Hey, cool project, code should run faster if this is initialized on the gpu instead of cpu `token_grads = torch.empty( (input_ids.shape[0], input_ids.shape[1], embed.num_embeddings), dtype=torch.float, )`
Hey, cool project, code should run faster if this is initialized on the gpu instead of cpu
token_grads = torch.empty( (input_ids.shape[0], input_ids.shape[1], embed.num_embeddings), dtype=torch.float, )