Skip to content

Add fused AdamW option and warn on torch attention mask memory usage #1270

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

Merged
SWivid merged 2 commits into from
Mar 4, 2026
Merged

Add fused AdamW option and warn on torch attention mask memory usage #1270

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
c3603d2
Add fused AdamW option and warn on torch attention mask memory usage
ZhikangNiu Mar 2, 2026
fe4e6a4
least code, directly fused=True
SWivid Mar 4, 2026
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
13 changes: 13 additions & 0 deletions src/f5_tts/model/modules.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -11,6 +11,7 @@
from __future__ import annotations

import math
import warnings
from typing import Optional

import torch
Expand Down Expand Up @@ -452,6 +453,12 @@ def __init__(
):
if attn_backend == "flash_attn":
assert is_package_available("flash_attn"), "Please install flash-attn first."
if attn_backend == "torch" and attn_mask_enabled:
warnings.warn(
"attn_mask_enabled=True with attn_backend='torch' can consume large GPU memory. "
"Please switch attn_backend to 'flash_attn'.",
UserWarning,
)

self.pe_attn_head = pe_attn_head
self.attn_backend = attn_backend
Expand Down Expand Up @@ -557,6 +564,12 @@ def __init__(
):
if attn_backend == "flash_attn":
assert is_package_available("flash_attn"), "Please install flash-attn first."
if attn_backend == "torch" and attn_mask_enabled:
warnings.warn(
"attn_mask_enabled=True with attn_backend='torch' can consume large GPU memory. "
"Please switch attn_backend to 'flash_attn'.",
UserWarning,
)

self.attn_backend = attn_backend
self.attn_mask_enabled = attn_mask_enabled
Expand Down
3 changes: 2 additions & 1 deletion src/f5_tts/model/trainer.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -85,6 +85,7 @@ def __init__(
"grad_accumulation_steps": grad_accumulation_steps,
"max_grad_norm": max_grad_norm,
"noise_scheduler": noise_scheduler,
"bnb_optimizer": bnb_optimizer,
}
model_cfg_dict["gpus"] = self.accelerator.num_processes
self.accelerator.init_trackers(
Expand Down Expand Up @@ -139,7 +140,7 @@ def __init__(

self.optimizer = bnb.optim.AdamW8bit(model.parameters(), lr=learning_rate)
else:
self.optimizer = AdamW(model.parameters(), lr=learning_rate)
self.optimizer = AdamW(model.parameters(), lr=learning_rate, fused=True)
self.model, self.optimizer = self.accelerator.prepare(self.model, self.optimizer)

@property
Expand Down