(tensor: torch.Tensor, dim)

(self, batch: Any, batch_idx: int, phase='train')

(self, batch: Any, is_train=True)

(self, path)

(x_sbhd, cos, sin)

(x_sbhd, cos, sin)

( qkv, cos, sin, cos_k=None, sin_k=None, interleaved=False, inplace=False, con

(dk, dv, # Q, k, v, sm_scale, # DO, # M, D, #

(dq, q, K, V, # do, m, D, # shared by Q/K/V/DO. stride_tok

(acc, l_i, m_i, q, # K_block_ptr, V_block_ptr, # start_m, qk_scale,

(acc, l_i, m_i, q, k_ptrs, v_ptrs, bias_ptrs, stride_kn, stride_vk, stride_bn, start_m, ac

(dk, dv, Q, k, v, sm_scale, alibi_slope, DO, M, D, # shared by Q/K/V/DO.

(dq, q, K, V, do, m, D, alibi_slope, # shared by Q/K/V/DO. stride_tok, s

(self, device=None)

(self, dataset, **loader_args)

(self, path, skip_warmup)

( dy, x, weight, bias, eps, mean, rstd, dresidual=None, dy1=None, weig

( x, weight, bias, eps, residual=None, x1=None, weight1=None, bias1=None,

(self, dataset, **kwargs)

(self, seqlen, device=None, dtype=None)

(path)

(x, y)

(self, i)

(dtype)

(flop, time)

(x, cos_in, sin_in, ro_dim)

(x, cos_in, sin_in, ro_dim)

()

()

(self)

(self)

(x, residual, norm)

(x, residual, norm)

(s)

(s)

(self, path)

(ref, pred)

(ref, pred)

(ref, pred)

(debug: bool)

(self, persistent)

(self, **kwargs)

(self, preds: Tensor, target: Tensor, loss: Optional[Tensor] = None)