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

[Feature] refractor dist came; fix percision error; add low level zero test with bert model zoo; #5676

Merged
merged 7 commits into from
Apr 30, 2024
Merged

[Feature] refractor dist came; fix percision error; add low level zero test with bert model zoo; #5676

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
7 commits
Select commit Hold shift + click to select a range
c75ce7e
[feature] daily update;
duanjunwen Apr 4, 2024
901e7ea
Merge branch 'feature/dist-optim' of https://github.com/hpcaitech/Col…
duanjunwen Apr 30, 2024
c5794de
[fix] fix dist came;
duanjunwen Apr 30, 2024
30a3519
[feature] refractor dist came; fix percision error; add low level zer…
duanjunwen Apr 30, 2024
74d4fac
[fix] open rms; fix low level zero test; fix dist came test function …
duanjunwen Apr 30, 2024
e82e2b2
[fix] remove redundant test;
duanjunwen Apr 30, 2024
c958af1
[pre-commit.ci] auto fixes from pre-commit.com hooks
pre-commit-ci[bot] Apr 30, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
6 changes: 3 additions & 3 deletions colossalai/nn/optimizer/came.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -75,7 +75,7 @@ def step(self, closure=None):
for p in group["params"]:
if p.grad is None:
continue
grad = p.grad.data
grad = p.grad
if grad.is_sparse:
raise RuntimeError("CAME does not support sparse gradients.")

Expand Down Expand Up @@ -104,6 +104,7 @@ def step(self, closure=None):
state["step"] += 1

update = (grad**2) + group["eps"][0]

if factored:
exp_avg_sq_row = state["exp_avg_sq_row"]
exp_avg_sq_col = state["exp_avg_sq_col"]
Expand Down Expand Up @@ -132,15 +133,14 @@ def step(self, closure=None):
if factored:
exp_avg_res_row = state["exp_avg_res_row"]
exp_avg_res_col = state["exp_avg_res_col"]

exp_avg_res_row.mul_(group["betas"][2]).add_(res.mean(dim=-1), alpha=1.0 - group["betas"][2])
exp_avg_res_col.mul_(group["betas"][2]).add_(res.mean(dim=-2), alpha=1.0 - group["betas"][2])

# Approximation of exponential moving average of instability
res_approx = self._approx_sq_grad(exp_avg_res_row, exp_avg_res_col)
update = res_approx.mul_(exp_avg)
else:
update = exp_avg
update = exp_avg.clone()

if group["weight_decay"] != 0:
p.data.add_(p.data, alpha=-group["weight_decay"] * group["lr"])
Expand Down
Loading
Loading