Skip to content

Reference for ultralytics/utils/autobatch.py

Note

This file is available at https://github.com/ultralytics/ultralytics/blob/main/ultralytics/utils/autobatch.py. If you spot a problem please help fix it by contributing a Pull Request 🛠️. Thank you 🙏!


ultralytics.utils.autobatch.check_train_batch_size

check_train_batch_size(model, imgsz=640, amp=True, batch=-1)

Compute optimal YOLO training batch size using the autobatch() function.

Parameters:

Name Type Description Default
model Module

YOLO model to check batch size for.

required
imgsz int

Image size used for training.

640
amp bool

Use automatic mixed precision if True.

True
batch float

Fraction of GPU memory to use. If -1, use default.

-1

Returns:

Type Description
int

Optimal batch size computed using the autobatch() function.

Note

If 0.0 < batch < 1.0, it's used as the fraction of GPU memory to use. Otherwise, a default fraction of 0.6 is used.

Source code in ultralytics/utils/autobatch.py
def check_train_batch_size(model, imgsz=640, amp=True, batch=-1):
    """
    Compute optimal YOLO training batch size using the autobatch() function.

    Args:
        model (torch.nn.Module): YOLO model to check batch size for.
        imgsz (int, optional): Image size used for training.
        amp (bool, optional): Use automatic mixed precision if True.
        batch (float, optional): Fraction of GPU memory to use. If -1, use default.

    Returns:
        (int): Optimal batch size computed using the autobatch() function.

    Note:
        If 0.0 < batch < 1.0, it's used as the fraction of GPU memory to use.
        Otherwise, a default fraction of 0.6 is used.
    """
    with autocast(enabled=amp):
        return autobatch(deepcopy(model).train(), imgsz, fraction=batch if 0.0 < batch < 1.0 else 0.6)





ultralytics.utils.autobatch.autobatch

autobatch(model, imgsz=640, fraction=0.6, batch_size=DEFAULT_CFG.batch)

Automatically estimate the best YOLO batch size to use a fraction of the available CUDA memory.

Parameters:

Name Type Description Default
model module

YOLO model to compute batch size for.

required
imgsz int

The image size used as input for the YOLO model. Defaults to 640.

640
fraction float

The fraction of available CUDA memory to use. Defaults to 0.60.

0.6
batch_size int

The default batch size to use if an error is detected. Defaults to 16.

batch

Returns:

Type Description
int

The optimal batch size.

Source code in ultralytics/utils/autobatch.py
def autobatch(model, imgsz=640, fraction=0.60, batch_size=DEFAULT_CFG.batch):
    """
    Automatically estimate the best YOLO batch size to use a fraction of the available CUDA memory.

    Args:
        model (torch.nn.module): YOLO model to compute batch size for.
        imgsz (int, optional): The image size used as input for the YOLO model. Defaults to 640.
        fraction (float, optional): The fraction of available CUDA memory to use. Defaults to 0.60.
        batch_size (int, optional): The default batch size to use if an error is detected. Defaults to 16.

    Returns:
        (int): The optimal batch size.
    """
    # Check device
    prefix = colorstr("AutoBatch: ")
    LOGGER.info(f"{prefix}Computing optimal batch size for imgsz={imgsz} at {fraction * 100}% CUDA memory utilization.")
    device = next(model.parameters()).device  # get model device
    if device.type in {"cpu", "mps"}:
        LOGGER.info(f"{prefix} ⚠️ intended for CUDA devices, using default batch-size {batch_size}")
        return batch_size
    if torch.backends.cudnn.benchmark:
        LOGGER.info(f"{prefix} ⚠️ Requires torch.backends.cudnn.benchmark=False, using default batch-size {batch_size}")
        return batch_size

    # Inspect CUDA memory
    gb = 1 << 30  # bytes to GiB (1024 ** 3)
    d = f"CUDA:{os.getenv('CUDA_VISIBLE_DEVICES', '0').strip()[0]}"  # 'CUDA:0'
    properties = torch.cuda.get_device_properties(device)  # device properties
    t = properties.total_memory / gb  # GiB total
    r = torch.cuda.memory_reserved(device) / gb  # GiB reserved
    a = torch.cuda.memory_allocated(device) / gb  # GiB allocated
    f = t - (r + a)  # GiB free
    LOGGER.info(f"{prefix}{d} ({properties.name}) {t:.2f}G total, {r:.2f}G reserved, {a:.2f}G allocated, {f:.2f}G free")

    # Profile batch sizes
    batch_sizes = [1, 2, 4, 8, 16] if t < 16 else [1, 2, 4, 8, 16, 32, 64]
    try:
        img = [torch.empty(b, 3, imgsz, imgsz) for b in batch_sizes]
        results = profile(img, model, n=1, device=device)

        # Fit a solution
        y = [x[2] for x in results if x]  # memory [2]
        p = np.polyfit(batch_sizes[: len(y)], y, deg=1)  # first degree polynomial fit
        b = int((f * fraction - p[1]) / p[0])  # y intercept (optimal batch size)
        if None in results:  # some sizes failed
            i = results.index(None)  # first fail index
            if b >= batch_sizes[i]:  # y intercept above failure point
                b = batch_sizes[max(i - 1, 0)]  # select prior safe point
        if b < 1 or b > 1024:  # b outside of safe range
            b = batch_size
            LOGGER.info(f"{prefix}WARNING ⚠️ CUDA anomaly detected, using default batch-size {batch_size}.")

        fraction = (np.polyval(p, b) + r + a) / t  # actual fraction predicted
        LOGGER.info(f"{prefix}Using batch-size {b} for {d} {t * fraction:.2f}G/{t:.2f}G ({fraction * 100:.0f}%) ✅")
        return b
    except Exception as e:
        LOGGER.warning(f"{prefix}WARNING ⚠️ error detected: {e},  using default batch-size {batch_size}.")
        return batch_size
    finally:
        torch.cuda.empty_cache()




📅 Created 11 months ago ✏️ Updated 1 month ago