diffusers/tests/others/test_training.py

# coding=utf-8
# Copyright 2026 HuggingFace Inc.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

import torch

from diffusers import DDIMScheduler, DDPMScheduler, UNet2DModel
from diffusers.training_utils import compute_confidence_aware_loss, set_seed

from ..testing_utils import slow


torch.backends.cuda.matmul.allow_tf32 = False


class TestTraining:
    def get_model_optimizer(self, resolution=32):
        set_seed(0)
        model = UNet2DModel(sample_size=resolution, in_channels=3, out_channels=3)
        optimizer = torch.optim.SGD(model.parameters(), lr=0.0001)
        return model, optimizer

    @slow
    def test_training_step_equality(self):
        device = "cpu"  # ensure full determinism without setting the CUBLAS_WORKSPACE_CONFIG env variable
        ddpm_scheduler = DDPMScheduler(
            num_train_timesteps=1000,
            beta_start=0.0001,
            beta_end=0.02,
            beta_schedule="linear",
            clip_sample=True,
        )
        ddim_scheduler = DDIMScheduler(
            num_train_timesteps=1000,
            beta_start=0.0001,
            beta_end=0.02,
            beta_schedule="linear",
            clip_sample=True,
        )

        assert ddpm_scheduler.config.num_train_timesteps == ddim_scheduler.config.num_train_timesteps

        # shared batches for DDPM and DDIM
        set_seed(0)
        clean_images = [torch.randn((4, 3, 32, 32)).clip(-1, 1).to(device) for _ in range(4)]
        noise = [torch.randn((4, 3, 32, 32)).to(device) for _ in range(4)]
        timesteps = [torch.randint(0, 1000, (4,)).long().to(device) for _ in range(4)]

        # train with a DDPM scheduler
        model, optimizer = self.get_model_optimizer(resolution=32)
        model.train().to(device)
        for i in range(4):
            optimizer.zero_grad()
            ddpm_noisy_images = ddpm_scheduler.add_noise(clean_images[i], noise[i], timesteps[i])
            ddpm_noise_pred = model(ddpm_noisy_images, timesteps[i]).sample
            loss = torch.nn.functional.mse_loss(ddpm_noise_pred, noise[i])
            loss.backward()
            optimizer.step()
        del model, optimizer

        # recreate the model and optimizer, and retry with DDIM
        model, optimizer = self.get_model_optimizer(resolution=32)
        model.train().to(device)
        for i in range(4):
            optimizer.zero_grad()
            ddim_noisy_images = ddim_scheduler.add_noise(clean_images[i], noise[i], timesteps[i])
            ddim_noise_pred = model(ddim_noisy_images, timesteps[i]).sample
            loss = torch.nn.functional.mse_loss(ddim_noise_pred, noise[i])
            loss.backward()
            optimizer.step()
        del model, optimizer

        assert torch.allclose(ddpm_noisy_images, ddim_noisy_images, atol=1e-5)
        assert torch.allclose(ddpm_noise_pred, ddim_noise_pred, atol=1e-5)

    def test_confidence_aware_loss(self):
        logits = torch.tensor([[[5.0, 0.0], [0.0, 5.0]]])
        labels = torch.tensor([[0, 0]])
        weights = torch.tensor([[1.0, 2.0]])

        loss, loss_sft, loss_conf = compute_confidence_aware_loss(
            logits, labels, lambda_conf=0.0, per_token_weights=weights
        )
        assert torch.allclose(loss, loss_sft)
        assert torch.allclose(loss_conf, torch.zeros_like(loss_conf))

        lambda_conf = 0.25
        loss, loss_sft, loss_conf = compute_confidence_aware_loss(
            logits, labels, lambda_conf=lambda_conf, per_token_weights=weights
        )

        # Manual expected values for the small 2-class case.
        per_token_nll = torch.nn.functional.cross_entropy(logits.view(-1, 2), labels.view(-1), reduction="none").view(
            1, 2
        )
        expected_sft = (per_token_nll * weights).sum() / weights.sum()

        pred = logits.argmax(dim=-1)
        correct = pred.eq(labels)
        log_probs = torch.log_softmax(logits.float(), dim=-1)
        probs = log_probs.exp()
        entropy = -(probs * log_probs).sum(dim=-1).to(dtype=logits.dtype)
        expected_conf = (entropy * weights * correct.to(entropy.dtype)).sum() / (
            weights * correct.to(weights.dtype)
        ).sum().clamp_min(1)

        expected = expected_sft + lambda_conf * expected_conf
        assert torch.allclose(loss_sft, expected_sft)
        assert torch.allclose(loss_conf, expected_conf)
        assert torch.allclose(loss, expected)

        # Temperature affects only the confidence term.
        loss_t, loss_sft_t, loss_conf_t = compute_confidence_aware_loss(
            logits, labels, lambda_conf=lambda_conf, temperature=0.5, per_token_weights=weights
        )
        assert torch.allclose(loss_sft_t, expected_sft)
        assert not torch.allclose(loss_conf_t, expected_conf)
        assert torch.allclose(loss_t, loss_sft_t + lambda_conf * loss_conf_t)