Missing trainer.py file in src/diffusers/trainer.py causing import error

[bobrathbone]

Describe the bug

There is no file called trainer.py at /src/diffusers/ in the official diffusers repository.
I(myenv) C:\Temp\jan\scripts\diffusers>python
Python 3.13.14 (tags/v3.13.14:fd17997, Jun 10 2026, 13:03:48) [MSC v.1944 64 bit (AMD64)] on win32
Type "help", "copyright", "credits" or "license" for more information.

from diffusers.trainer import Trainer as DiffusersTrainer
Traceback (most recent call last):
File "", line 1, in
from diffusers.trainer import Trainer as DiffusersTrainer
ModuleNotFoundError: No module named 'diffusers.trainer'

These are my imports

import torch
from diffusers import (
    StableDiffusionPipeline,
    UNet2DConditionModel,
)

from diffusers.loaders import LoraLoaderMixin
from peft import LoraConfig, get_peft_model
from datasets import Dataset
from diffusers.trainer import Trainer as DiffusersTrainer  

Below is my training subroutine

def train():
    # Load dataset
    print("Loading training data...")
    raw_data = load_dataset_from_jsonl(LABELS_FILE)

    if not raw_data:
        raise ValueError(f"No valid entries found in {LABELS_FILE}!")

    print(f"Loaded {len(raw_data)} image-prompt pairs.")

    # Create dataset object (for trainer compatibility)
    prompts = [item["prompt"] for item in raw_data]
    img_paths = [Path(item["image_path"]) for item in raw_data]

    train_dataset = ImagePromptDataset(prompts=prompts, image_paths=img_paths)

    # Setup model
    pipe = setup_model()

    # Replace UNet with LoRA-enabled version (if not already done)
    if hasattr(pipe.unet, "model"):
        print("✅ Successfully applied LoRA to SD pipeline.")

    # Training arguments — use LoRA-specific config
    training_args = {
        "output_dir": "./lora_finetuned_sd",
        "logging_steps": 10,
        "save_steps": 50,
        "learning_rate": 2e-4,
        "num_train_epochs": EPOCHS,
        "gradient_accumulation_steps": BATCH_SIZE,  # Adjust per GPU batch size
        "per_device_train_batch_size": BATCH_SIZE,
        "fp16": True,
        "warmup_ratio": 0.05,
        "weight_decay": 0.01,
        "report_to": None,
    }

    trainer = DiffusersTrainer(
        model=pipe.unet,
        args=training_args,
        train_dataset=train_dataset,
        tokenizer=None
    )

    print("Starting training...")
    trainer.train()

    # Save final LoRA weights
    pipe.save_lora_weights("./lora_finetuned_sd", safe=True)
    print("✅ Final LoRA model saved to ./lora_finetuned_sd")

Reproduction

python

from diffusers.trainer import Trainer as DiffusersTrainer

Logs

>python
Python 3.13.14 (tags/v3.13.14:fd17997, Jun 10 2026, 13:03:48) [MSC v.1944 64 bit (AMD64)] on win32
Type "help", "copyright", "credits" or "license" for more information.
>>> **from diffusers.trainer import Trainer as DiffusersTrainer**
Traceback (most recent call last):
  File "<python-input-0>", line 1, in <module>
    from diffusers.trainer import Trainer as DiffusersTrainer
ModuleNotFoundError: No module named 'diffusers.trainer'
>>>

System Info

Python 3.13.14

import diffusers
print (diffusers.version)
0.40.0.dev0

Who can help?

No response

[indrajeetapache]

This doesn’t look like a bug. Diffusers has never shipped diffusers.trainer.Trainer — that API is from Transformers (from transformers import Trainer).

Diffusers training is done via the scripts in examples/ (Accelerate + an explicit training loop), not a Trainer class. For LoRA fine-tuning of Stable Diffusion, see:

https://huggingface.co/docs/diffusers/en/training/lora
examples/text_to_image/train_text_to_image_lora.py
Suggest closing this as not-a-bug