DeepSpeed
Z3: optimizations for grad norm calculation and gradient clipping
#5504
Merged

Z3: optimizations for grad norm calculation and gradient clipping #5504

nelyahu
nelyahu nelyahu requested a review from tjruwase tjruwase 1 year ago
nelyahu nelyahu requested a review from mrwyattii mrwyattii 1 year ago
tjruwase
tjruwase commented on 2024-05-07
z3 scaled_global_grad_norm: repalce get_global_norm with torch.norm
43792bb7
nelyahu nelyahu force pushed from 299e3b62 to 43792bb7 1 year ago
tjruwase Merge branch 'master' into zero3_scaled_global
cac04d9c
tjruwase
tjruwase approved these changes on 2024-05-20
loadams Merge branch 'master' into zero3_scaled_global
cbb6b6a5
jomayeri
tjruwase Merge branch 'master' into zero3_scaled_global
37b4cb77
fix grad norm calc in cpu offload and use torch.clip for grad clipping
5dd50c32
nelyahu nelyahu changed the title z3 scaled_global_grad_norm: repalce get_global_norm with torch.norm Z3: optimizations for grad norm calculation and gradient clipping 1 year ago
lekurile Merge branch 'master' into zero3_scaled_global
f918054a
nelyahu Merge branch 'master' into zero3_scaled_global
ba9fd426
nelyahu Merge branch 'master' into zero3_scaled_global
238ab343
loadams Merge branch 'master' into zero3_scaled_global
6b6a834f
loadams
nelyahu
tjruwase Merge branch 'master' into zero3_scaled_global
b14f920f
tjruwase
tjruwase Merge branch 'master' into zero3_scaled_global
45e62d29
loadams
loadams Merge branch 'master' into zero3_scaled_global
ffdb7f7d
nelyahu
nelyahu nelyahu requested a review from loadams loadams 1 year ago
adding gradient clipping to TestZeroPartialOffloadConfigSweep
d99524ba
nelyahu nelyahu force pushed from d2919c52 to d99524ba 1 year ago
nelyahu
tjruwase Merge branch 'master' into zero3_scaled_global
e5d5d7c2
nelyahu
loadams Merge branch 'master' into zero3_scaled_global
67d2e35a
loadams Merge branch 'master' into zero3_scaled_global
214103df
loadams loadams enabled auto-merge 1 year ago
loadams Merge branch 'master' into zero3_scaled_global
6134da44
loadams Merge branch 'master' into zero3_scaled_global
4b16d7ca
tjruwase Merge branch 'master' into zero3_scaled_global
38bb8608
loadams Merge branch 'master' into zero3_scaled_global
ec1aa8c0
loadams Merge branch 'master' into zero3_scaled_global
95cd3a1f
Merge branch 'master' into zero3_scaled_global
26676f26
loadams Merge branch 'master' into zero3_scaled_global
e4559624
loadams Merge branch 'master' into zero3_scaled_global
3407e93c
loadams Merge branch 'master' into zero3_scaled_global
cb51597f
loadams loadams merged 6eed634e into master 1 year ago

Login to write a write a comment.

Login via GitHub

Assignees
No one assigned
Labels
Milestone