Skip to content

Profile and speedup pipelines #9475

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

Closed
wants to merge 6 commits into from
Closed

Profile and speedup pipelines #9475

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
5687dc6
profile cogvideox
a-r-r-o-w Sep 19, 2024
fb0ebbb
update
a-r-r-o-w Sep 19, 2024
3ae6094
dump
a-r-r-o-w Sep 24, 2024
b7dd6ba
Merge branch 'main' into cogvideox-profiling
a-r-r-o-w Sep 24, 2024
8e78c9d
update
a-r-r-o-w Sep 24, 2024
3bf3961
update
a-r-r-o-w Sep 24, 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
21 changes: 16 additions & 5 deletions src/diffusers/schedulers/scheduling_ddim.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -231,7 +231,9 @@ def __init__(

# setable values
self.num_inference_steps = None
self.timesteps = torch.from_numpy(np.arange(0, num_train_timesteps)[::-1].copy().astype(np.int64))

# TODO: discuss with YiYi why we have a .copy() here and if it's really needed. I've removed it for now
self.timesteps = torch.from_numpy(np.arange(0, num_train_timesteps)[::-1].astype(np.int64))

def scale_model_input(self, sample: torch.Tensor, timestep: Optional[int] = None) -> torch.Tensor:
"""
Expand All @@ -251,8 +253,12 @@ def scale_model_input(self, sample: torch.Tensor, timestep: Optional[int] = None
return sample

def _get_variance(self, timestep, prev_timestep):
alpha_prod_t = self.alphas_cumprod[timestep]
alpha_prod_t_prev = self.alphas_cumprod[prev_timestep] if prev_timestep >= 0 else self.final_alpha_cumprod
alpha_prod_t = torch.gather(self.alphas_cumprod, 0, timestep)

safe_prev_timestep = torch.clamp(prev_timestep, min=0)
safe_alpha_prod_t_prev = torch.gather(self.alphas_cumprod, 0, safe_prev_timestep)
alpha_prod_t_prev = torch.where(prev_timestep >= 0, safe_alpha_prod_t_prev, self.final_alpha_cumprod)

beta_prod_t = 1 - alpha_prod_t
beta_prod_t_prev = 1 - alpha_prod_t_prev

Expand Down Expand Up @@ -338,6 +344,8 @@ def set_timesteps(self, num_inference_steps: int, device: Union[str, torch.devic
)

self.timesteps = torch.from_numpy(timesteps).to(device)
self.alphas_cumprod = self.alphas_cumprod.to(device)
self.final_alpha_cumprod = self.final_alpha_cumprod.to(device)

def step(
self,
Expand Down Expand Up @@ -402,8 +410,11 @@ def step(
prev_timestep = timestep - self.config.num_train_timesteps // self.num_inference_steps

# 2. compute alphas, betas
alpha_prod_t = self.alphas_cumprod[timestep]
alpha_prod_t_prev = self.alphas_cumprod[prev_timestep] if prev_timestep >= 0 else self.final_alpha_cumprod
alpha_prod_t = torch.gather(self.alphas_cumprod, 0, timestep)

safe_prev_timestep = torch.clamp(prev_timestep, min=0)
safe_alpha_prod_t_prev = torch.gather(self.alphas_cumprod, 0, safe_prev_timestep)
alpha_prod_t_prev = torch.where(prev_timestep >= 0, safe_alpha_prod_t_prev, self.final_alpha_cumprod)

beta_prod_t = 1 - alpha_prod_t

Expand Down
21 changes: 16 additions & 5 deletions src/diffusers/schedulers/scheduling_ddim_cogvideox.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -228,11 +228,17 @@ def __init__(

# setable values
self.num_inference_steps = None
self.timesteps = torch.from_numpy(np.arange(0, num_train_timesteps)[::-1].copy().astype(np.int64))

# TODO: discuss with YiYi why we have a .copy() here and if it's really needed. I've removed it for now
self.timesteps = torch.from_numpy(np.arange(0, num_train_timesteps)[::-1].astype(np.int64))

def _get_variance(self, timestep, prev_timestep):
alpha_prod_t = self.alphas_cumprod[timestep]
alpha_prod_t_prev = self.alphas_cumprod[prev_timestep] if prev_timestep >= 0 else self.final_alpha_cumprod
alpha_prod_t = torch.gather(self.alphas_cumprod, 0, timestep)

safe_prev_timestep = torch.clamp(prev_timestep, min=0)
safe_alpha_prod_t_prev = torch.gather(self.alphas_cumprod, 0, safe_prev_timestep)
alpha_prod_t_prev = torch.where(prev_timestep >= 0, safe_alpha_prod_t_prev, self.final_alpha_cumprod)

beta_prod_t = 1 - alpha_prod_t
beta_prod_t_prev = 1 - alpha_prod_t_prev

Expand Down Expand Up @@ -301,6 +307,8 @@ def set_timesteps(self, num_inference_steps: int, device: Union[str, torch.devic
)

self.timesteps = torch.from_numpy(timesteps).to(device)
self.alphas_cumprod = self.alphas_cumprod.to(device)
self.final_alpha_cumprod = self.final_alpha_cumprod.to(device)

def step(
self,
Expand Down Expand Up @@ -365,8 +373,11 @@ def step(
prev_timestep = timestep - self.config.num_train_timesteps // self.num_inference_steps

# 2. compute alphas, betas
alpha_prod_t = self.alphas_cumprod[timestep]
alpha_prod_t_prev = self.alphas_cumprod[prev_timestep] if prev_timestep >= 0 else self.final_alpha_cumprod
alpha_prod_t = torch.gather(self.alphas_cumprod, 0, timestep)

safe_prev_timestep = torch.clamp(prev_timestep, min=0)
safe_alpha_prod_t_prev = torch.gather(self.alphas_cumprod, 0, safe_prev_timestep)
alpha_prod_t_prev = torch.where(prev_timestep >= 0, safe_alpha_prod_t_prev, self.final_alpha_cumprod)

beta_prod_t = 1 - alpha_prod_t

Expand Down
Loading