torch.nn.utils.clip_grad.clip_grad_value_# torch.nn.utils.clip_grad.clip_grad_value_(parameters, clip_value, foreach=None)[source]# Clip the gradients of an iterable of parameters at specified value. Gradients are modified in-place. Parameters parameters (Iterable[Tensor] or Tensor) – an iterable of Tensors or a single Tensor that will have gradients normalized clip_value (float) – maximum allowed value of the gradients. The gradients are clipped in the range [-clip_value,clip_value]\left[\text{-clip\_value}, \text{clip\_value}\right][-clip_value,clip_value] foreach (bool, optional) – use the faster foreach-based implementation If None, use the foreach implementation for CUDA and CPU native tensors and silently fall back to the slow implementation for other device types. Default: None
