: Jeff Rasley <[email protected]> Co-authored-by: Logan Adams <[email protected]>
Co-authored-by: Jeff Rasley <[email protected]>
* Bug fix * Fixed formatting error --------- Co-authored-by: Logan Adams <[email protected]>
Co-authored-by: Stephen Youn <[email protected]> Co-authored-by: Jeff Rasley <[email protected]>
jeffra
requested review from
awan-10,
RezaYazdaniAminabadi,
cmikeh2 and
arashb
as code owners
|
thanks, good catch, yes, this is the grad clip to avoid exploding, no matter which format is used. It is independent with loss scale of fp16. |
tjruwase
approved these changes
Jun 26, 2023
loadams
approved these changes
Jun 26, 2023
zhangir-azerbayev
pushed a commit
to EleutherAI/DeeperSpeed
that referenced
this pull request
Aug 4, 2023
…osoft#3790) * zero++ tutorial PR (microsoft#3783) * [Fix] _conv_flops_compute when padding is a str and stride=1 (microsoft#3169) * fix conv_flops_compute when padding is a str when stride=1 * fix error * change type of paddings to tuple * fix padding calculation * apply formatting check --------- Co-authored-by: Cheng Li <[email protected]> Co-authored-by: Olatunji Ruwase <[email protected]> * fix interpolate flops compute (microsoft#3782) * use `Flops Profiler` to test `model.generate()` (microsoft#2515) * Update profiler.py * pre-commit run --all-files * Delete .DS_Store * Delete .DS_Store * Delete .DS_Store --------- Co-authored-by: Jeff Rasley <[email protected]> Co-authored-by: Cheng Li <[email protected]> * revert PR microsoft#3166, it disabled grad clip for bf16 * ensure no loss scaling for non-fp16 dtypes * revert PR microsoft#3611 (microsoft#3786) * bump to 0.9.6 * ZeRO++ chinese blog (microsoft#3793) * zeropp chinese blog * try better quality images * make title larger * even larger... * various fix * center captions * more fixes * fix format * remove staging trigger (microsoft#3792) * DeepSpeed-Triton for Inference (microsoft#3748) Co-authored-by: Stephen Youn <[email protected]> Co-authored-by: Arash Bakhtiari <[email protected]> Co-authored-by: Cheng Li <[email protected]> Co-authored-by: Ethan Doe <[email protected]> Co-authored-by: yidoe <[email protected]> Co-authored-by: Jeff Rasley <[email protected]> * ZeRO++ (microsoft#3784) Co-authored-by: HeyangQin <[email protected]> Co-authored-by: GuanhuaWang <[email protected]> Co-authored-by: cmikeh2 <[email protected]> Co-authored-by: Ammar Ahmad Awan <[email protected]> Co-authored-by: Jeff Rasley <[email protected]> Co-authored-by: Michael Wyatt <[email protected]> Co-authored-by: Olatunji Ruwase <[email protected]> Co-authored-by: Reza Yazdani <[email protected]> * adding zero++ to navigation panel of deepspeed.ai (microsoft#3796) * Add ZeRO++ Japanese blog (microsoft#3797) * zeropp chinese blog * try better quality images * make title larger * even larger... * various fix * center captions * more fixes * fix format * add ZeRO++ Japanese blog * add links --------- Co-authored-by: HeyangQin <[email protected]> Co-authored-by: Conglong Li <[email protected]> * Bug Fixes for autotuner and flops profiler (microsoft#1880) * fix autotuner when backward is not called * fix format --------- Co-authored-by: Olatunji Ruwase <[email protected]> * Missing strided copy for gated MLP (microsoft#3788) Co-authored-by: Ammar Ahmad Awan <[email protected]> Co-authored-by: Jeff Rasley <[email protected]> Co-authored-by: Logan Adams <[email protected]> * Requires grad checking. (microsoft#3789) Co-authored-by: Jeff Rasley <[email protected]> * bump to 0.10.0 * Fix Bug in transform.cu (microsoft#3534) * Bug fix * Fixed formatting error --------- Co-authored-by: Logan Adams <[email protected]> * bug fix: triton importing error (microsoft#3799) Co-authored-by: Stephen Youn <[email protected]> Co-authored-by: Jeff Rasley <[email protected]> --------- Co-authored-by: Heyang Qin <[email protected]> Co-authored-by: Bill Luo <[email protected]> Co-authored-by: Cheng Li <[email protected]> Co-authored-by: Olatunji Ruwase <[email protected]> Co-authored-by: Guorun <[email protected]> Co-authored-by: stephen youn <[email protected]> Co-authored-by: Stephen Youn <[email protected]> Co-authored-by: Arash Bakhtiari <[email protected]> Co-authored-by: Ethan Doe <[email protected]> Co-authored-by: yidoe <[email protected]> Co-authored-by: GuanhuaWang <[email protected]> Co-authored-by: cmikeh2 <[email protected]> Co-authored-by: Ammar Ahmad Awan <[email protected]> Co-authored-by: Michael Wyatt <[email protected]> Co-authored-by: Reza Yazdani <[email protected]> Co-authored-by: Masahiro Tanaka <[email protected]> Co-authored-by: Conglong Li <[email protected]> Co-authored-by: Logan Adams <[email protected]> Co-authored-by: Joe Mayer <[email protected]> Co-authored-by: Ramya Ramineni <[email protected]>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
#3166 inadvertently disabled gradient clipping when using bf16. The original intention was to remove grad scaling when using bf16, however when using bf16 the loss scale is forced to 1.0 which will result in no grad scaling occurring.