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

[low-bit optim] Upcast everything to FP32 for internal calculations #1068

Merged
merged 2 commits into from
Oct 14, 2024
Merged

[low-bit optim] Upcast everything to FP32 for internal calculations #1068

Changes from 1 commit
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
7683f61
fix dtype
gau-nernst Oct 14, 2024
72ca834
Update regression_test.yml
msaroufim Oct 14, 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
29 changes: 15 additions & 14 deletions torchao/prototype/low_bit_optim/adam.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -109,9 +109,6 @@ def step(self, closure=None):

# this will work with any optim state tensor subclass that implements aten.lerp.Scalar and aten.copy_.default
# and param tensor subclass that implements aten.add_.Tensor, and aten.addcdiv_.default
# NOTE: right now all of our optimizer state subclasses will dequant to FP32, thus adam computation
# will be done in FP32 (not purposely). we should explicitly cast all inputs to FP32 to ensure FP32
# computation. will need to benchmark to ensure no slowdown.
def single_param_adam(
p: Tensor,
grad: Tensor,
Expand All @@ -126,32 +123,36 @@ def single_param_adam(
eps: float,
is_adamw: bool,
):
# compute in FP32 for accurate calculations
p_f32 = p.float()
grad_f32 = grad.float()

if not is_adamw:
grad = grad.add(p, alpha=weight_decay)
grad_f32 = grad_f32.add(p_f32, alpha=weight_decay)

bias_correction1 = 1 - beta1**step
bias_correction2 = 1 - beta2**step

# keep high precision copy for param update
new_exp_avg = exp_avg.lerp(grad, 1 - beta1)
new_exp_avg_sq = exp_avg_sq.lerp(grad.square(), 1 - beta2)
exp_avg_f32 = exp_avg.float().lerp(grad_f32, 1 - beta1)
exp_avg_sq_f32 = exp_avg_sq.float().lerp(grad_f32.square(), 1 - beta2)

exp_avg.copy_(new_exp_avg)
exp_avg_sq.copy_(new_exp_avg_sq)
exp_avg.copy_(exp_avg_f32)
exp_avg_sq.copy_(exp_avg_sq_f32)

if max_exp_avg_sq is not None:
new_max_exp_avg_sq = torch.maximum(max_exp_avg_sq, new_exp_avg_sq)
max_exp_avg_sq.copy_(new_max_exp_avg_sq)
denom = (new_max_exp_avg_sq.sqrt() / bias_correction2.sqrt()).add_(eps)
max_exp_avg_sq_f32 = torch.maximum(max_exp_avg_sq.float(), exp_avg_sq_f32)
max_exp_avg_sq.copy_(max_exp_avg_sq_f32)
denom = (max_exp_avg_sq_f32.sqrt() / bias_correction2.sqrt()).add_(eps)
else:
denom = (new_exp_avg_sq.sqrt() / bias_correction2.sqrt()).add_(eps)
denom = (exp_avg_sq_f32.sqrt() / bias_correction2.sqrt()).add_(eps)

step_size = lr / bias_correction1
if is_adamw:
# merge weight decay and param update in a single .add_() to make this work with quantized param
p.add_(-lr * weight_decay * p - step_size * new_exp_avg / denom)
p.add_(-lr * weight_decay * p_f32 - step_size * exp_avg_f32 / denom)
else:
p.addcdiv_(new_exp_avg, denom, value=-step_size)
p.addcdiv_(exp_avg_f32, denom, value=-step_size)


class Adam8bit(_AdamBase):
Expand Down
Loading