[training scripts] `--logit_mean` biases timesteps in the opposite direction to the SD3 paper (positive m oversamples low noise)
維護者通常 1 天內回覆
@mlzerotoheroyoutube 已經在處理了。
開始於 2026年11月3日。
評估
- 難度
- 4/5
- 預估耗時
- 3-5 天
- 新手友好度
- 35/100
- Issue 類型
- 缺陷
- 描述清晰度
- 描述清楚
- 活躍度
- 活躍
研究方向
Start with the sampling helper compute_density_for_timestep_sampling in src/diffusers/training_utils.py, then the indexing line in examples/dreambooth/train_dreambooth_lora_sd3.py that maps u onto noise_scheduler_copy.timesteps. Run the issue's reproduction script first to confirm the reversed E[sigma]. Done means the sign of logit_mean matches the SD3 paper across the 28 affected scripts, with the reproduction assertion passing. The maintainers still need to decide whether to flip the sampler or the indexing, since that changes behaviour for existing users.
由索引模型根據 Issue 內容生成。
描述
Describe the bug
The SD3/FLUX-style training scripts sample timesteps with compute_density_for_timestep_sampling(weighting_scheme="logit_normal", logit_mean=m, logit_std=s), which returns u = sigmoid(N(m, s)). They then index the scheduler's timestep table with indices = (u * num_train_timesteps).long(). FlowMatchEulerDiscreteScheduler's table is ordered noise-first (timesteps[0] = 1000, sigmas[0] = 1), so the realised noise level is sigma ≈ shift(1 - u), i.e. sigma follows a (shifted) logit-normal(-m, s) distribution.
The SD3 paper (Esser et al. 2024, Sec. 3.1, Eq. 19), which the helper's docstring and README_sd3.md refer to, samples the noise level t of z_t = (1 - t) x_0 + t eps (the scripts' own comment: zt = (1 - texp) * x + texp * z1) from logit-normal(m, s), and says m biases training "towards either data p0 (negative m) or noise p1 (positive m)". In the scripts the sign of --logit_mean is therefore reversed: --logit_mean 1.0 --logit_std 0.6 (the paper's lognorm(1.00, 0.60)) concentrates training on low noise levels (FLUX: E[sigma] 0.284 instead of 0.717; P(sigma > 0.5) 0.048 instead of 0.952).
The defaults are not affected: logit_mean=0 and the mode / uniform schemes are symmetric. Only users who set --logit_mean get the opposite of what the paper describes.
Affected: the 28 scripts that use this indexing line (list below). Not affected:
train_dreambooth_lora_lumina2.py, which also reverses the interpolation ((1 - sigma) * noise + sigma * x0);train_dreambooth_lora_ideogram4.py, which uses the sampled value directly as sigma, i.e. the paper's direction.
Reproduction
This uses the real compute_density_for_timestep_sampling and FlowMatchEulerDiscreteScheduler with the FLUX.1 and SD3 scheduler configs, and the scripts' indexing:
# The SD3/FLUX training scripts sample timesteps like this (e.g. examples/dreambooth/train_dreambooth_sd3.py):
# u = compute_density_for_timestep_sampling(weighting_scheme="logit_normal", logit_mean=m, logit_std=s, ...)
# indices = (u * noise_scheduler_copy.config.num_train_timesteps).long()
# timesteps = noise_scheduler_copy.timesteps[indices] -> sigmas via get_sigmas(timesteps)
# noisy_model_input = (1.0 - sigmas) * model_input + sigmas * noise
# SD3 paper (Esser et al. 2024, Sec. 3.1, Eq. 19): t ~ logit-normal(m, s) is the noise level of
# z_t = (1 - t) x_0 + t eps, and m > 0 biases training towards noise. Here sigma plays the role of t.
import torch
import diffusers
from diffusers import FlowMatchEulerDiscreteScheduler
from diffusers.training_utils import compute_density_for_timestep_sampling
print(f"diffusers {diffusers.__version__}, torch {torch.__version__}")
m, s, n = 1.0, 0.6, 200_000
configs = { # scheduler configs of the FLUX.1 and SD3 checkpoints (scheduler/scheduler_config.json)
"FLUX.1": dict(shift=3.0, use_dynamic_shifting=True, base_shift=0.5, max_shift=1.15,
base_image_seq_len=256, max_image_seq_len=4096),
"SD3": dict(shift=3.0),
}
for name, cfg in configs.items():
sched = FlowMatchEulerDiscreteScheduler(num_train_timesteps=1000, **cfg)
torch.manual_seed(0)
u = compute_density_for_timestep_sampling(weighting_scheme="logit_normal", batch_size=n,
logit_mean=m, logit_std=s, mode_scale=1.29)
indices = (u * sched.config.num_train_timesteps).long() # as in the scripts
timesteps = sched.timesteps[indices]
# get_sigmas() looks up each timestep's position in the same table, i.e. `indices` (checked on a subset)
assert all((sched.timesteps == t).nonzero().item() == i for t, i in zip(timesteps[:500], indices[:500]))
sigma = sched.sigmas[indices]
# Paper direction: the noise level itself is logit-normal(m, s) (then the same table/shift)
sigma_paper = sched.sigmas[((1 - u) * 1000).long().clamp(max=999)]
print(f"{name:6s} logit_mean={m:+.1f}, logit_std={s}: E[sigma] scripts {sigma.mean():.3f}, paper {sigma_paper.mean():.3f}"
f" | P(sigma > 0.5) scripts {(sigma > 0.5).float().mean():.3f}, paper {(sigma_paper > 0.5).float().mean():.3f}")
assert sigma.mean() < 0.55 < sigma_paper.mean() # m > 0 should bias towards noise, the scripts bias towards data
print("BUG REPRODUCED: the sign of --logit_mean is reversed relative to the SD3 paper")
Logs
diffusers 0.42.1, torch 2.15.0+cpu
FLUX.1 logit_mean=+1.0, logit_std=0.6: E[sigma] scripts 0.284, paper 0.717 | P(sigma > 0.5) scripts 0.048, paper 0.952
SD3 logit_mean=+1.0, logit_std=0.6: E[sigma] scripts 0.524, paper 0.877 | P(sigma > 0.5) scripts 0.567, paper 1.000
BUG REPRODUCED: the sign of --logit_mean is reversed relative to the SD3 paper
Expected: with logit_mean=+1.0 the noise level should be biased towards noise, as in the paper (E[sigma] 0.717 for FLUX.1, 0.877 for SD3).
Where: training_utils.py#L366-L390 (v0.42.1, 2a14339) returns u, which correctly plays the role of the paper's t. The table is noise-first (scheduling_flow_match_euler_discrete.py#L127-L136). The scripts index it with u, e.g. train_dreambooth_lora_sd3.py#L1784-L1797:
u = compute_density_for_timestep_sampling(
weighting_scheme=args.weighting_scheme,
batch_size=bsz,
logit_mean=args.logit_mean,
logit_std=args.logit_std,
mode_scale=args.mode_scale,
)
indices = (u * noise_scheduler_copy.config.num_train_timesteps).long()
timesteps = noise_scheduler_copy.timesteps[indices].to(device=model_input.device)
# Add noise according to flow matching.
# zt = (1 - texp) * x + texp * z1
sigmas = get_sigmas(timesteps, n_dim=model_input.ndim, dtype=model_input.dtype)
noisy_model_input = (1.0 - sigmas) * model_input + sigmas * noise
The indexing line is identical, at the same line number, in all 28 scripts at v0.42.1 and on main (86ef420).
The 28 affected scripts (indexing line)
examples/advanced_diffusion_training/train_dreambooth_lora_flux_advanced.py#L2206examples/cogview4-control/train_control_cogview4.py#L1077examples/controlnet/train_controlnet_flux.py#L1288examples/controlnet/train_controlnet_sd3.py#L1320examples/dreambooth/train_dreambooth_flux.py#L1634examples/dreambooth/train_dreambooth_lora_bria_fibo2_img2img.py#L1731examples/dreambooth/train_dreambooth_lora_bria_fibo2.py#L1898examples/dreambooth/train_dreambooth_lora_flux2_img2img.py#L1779examples/dreambooth/train_dreambooth_lora_flux2_klein_img2img.py#L1725examples/dreambooth/train_dreambooth_lora_flux2_klein.py#L1845examples/dreambooth/train_dreambooth_lora_flux2.py#L1876examples/dreambooth/train_dreambooth_lora_flux3_klein_img2img.py#L1870examples/dreambooth/train_dreambooth_lora_flux3_klein.py#L1989examples/dreambooth/train_dreambooth_lora_flux_kontext.py#L2069examples/dreambooth/train_dreambooth_lora_flux.py#L1773examples/dreambooth/train_dreambooth_lora_hidream.py#L1597examples/dreambooth/train_dreambooth_lora_krea2.py#L1857examples/dreambooth/train_dreambooth_lora_qwenimage21_img2img.py#L1886examples/dreambooth/train_dreambooth_lora_qwenimage21.py#L1808examples/dreambooth/train_dreambooth_lora_qwen_image.py#L1753examples/dreambooth/train_dreambooth_lora_sana.py#L1410examples/dreambooth/train_dreambooth_lora_sd3.py#L1791examples/dreambooth/train_dreambooth_lora_z_image.py#L1741examples/dreambooth/train_dreambooth_sd3.py#L1587examples/flux-control/train_control_flux.py#L1067examples/flux-control/train_control_lora_flux.py#L1213examples/research_projects/flux_lora_quantization/train_dreambooth_lora_flux_miniature.py#L1063examples/research_projects/sd3_lora_colab/train_dreambooth_lora_sd3_miniature.py#L985
Proposed fix. The same one-line change in each of the 28 scripts. The shared helper cannot be flipped, because the ideogram4 script uses its output directly as sigma:
- indices = (u * noise_scheduler_copy.config.num_train_timesteps).long()
+ indices = ((1 - u) * noise_scheduler_copy.config.num_train_timesteps).long().clamp(
+ max=noise_scheduler_copy.config.num_train_timesteps - 1
+ )
The mode and uniform schemes, and logit_normal with logit_mean=0, are symmetric under u -> 1 - u, so their distributions do not change.
Verifying the fix. fix_check.py applies the fixed indexing with the same real helper and schedulers. It compares the realised sigmas with the paper's distribution sigma = shift(sigmoid(x)), x ~ N(m, s) (quantiles at 5/25/50/75/95%), and checks that the default settings keep the same distribution:
released = sched.sigmas[(u * N).long()]
fixed = sched.sigmas[((1 - u) * N).long().clamp(max=N - 1)]
t = torch.sigmoid(torch.randn(n) * s + m) # paper: the noise level t ~ logit-normal(m, s)
paper = shift * t / (1 + (shift - 1) * t) # shift = 1 for FLUX.1 (dynamic shifting), 3 for SD3
assert (fixed.quantile(qs) - paper.quantile(qs)).abs().max() < 5e-3
diffusers 0.42.1, torch 2.15.0+cpu
FLUX.1 logit_normal m=+1.0 s=0.6: E[sigma] released 0.284, fixed 0.717, paper 0.716 | max quantile gap fixed vs paper 0.0012
FLUX.1 logit_normal m=-0.5 s=1.0: E[sigma] released 0.603, fixed 0.398, paper 0.398 | max quantile gap fixed vs paper 0.0007
FLUX.1 logit_normal m=+0.0 s=1.0: E[sigma] released 0.501, fixed 0.500, paper 0.500 | max quantile gap fixed vs paper 0.0007 | max quantile gap fixed vs released 0.0020
FLUX.1 mode m=+0.0 s=1.0: E[sigma] released 0.501, fixed 0.500 | max quantile gap fixed vs released 0.0020
FLUX.1 none m=+0.0 s=1.0: E[sigma] released 0.500, fixed 0.501 | max quantile gap fixed vs released 0.0030
SD3 logit_normal m=+1.0 s=0.6: E[sigma] released 0.524, fixed 0.877, paper 0.877 | max quantile gap fixed vs paper 0.0009
SD3 logit_normal m=-0.5 s=1.0: E[sigma] released 0.794, fixed 0.621, paper 0.621 | max quantile gap fixed vs paper 0.0010
SD3 logit_normal m=+0.0 s=1.0: E[sigma] released 0.715, fixed 0.714, paper 0.714 | max quantile gap fixed vs paper 0.0011 | max quantile gap fixed vs released 0.0021
SD3 mode m=+0.0 s=1.0: E[sigma] released 0.710, fixed 0.710 | max quantile gap fixed vs released 0.0020
SD3 none m=+0.0 s=1.0: E[sigma] released 0.676, fixed 0.677 | max quantile gap fixed vs released 0.0027
FIX OK: the realised noise levels follow the SD3 paper's logit-normal(m, s); defaults unchanged
System Info
- 🤗 Diffusers version: 0.42.1
- Platform: Linux x86_64
- Python version: 3.13.13
- PyTorch version: 2.15.0+cpu (CPU only)
- Transformers version: 5.22.1
- Accelerate version: 1.15.0
Who can help?
No response
I would be happy to open a PR with this fix for the 28 scripts.
- 主要語言
- Python
- 星號
- 34.9k
- 分支
- 7.5k
- 平均合併
- 3 天 16 小時
- 30 天內合併 PR
- 65
環境準備
- 沒有 Dockerfile 或 Docker Compose 檔案
- 有 Pull Request 範本
- 閱讀貢獻指南
從這裡開始
- 先讀完整個 Issue,再讀專案的貢獻指南。
- 在 Issue 下留言說明你要接手 —— 這能避免兩個人做同樣的事。
- Fork 儲存庫,在一個分支上完成修改。
- 送出 Pull Request,並在描述裡引用這個 Issue 編號。
huggingface/diffusers 的其他 Issue
-
難度 2/5 1-3 小時 新手友好度 84/100
huggingface/diffusers#15363 ·
維護者通常 1 天內回覆
-
`ZImageImg2ImgPipeline` / `ZImageInpaintPipeline`: explicit height/width different from the input image raises RuntimeError可能已有人在做 @chessing234 於 1 天前認領。 未關閉bug
難度 2/5 1-3 小時 新手友好度 70/100
huggingface/diffusers#15337 · 2 則留言 ·
維護者通常 1 天內回覆
-
Multi-vector textual inversion uses only the first vector when the token has uppercase letters (transformers v5)可能已有人在做 @sunhouy 於 2 天前認領。 未關閉
難度 2/5 1-3 小時 新手友好度 75/100
huggingface/diffusers#15313 ·
維護者通常 1 天內回覆
-
TCDScheduler advances the generator on the final step of a strength-truncated custom schedule可能已有人在做 @SIDDARTHAREDDY8 於 5 天前認領。 未關閉bug
難度 2/5 1-3 小時 新手友好度 74/100
huggingface/diffusers#15288 · 2 則留言 ·
維護者通常 1 天內回覆
-
unload_textual_inversion(token) leaves the extra tokens of a multi-vector embedding behind, so it can't be loaded again可能已有人在做 @sohailshk 於 11 天前認領。 未關閉
難度 2/5 1-3 小時 新手友好度 72/100
huggingface/diffusers#15211 · 3 則留言 ·
維護者通常 1 天內回覆
查看 huggingface/diffusers 的全部 Issue
相似的 Issue
-
難度 2/5 1-3 小時 新手友好度 73/100
維護者通常 1 天內回覆
-
Unused loot bag group names left in dreamcraft en_US.lang可能已有人在做 關聯的 PR 仍在進行中或已合併。 未關閉Status: Triage Suggestion
難度 1/5 1 小時以內 新手友好度 90/100
GTNewHorizons/GT-New-Horizons-Modpack#27789 ·
維護者通常 1 天內回覆
-
bug frontend sev-5
難度 2/5 1-3 小時 新手友好度 83/100
-
core good first issue
難度 2/5 1-3 小時 新手友好度 65/100
-
b4b code health next
難度 2/5 1-3 小時 新手友好度 68/100
維護者通常 1 天內回覆