{ "function_name": "training_losses", "file_path": "LightningDiT/transport/transport.py", "function_calls": [ { "function_call_name": "sample", "functionality": "Samples a random time 't' and the corresponding start (x0, typically noise) and end (x1, the data) points of the transport path.", "input": "self, x1, sp_timesteps, shifted_mu", "output": "A tuple (t, x0, x1) containing the time, start point, and end point.", "location_or_namespace": "LightningDiT/transport/transport.py" }, { "function_call_name": "plan", "functionality": "Based on a sampled time 't' and path endpoints 'x0' and 'x1', computes the interpolated state 'xt' and the target velocity 'ut' at that point on the path.", "input": "self, t, x0, x1", "output": "A tuple (t, xt, ut) containing the time, interpolated state, and target velocity.", "location_or_namespace": "LightningDiT/transport/path_sampler.py" }, { "function_call_name": "mean_flat", "functionality": "Calculates the mean of a tensor over all dimensions except the first (batch) dimension.", "input": "tensor", "output": "A tensor of shape (B,) where B is the batch size of the input tensor.", "location_or_namespace": "LightningDiT/transport/utils.py" }, { "function_call_name": "cosine_similarity", "functionality": "Computes the cosine similarity between two tensors along a given dimension.", "input": "input1, input2, dim=1, eps=1e-8", "output": "A tensor containing the cosine similarity values.", "location_or_namespace": "torch.nn.functional" }, { "function_call_name": "compute_drift", "functionality": "Computes the drift and its variance for the transport path at a given state and time.", "input": "self, x, t", "output": "A tuple containing the drift and its variance.", "location_or_namespace": "LightningDiT/transport/path_sampler.py" }, { "function_call_name": "compute_sigma_t", "functionality": "Computes the noise standard deviation 'sigma_t' at a given time 't'.", "input": "self, t", "output": "A tuple containing sigma_t and its time derivative.", "location_or_namespace": "LightningDiT/transport/path_sampler.py" }, { "function_call_name": "expand_t_like_x", "functionality": "Reshapes the time tensor 't' to be broadcastable with the data tensor 'x'.", "input": "t, x", "output": "A reshaped time tensor.", "location_or_namespace": "LightningDiT/transport/path.py" }, { "function_call_name": "NotImplementedError", "functionality": "Raises a NotImplementedError exception, indicating that an unimplemented code path has been reached.", "input": "message", "output": "Raises an exception.", "location_or_namespace": "builtins" } ] }