# Adapt from https://github.com/fla-org/flash-linear-attention/blob/main/fla/ops/gated_delta_rule/fused_recurrent.py # -*- coding: utf-8 -*- # Copyright (c) 2023-2025, Songlin Yang, Yu Zhang from typing import Optional, Tuple import torch import triton import triton.language as tl from sglang.srt.layers.attention.fla.op import exp from sglang.srt.layers.attention.fla.utils import input_guard @triton.jit(do_not_specialize=["T"]) def fused_recurrent_gated_delta_rule_fwd_kernel( q, k, v, g, beta, o, h0, ht, cu_seqlens, scale, T, B: tl.constexpr, H: tl.constexpr, HV: tl.constexpr, K: tl.constexpr, V: tl.constexpr, BK: tl.constexpr, BV: tl.constexpr, USE_INITIAL_STATE: tl.constexpr, # whether to use initial state STORE_FINAL_STATE: tl.constexpr, # whether to store final state IS_BETA_HEADWISE: tl.constexpr, # whether beta is headwise vector or scalar, USE_QK_L2NORM_IN_KERNEL: tl.constexpr, IS_VARLEN: tl.constexpr, IS_KDA: tl.constexpr, ): i_k, i_v, i_nh = tl.program_id(0), tl.program_id(1), tl.program_id(2) i_n, i_hv = i_nh // HV, i_nh % HV i_h = i_hv // (HV // H) if IS_VARLEN: bos, eos = tl.load(cu_seqlens + i_n).to(tl.int64), tl.load( cu_seqlens + i_n + 1 ).to(tl.int64) all = T T = eos - bos else: bos, eos = i_n * T, i_n * T + T all = B * T o_k = i_k * BK + tl.arange(0, BK) o_v = i_v * BV + tl.arange(0, BV) p_q = q + (bos * H + i_h) * K + o_k p_k = k + (bos * H + i_h) * K + o_k p_v = v + (bos * HV + i_hv) * V + o_v if IS_BETA_HEADWISE: p_beta = beta + (bos * HV + i_hv) * V + o_v else: p_beta = beta + bos * HV + i_hv if not IS_KDA: p_g = g + bos * HV + i_hv else: p_gk = g + (bos * H + i_h) * K + o_k p_o = o + ((i_k * all + bos) * HV + i_hv) * V + o_v mask_k = o_k < K mask_v = o_v < V mask_h = mask_v[:, None] & mask_k[None, :] b_h = tl.zeros([BV, BK], dtype=tl.float32) if USE_INITIAL_STATE: p_h0 = h0 + i_nh * V * K + o_v[:, None] * K + o_k[None, :] b_h += tl.load(p_h0, mask=mask_h, other=0).to(tl.float32) for _ in range(0, T): b_q = tl.load(p_q, mask=mask_k, other=0).to(tl.float32) b_k = tl.load(p_k, mask=mask_k, other=0).to(tl.float32) b_v = tl.load(p_v, mask=mask_v, other=0).to(tl.float32) if USE_QK_L2NORM_IN_KERNEL: b_q = b_q / (tl.sqrt(tl.sum(b_q * b_q) + 1e-6)) b_k = b_k / (tl.sqrt(tl.sum(b_k * b_k) + 1e-6)) b_q = b_q * scale # [BV, BK] if not IS_KDA: b_g = tl.load(p_g).to(tl.float32) b_h *= exp(b_g) else: b_gk = tl.load(p_gk, mask=mask_k, other=0).to(tl.float32) b_h *= exp(b_gk[None, :]) # [BV] b_v -= tl.sum(b_h * b_k[None, :], 1) if IS_BETA_HEADWISE: b_beta = tl.load(p_beta, mask=mask_v, other=0).to(tl.float32) else: b_beta = tl.load(p_beta).to(tl.float32) b_v *= b_beta # [BV, BK] b_h += b_v[:, None] * b_k[None, :] # [BV] b_o = tl.sum(b_h * b_q[None, :], 1) tl.store(p_o, b_o.to(p_o.dtype.element_ty), mask=mask_v) p_q += H * K p_k += H * K p_o += HV * V p_v += HV * V if not IS_KDA: p_g += HV else: p_gk += H * K p_beta += HV * (V if IS_BETA_HEADWISE else 1) if STORE_FINAL_STATE: p_ht = ht + i_nh * V * K + o_v[:, None] * K + o_k[None, :] tl.store(p_ht, b_h.to(p_ht.dtype.element_ty), mask=mask_h) def fused_recurrent_gated_delta_rule_fwd( q: torch.Tensor, k: torch.Tensor, v: torch.Tensor, g: torch.Tensor, beta: torch.Tensor, scale: float, initial_state: torch.Tensor, output_final_state: bool, use_qk_l2norm_in_kernel: bool = False, cu_seqlens: Optional[torch.LongTensor] = None, ) -> Tuple[torch.Tensor, torch.Tensor]: B, T, H, K, V = *k.shape, v.shape[-1] HV = v.shape[2] N = B if cu_seqlens is None else len(cu_seqlens) - 1 BK, BV = triton.next_power_of_2(K), min(triton.next_power_of_2(V), 32) NK, NV = triton.cdiv(K, BK), triton.cdiv(V, BV) assert NK == 1, "NK > 1 is not supported yet" num_stages = 3 num_warps = 1 o = q.new_empty(NK, *v.shape) if output_final_state: final_state = q.new_empty(N, HV, V, K, dtype=torch.float32) else: final_state = None grid = (NK, NV, N * HV) fused_recurrent_gated_delta_rule_fwd_kernel[grid]( q=q, k=k, v=v, g=g, beta=beta, o=o, h0=initial_state, ht=final_state, cu_seqlens=cu_seqlens, scale=scale, T=T, B=B, H=H, HV=HV, K=K, V=V, BK=BK, BV=BV, USE_INITIAL_STATE=initial_state is not None, STORE_FINAL_STATE=final_state is not None, IS_BETA_HEADWISE=beta.ndim == v.ndim, USE_QK_L2NORM_IN_KERNEL=use_qk_l2norm_in_kernel, IS_VARLEN=cu_seqlens is not None, IS_KDA=False, num_warps=num_warps, num_stages=num_stages, ) o = o.squeeze(0) return o, final_state # Adapted from vllm project. @triton.jit def fused_recurrent_gated_delta_rule_packed_decode_kernel( mixed_qkv, a, b, A_log, dt_bias, o, h0, ht, ssm_state_indices, scale, stride_mixed_qkv_tok: tl.constexpr, stride_a_tok: tl.constexpr, stride_b_tok: tl.constexpr, stride_init_state_token: tl.constexpr, stride_final_state_token: tl.constexpr, stride_indices_seq: tl.constexpr, H: tl.constexpr, HV: tl.constexpr, K: tl.constexpr, V: tl.constexpr, BK: tl.constexpr, BV: tl.constexpr, SOFTPLUS_THRESHOLD: tl.constexpr, USE_QK_L2NORM_IN_KERNEL: tl.constexpr, ): i_v, i_nh = tl.program_id(0), tl.program_id(1) i_n, i_hv = i_nh // HV, i_nh % HV i_h = i_hv // (HV // H) o_k = tl.arange(0, BK) o_v = i_v * BV + tl.arange(0, BV) mask_k = o_k < K mask_v = o_v < V mask_h = mask_v[:, None] & mask_k[None, :] state_idx = tl.load(ssm_state_indices + i_n * stride_indices_seq).to(tl.int64) p_o = o + (i_n * HV + i_hv) * V + o_v if state_idx < 0: zero = tl.zeros([BV], dtype=tl.float32).to(p_o.dtype.element_ty) tl.store(p_o, zero, mask=mask_v) return p_h0 = h0 + state_idx * stride_init_state_token p_h0 = p_h0 + i_hv * V * K + o_v[:, None] * K + o_k[None, :] b_h = tl.load(p_h0, mask=mask_h, other=0).to(tl.float32) p_mixed = mixed_qkv + i_n * stride_mixed_qkv_tok q_off = i_h * K + o_k k_off = (H * K) + i_h * K + o_k v_off = (2 * H * K) + i_hv * V + o_v b_q = tl.load(p_mixed + q_off, mask=mask_k, other=0).to(tl.float32) b_k = tl.load(p_mixed + k_off, mask=mask_k, other=0).to(tl.float32) b_v = tl.load(p_mixed + v_off, mask=mask_v, other=0).to(tl.float32) if USE_QK_L2NORM_IN_KERNEL: b_q = b_q / tl.sqrt(tl.sum(b_q * b_q) + 1e-6) b_k = b_k / tl.sqrt(tl.sum(b_k * b_k) + 1e-6) b_q = b_q * scale a_val = tl.load(a + i_n * stride_a_tok + i_hv).to(tl.float32) b_val = tl.load(b + i_n * stride_b_tok + i_hv).to(tl.float32) A_log_val = tl.load(A_log + i_hv).to(tl.float32) dt_bias_val = tl.load(dt_bias + i_hv).to(tl.float32) x = a_val + dt_bias_val softplus_x = tl.where(x <= SOFTPLUS_THRESHOLD, tl.log(1.0 + tl.exp(x)), x) g_val = -tl.exp(A_log_val) * softplus_x beta_val = tl.sigmoid(b_val).to(b.dtype.element_ty).to(tl.float32) b_h *= exp(g_val) b_v -= tl.sum(b_h * b_k[None, :], 1) b_v *= beta_val b_h += b_v[:, None] * b_k[None, :] b_o = tl.sum(b_h * b_q[None, :], 1) tl.store(p_o, b_o.to(p_o.dtype.element_ty), mask=mask_v) p_ht = ht + state_idx * stride_final_state_token p_ht = p_ht + i_hv * V * K + o_v[:, None] * K + o_k[None, :] tl.store(p_ht, b_h.to(p_ht.dtype.element_ty), mask=mask_h) def fused_recurrent_gated_delta_rule_packed_decode( mixed_qkv: torch.Tensor, a: torch.Tensor, b: torch.Tensor, A_log: torch.Tensor, dt_bias: torch.Tensor, scale: float, initial_state: torch.Tensor, out: torch.Tensor, ssm_state_indices: torch.Tensor, use_qk_l2norm_in_kernel: bool = False, ) -> tuple[torch.Tensor, torch.Tensor]: if mixed_qkv.ndim != 2: raise ValueError( f"`mixed_qkv` must be a 2D tensor (got ndim={mixed_qkv.ndim})." ) if mixed_qkv.stride(-1) != 1: raise ValueError("`mixed_qkv` must be contiguous in the last dim.") if a.ndim != 2 or b.ndim != 2: raise ValueError( f"`a` and `b` must be 2D tensors (got a.ndim={a.ndim}, b.ndim={b.ndim})." ) if a.stride(-1) != 1 or b.stride(-1) != 1: raise ValueError("`a`/`b` must be contiguous in the last dim.") if A_log.ndim != 1 or dt_bias.ndim != 1: raise ValueError("`A_log`/`dt_bias` must be 1D tensors.") if A_log.stride(0) != 1 or dt_bias.stride(0) != 1: raise ValueError("`A_log`/`dt_bias` must be contiguous.") if ssm_state_indices.ndim != 1: raise ValueError( f"`ssm_state_indices` must be 1D for packed decode (got ndim={ssm_state_indices.ndim})." ) if not out.is_contiguous(): raise ValueError("`out` must be contiguous.") dev = mixed_qkv.device if any( t.device != dev for t in (a, b, A_log, dt_bias, initial_state, out, ssm_state_indices) ): raise ValueError("All inputs must be on the same device.") B = mixed_qkv.shape[0] if a.shape[0] != B or b.shape[0] != B: raise ValueError( "Mismatched batch sizes: " f"mixed_qkv.shape[0]={B}, a.shape[0]={a.shape[0]}, b.shape[0]={b.shape[0]}." ) if ssm_state_indices.shape[0] != B: raise ValueError( f"`ssm_state_indices` must have shape [B] (got {tuple(ssm_state_indices.shape)}; expected ({B},))." ) if initial_state.ndim != 4: raise ValueError( f"`initial_state` must be a 4D tensor (got ndim={initial_state.ndim})." ) if initial_state.stride(-1) != 1: raise ValueError("`initial_state` must be contiguous in the last dim.") HV, V, K = initial_state.shape[-3:] if a.shape[1] != HV or b.shape[1] != HV: raise ValueError( f"`a`/`b` must have shape [B, HV] with HV={HV} (got a.shape={tuple(a.shape)}, b.shape={tuple(b.shape)})." ) if A_log.numel() != HV or dt_bias.numel() != HV: raise ValueError( f"`A_log` and `dt_bias` must have {HV} elements (got A_log.numel()={A_log.numel()}, dt_bias.numel()={dt_bias.numel()})." ) if out.shape != (B, 1, HV, V): raise ValueError( f"`out` must have shape {(B, 1, HV, V)} (got out.shape={tuple(out.shape)})." ) qkv_dim = mixed_qkv.shape[1] qk_dim = qkv_dim - HV * V if qk_dim <= 0 or qk_dim % 2 != 0: raise ValueError( f"Invalid packed `mixed_qkv` last dim={qkv_dim} for HV={HV}, V={V}." ) q_dim = qk_dim // 2 if q_dim % K != 0: raise ValueError(f"Invalid packed Q size {q_dim}: must be divisible by K={K}.") H = q_dim // K if H <= 0 or HV % H != 0: raise ValueError( f"Invalid head config inferred from mixed_qkv: H={H}, HV={HV}." ) BK = triton.next_power_of_2(K) if triton.cdiv(K, BK) != 1: raise ValueError( f"Packed decode kernel only supports NK=1 (got K={K}, BK={BK})." ) BV = min(triton.next_power_of_2(V), 32) num_stages = 3 num_warps = 1 stride_mixed_qkv_tok = mixed_qkv.stride(0) stride_a_tok = a.stride(0) stride_b_tok = b.stride(0) stride_init_state_token = initial_state.stride(0) stride_final_state_token = initial_state.stride(0) stride_indices_seq = ssm_state_indices.stride(0) NV = triton.cdiv(V, BV) grid = (NV, B * HV) fused_recurrent_gated_delta_rule_packed_decode_kernel[grid]( mixed_qkv=mixed_qkv, a=a, b=b, A_log=A_log, dt_bias=dt_bias, o=out, h0=initial_state, ht=initial_state, ssm_state_indices=ssm_state_indices, scale=scale, stride_mixed_qkv_tok=stride_mixed_qkv_tok, stride_a_tok=stride_a_tok, stride_b_tok=stride_b_tok, stride_init_state_token=stride_init_state_token, stride_final_state_token=stride_final_state_token, stride_indices_seq=stride_indices_seq, H=H, HV=HV, K=K, V=V, BK=BK, BV=BV, SOFTPLUS_THRESHOLD=20.0, USE_QK_L2NORM_IN_KERNEL=use_qk_l2norm_in_kernel, num_warps=num_warps, num_stages=num_stages, ) return out, initial_state @triton.jit def fused_recurrent_kda_packed_decode_kernel( mixed_qkv, a, b, A_log, dt_bias, o, h0, ht, ssm_state_indices, scale, stride_mixed_qkv_tok: tl.constexpr, stride_a_tok: tl.constexpr, stride_b_tok: tl.constexpr, stride_init_state_token: tl.constexpr, stride_final_state_token: tl.constexpr, stride_indices_seq: tl.constexpr, H: tl.constexpr, HV: tl.constexpr, K: tl.constexpr, V: tl.constexpr, BK: tl.constexpr, BV: tl.constexpr, SOFTPLUS_THRESHOLD: tl.constexpr, USE_QK_L2NORM_IN_KERNEL: tl.constexpr, ): """KDA packed decode: same shape as the GDN packed decode kernel, but with a per-K gate (``a`` is ``[B, HV*K]`` and ``dt_bias`` is ``[HV*K]``), so the state decay is a per-K vector ``exp(g)`` rather than a scalar.""" i_v, i_nh = tl.program_id(0), tl.program_id(1) i_n, i_hv = i_nh // HV, i_nh % HV i_h = i_hv // (HV // H) o_k = tl.arange(0, BK) o_v = i_v * BV + tl.arange(0, BV) mask_k = o_k < K mask_v = o_v < V mask_h = mask_v[:, None] & mask_k[None, :] state_idx = tl.load(ssm_state_indices + i_n * stride_indices_seq).to(tl.int64) p_o = o + (i_n * HV + i_hv) * V + o_v if state_idx < 0: zero = tl.zeros([BV], dtype=tl.float32).to(p_o.dtype.element_ty) tl.store(p_o, zero, mask=mask_v) return p_h0 = h0 + state_idx * stride_init_state_token p_h0 = p_h0 + i_hv * V * K + o_v[:, None] * K + o_k[None, :] b_h = tl.load(p_h0, mask=mask_h, other=0).to(tl.float32) p_mixed = mixed_qkv + i_n * stride_mixed_qkv_tok q_off = i_h * K + o_k k_off = (H * K) + i_h * K + o_k v_off = (2 * H * K) + i_hv * V + o_v b_q = tl.load(p_mixed + q_off, mask=mask_k, other=0).to(tl.float32) b_k = tl.load(p_mixed + k_off, mask=mask_k, other=0).to(tl.float32) b_v = tl.load(p_mixed + v_off, mask=mask_v, other=0).to(tl.float32) if USE_QK_L2NORM_IN_KERNEL: b_q = b_q / tl.sqrt(tl.sum(b_q * b_q) + 1e-6) b_k = b_k / tl.sqrt(tl.sum(b_k * b_k) + 1e-6) b_q = b_q * scale # KDA per-K gate: load BK values of ``a`` and ``dt_bias`` for this head. p_a = a + i_n * stride_a_tok + i_hv * K + o_k p_dt = dt_bias + i_hv * K + o_k b_a = tl.load(p_a, mask=mask_k, other=0).to(tl.float32) b_dt = tl.load(p_dt, mask=mask_k, other=0).to(tl.float32) A_log_val = tl.load(A_log + i_hv).to(tl.float32) x = b_a + b_dt softplus_x = tl.where(x <= SOFTPLUS_THRESHOLD, tl.log(1.0 + tl.exp(x)), x) b_g = -tl.exp(A_log_val) * softplus_x # [BK] b_val = tl.load(b + i_n * stride_b_tok + i_hv).to(tl.float32) # Keep beta in fp32 (no bf16 round-trip) to match the generic decode # kernel `fused_sigmoid_gating_delta_rule_update`, which is the reference # validated against torch. beta_val = tl.sigmoid(b_val).to(tl.float32) # Per-K decay: each K-row of the [V, K] state decays by its own exp(g_k). b_h *= exp(b_g)[None, :] b_v -= tl.sum(b_h * b_k[None, :], 1) b_v *= beta_val b_h += b_v[:, None] * b_k[None, :] b_o = tl.sum(b_h * b_q[None, :], 1) tl.store(p_o, b_o.to(p_o.dtype.element_ty), mask=mask_v) p_ht = ht + state_idx * stride_final_state_token p_ht = p_ht + i_hv * V * K + o_v[:, None] * K + o_k[None, :] tl.store(p_ht, b_h.to(p_ht.dtype.element_ty), mask=mask_h) def fused_recurrent_kda_packed_decode( mixed_qkv: torch.Tensor, a: torch.Tensor, b: torch.Tensor, A_log: torch.Tensor, dt_bias: torch.Tensor, scale: float, initial_state: torch.Tensor, out: torch.Tensor, ssm_state_indices: torch.Tensor, use_qk_l2norm_in_kernel: bool = False, ) -> tuple[torch.Tensor, torch.Tensor]: """KDA T=1 decode fast path. Mirrors ``fused_recurrent_gated_delta_rule_packed_decode`` but the gate ``g`` is a per-K vector instead of a scalar. Args: mixed_qkv: ``[B, 2*H*K + HV*V]`` packed projection output after conv1d. Requires ``num_q_heads == num_k_heads == H`` and ``head_q_dim == head_k_dim == K``. a: ``[B, HV*K]`` per-K gate input (typically reshaped from ``[B, HV, K]``). b: ``[B, HV]`` beta input (post-sigmoid scalar per head). A_log: ``[HV]`` log-space decay parameter. dt_bias: ``[HV*K]`` per-K time-step bias. scale: attention scale factor (typically ``head_k_dim ** -0.5``). initial_state: ``[num_slots, HV, V, K]`` full state pool, updated in place. out: ``[B, 1, HV, V]`` contiguous output buffer. ssm_state_indices: ``[B]`` per-request state slot indices (-1 = skip). use_qk_l2norm_in_kernel: apply per-head L2 norm to Q/K inside the kernel. """ if mixed_qkv.ndim != 2: raise ValueError( f"`mixed_qkv` must be a 2D tensor (got ndim={mixed_qkv.ndim})." ) if mixed_qkv.stride(-1) != 1: raise ValueError("`mixed_qkv` must be contiguous in the last dim.") if a.ndim != 2 or b.ndim != 2: raise ValueError( f"`a` and `b` must be 2D tensors (got a.ndim={a.ndim}, b.ndim={b.ndim})." ) if a.stride(-1) != 1 or b.stride(-1) != 1: raise ValueError("`a`/`b` must be contiguous in the last dim.") if A_log.ndim != 1 or dt_bias.ndim != 1: raise ValueError("`A_log`/`dt_bias` must be 1D tensors.") if A_log.stride(0) != 1 or dt_bias.stride(0) != 1: raise ValueError("`A_log`/`dt_bias` must be contiguous.") if ssm_state_indices.ndim != 1: raise ValueError( f"`ssm_state_indices` must be 1D for packed decode (got ndim={ssm_state_indices.ndim})." ) if not out.is_contiguous(): raise ValueError("`out` must be contiguous.") dev = mixed_qkv.device if any( t.device != dev for t in (a, b, A_log, dt_bias, initial_state, out, ssm_state_indices) ): raise ValueError("All inputs must be on the same device.") B = mixed_qkv.shape[0] if a.shape[0] != B or b.shape[0] != B: raise ValueError( "Mismatched batch sizes: " f"mixed_qkv.shape[0]={B}, a.shape[0]={a.shape[0]}, b.shape[0]={b.shape[0]}." ) if ssm_state_indices.shape[0] != B: raise ValueError( f"`ssm_state_indices` must have shape [B] (got {tuple(ssm_state_indices.shape)}; expected ({B},))." ) if initial_state.ndim != 4: raise ValueError( f"`initial_state` must be a 4D tensor (got ndim={initial_state.ndim})." ) if initial_state.stride(-1) != 1: raise ValueError("`initial_state` must be contiguous in the last dim.") HV, V, K = initial_state.shape[-3:] if a.shape[1] != HV * K: raise ValueError( f"`a` must have shape [B, HV*K] with HV={HV}, K={K} " f"(got a.shape={tuple(a.shape)})." ) if b.shape[1] != HV: raise ValueError( f"`b` must have shape [B, HV] with HV={HV} (got b.shape={tuple(b.shape)})." ) if A_log.numel() != HV: raise ValueError(f"`A_log` must have {HV} elements (got {A_log.numel()}).") if dt_bias.numel() != HV * K: raise ValueError( f"`dt_bias` must have {HV * K} elements (got {dt_bias.numel()})." ) if out.shape != (B, 1, HV, V): raise ValueError( f"`out` must have shape {(B, 1, HV, V)} (got out.shape={tuple(out.shape)})." ) qkv_dim = mixed_qkv.shape[1] qk_dim = qkv_dim - HV * V if qk_dim <= 0 or qk_dim % 2 != 0: raise ValueError( f"Invalid packed `mixed_qkv` last dim={qkv_dim} for HV={HV}, V={V}." ) q_dim = qk_dim // 2 if q_dim % K != 0: raise ValueError( f"Invalid packed Q size {q_dim}: must be divisible by K={K}. " "KDA packed decode requires num_q_heads == num_k_heads and " "head_q_dim == head_k_dim." ) H = q_dim // K if H <= 0 or HV % H != 0: raise ValueError( f"Invalid head config inferred from mixed_qkv: H={H}, HV={HV}." ) BK = triton.next_power_of_2(K) if triton.cdiv(K, BK) != 1: raise ValueError( f"Packed decode kernel only supports NK=1 (got K={K}, BK={BK})." ) BV = min(triton.next_power_of_2(V), 32) num_stages = 3 num_warps = 1 stride_mixed_qkv_tok = mixed_qkv.stride(0) stride_a_tok = a.stride(0) stride_b_tok = b.stride(0) stride_init_state_token = initial_state.stride(0) stride_final_state_token = initial_state.stride(0) stride_indices_seq = ssm_state_indices.stride(0) NV = triton.cdiv(V, BV) grid = (NV, B * HV) fused_recurrent_kda_packed_decode_kernel[grid]( mixed_qkv=mixed_qkv, a=a, b=b, A_log=A_log, dt_bias=dt_bias, o=out, h0=initial_state, ht=initial_state, ssm_state_indices=ssm_state_indices, scale=scale, stride_mixed_qkv_tok=stride_mixed_qkv_tok, stride_a_tok=stride_a_tok, stride_b_tok=stride_b_tok, stride_init_state_token=stride_init_state_token, stride_final_state_token=stride_final_state_token, stride_indices_seq=stride_indices_seq, H=H, HV=HV, K=K, V=V, BK=BK, BV=BV, SOFTPLUS_THRESHOLD=20.0, USE_QK_L2NORM_IN_KERNEL=use_qk_l2norm_in_kernel, num_warps=num_warps, num_stages=num_stages, ) return out, initial_state class FusedRecurrentFunction(torch.autograd.Function): @staticmethod @input_guard def forward( ctx, q: torch.Tensor, k: torch.Tensor, v: torch.Tensor, g: torch.Tensor, beta: torch.Tensor, scale: float, initial_state: torch.Tensor, output_final_state: bool, cu_seqlens: Optional[torch.LongTensor] = None, use_qk_l2norm_in_kernel: bool = False, ): o, final_state = fused_recurrent_gated_delta_rule_fwd( q=q, k=k, v=v, g=g, beta=beta, scale=scale, initial_state=initial_state, output_final_state=output_final_state, use_qk_l2norm_in_kernel=use_qk_l2norm_in_kernel, cu_seqlens=cu_seqlens, ) return o, final_state @staticmethod @input_guard def backward(ctx, do, dht): raise NotImplementedError( "Backward pass is not implemented yet and we do not have plans to implement it " "because we haven't figured out how to compute dg without materializing the full " "hidden states for all time steps." ) def fused_recurrent_gated_delta_rule( q: torch.Tensor, k: torch.Tensor, v: torch.Tensor, g: torch.Tensor, beta: torch.Tensor = None, scale: float = None, initial_state: torch.Tensor = None, output_final_state: bool = False, cu_seqlens: Optional[torch.LongTensor] = None, use_qk_l2norm_in_kernel: bool = False, ) -> Tuple[torch.Tensor, torch.Tensor]: r""" Args: q (torch.Tensor): queries of shape `[B, T, H, K]`. k (torch.Tensor): keys of shape `[B, T, H, K]`. v (torch.Tensor): values of shape `[B, T, HV, V]`. GVA is applied if `HV > H`. g (torch.Tensor): g (decays) of shape `[B, T, HV]`. beta (torch.Tensor): betas of shape `[B, T, HV]`. scale (Optional[int]): Scale factor for the RetNet attention scores. If not provided, it will default to `1 / sqrt(K)`. Default: `None`. initial_state (Optional[torch.Tensor]): Initial state of shape `[N, HV, V, K]` for `N` input sequences. For equal-length input sequences, `N` equals the batch size `B`. Default: `None`. output_final_state (Optional[bool]): Whether to output the final state of shape `[N, HV, V, K]`. Default: `False`. cu_seqlens (torch.LongTensor): Cumulative sequence lengths of shape `[N+1]` used for variable-length training, consistent with the FlashAttention API. Returns: o (torch.Tensor): Outputs of shape `[B, T, HV, V]`. final_state (torch.Tensor): Final state of shape `[N, HV, V, K]` if `output_final_state=True` else `None`. Examples:: >>> import torch >>> import torch.nn.functional as F >>> from einops import rearrange >>> from fla.ops.gated_delta_rule import fused_recurrent_gated_delta_rule # inputs with equal lengths >>> B, T, H, HV, K, V = 4, 2048, 4, 8, 512, 512 >>> q = torch.randn(B, T, H, K, device='cuda') >>> k = F.normalize(torch.randn(B, T, H, K, device='cuda'), p=2, dim=-1) >>> v = torch.randn(B, T, HV, V, device='cuda') >>> g = F.logsigmoid(torch.rand(B, T, HV, device='cuda')) >>> beta = torch.rand(B, T, HV, device='cuda').sigmoid() >>> h0 = torch.randn(B, HV, V, K, device='cuda') >>> o, ht = fused_gated_recurrent_delta_rule( q, k, v, g, beta, initial_state=h0, output_final_state=True ) # for variable-length inputs, the batch size `B` is expected to be 1 and `cu_seqlens` is required >>> q, k, v, g, beta = map(lambda x: rearrange(x, 'b t ... -> 1 (b t) ...'), (q, k, v, g, beta)) # for a batch with 4 sequences, `cu_seqlens` with 5 start/end positions are expected >>> cu_seqlens = q.new_tensor([0, 2048, 4096, 6144, 8192], dtype=torch.long) >>> o_var, ht_var = fused_gated_recurrent_delta_rule( q, k, v, g, beta, initial_state=h0, output_final_state=True, cu_seqlens=cu_seqlens ) """ if cu_seqlens is not None: if q.shape[0] != 1: raise ValueError( f"The batch size is expected to be 1 rather than {q.shape[0]} when using `cu_seqlens`." f"Please flatten variable-length inputs before processing." ) if initial_state is not None and initial_state.shape[0] != len(cu_seqlens) - 1: raise ValueError( f"The number of initial states is expected to be equal to the number of input sequences, " f"i.e., {len(cu_seqlens) - 1} rather than {initial_state.shape[0]}." ) if scale is None: scale = k.shape[-1] ** -0.5 else: assert scale > 0, "scale must be positive" if beta is None: beta = torch.ones_like(q[..., 0]) o, final_state = FusedRecurrentFunction.apply( q, k, v, g, beta, scale, initial_state, output_final_state, cu_seqlens, use_qk_l2norm_in_kernel, ) return o, final_state # HAS_EAGLE_TREE_CUSTOM_ATTN_MASK is added to support eagle tree attention mask # retrieve_parent_token_ptr: [N, NP2_T], retrieve_next_sibling_ptr: [N, NP2_T] # e.g. for a sequence of length 4, the eagle tree attention structure is: # retrieve_next_token=[1, 3, -1, -1] -> retrieve_next_token[i]: the 1st child token of token i # retrieve_next_sibling=[-1, 2, -1, -1] -> retrieve_next_sibling[i]: the 1st tree sibling token of token i # retrieve_parent_token=[n/a, 0, 0, 1] -> retrieve_parent_token[i]: the parent token of token i # Tree: # 0 # / \ # 1 2 # / # 3 # When calculating token 3's attention, it should attend to token 1 (parent) and token 0 (grand-parent) # When calculating token 2's attention, it should attend to token 0 (parent) @triton.jit(do_not_specialize=["T"]) def fused_recurrent_gated_delta_rule_update_fwd_kernel( q, k, v, g, beta, o, h0_source, h0_indices, cu_seqlens, scale, intermediate_states_buffer, intermediate_state_indices, cache_steps, retrieve_parent_token_ptr, stride_retrieve_parent_token_seq: tl.constexpr, stride_retrieve_parent_token_token: tl.constexpr, T, NP2_T: tl.constexpr, B: tl.constexpr, H: tl.constexpr, HV: tl.constexpr, K: tl.constexpr, V: tl.constexpr, BK: tl.constexpr, BV: tl.constexpr, USE_INITIAL_STATE: tl.constexpr, # whether to use initial state IS_BETA_HEADWISE: tl.constexpr, # whether beta is headwise vector or scalar, USE_QK_L2NORM_IN_KERNEL: tl.constexpr, IS_VARLEN: tl.constexpr, DISABLE_STATE_UPDATE: tl.constexpr, # whether to disable final state update DISABLE_OUTPUT_CALCULATION: tl.constexpr, # whether to disable output calculation CACHE_INTERMEDIATE_STATES: tl.constexpr, HAS_EAGLE_TREE_CUSTOM_ATTN_MASK: tl.constexpr, ): i_k, i_v, i_nh = tl.program_id(0), tl.program_id(1), tl.program_id(2) i_n, i_hv = i_nh // HV, i_nh % HV i_h = i_hv // (HV // H) if IS_VARLEN: bos, eos = tl.load(cu_seqlens + i_n).to(tl.int64), tl.load( cu_seqlens + i_n + 1 ).to(tl.int64) all = T T = eos - bos else: bos, eos = i_n * T, i_n * T + T all = B * T o_k = i_k * BK + tl.arange(0, BK) o_v = i_v * BV + tl.arange(0, BV) p_q = q + (bos * H + i_h) * K + o_k p_k = k + (bos * H + i_h) * K + o_k p_v = v + (bos * HV + i_hv) * V + o_v if IS_BETA_HEADWISE: p_beta = beta + (bos * HV + i_hv) * V + o_v else: p_beta = beta + bos * HV + i_hv p_g = g + bos * HV + i_hv p_o = o + ((i_k * all + bos) * HV + i_hv) * V + o_v if HAS_EAGLE_TREE_CUSTOM_ATTN_MASK: token_indices = tl.arange(0, NP2_T) mask_retrieve = token_indices < T retrieve_parent_token_base = ( retrieve_parent_token_ptr + (i_n * stride_retrieve_parent_token_seq) + token_indices * stride_retrieve_parent_token_token ) parent_idx_tokens = tl.load(retrieve_parent_token_base, mask_retrieve) mask_k = o_k < K mask_v = o_v < V mask_h = mask_v[:, None] & mask_k[None, :] b_h = tl.zeros([BV, BK], dtype=tl.float32) if USE_INITIAL_STATE: idx = tl.load(h0_indices + i_n) # Add bounds checking for idx if idx >= 0: # Assuming negative indices are invalid p_h0 = ( h0_source + idx * HV * K * V + i_hv * K * V + o_v[:, None] * K + o_k[None, :] ) b_h += tl.load(p_h0, mask=mask_h, other=0).to(tl.float32) # Prepare intermediate state cache variables if enabled cache_idx = -1 if CACHE_INTERMEDIATE_STATES: cache_idx = tl.load(intermediate_state_indices + i_n) step_idx = 0 for _ in range(0, T): if HAS_EAGLE_TREE_CUSTOM_ATTN_MASK: # step_idx = 0 should use the b_h from USE_INITIAL_STATE if step_idx != 0 and cache_idx >= 0: # when calculating current step's attention, load the state from the parent token parent_step_idx = tl.sum( tl.where(token_indices == step_idx, parent_idx_tokens, 0) ) step_offset = parent_step_idx * HV * K * V cache_ptr = ( intermediate_states_buffer + cache_idx * cache_steps * HV * K * V + step_offset + i_hv * K * V + o_v[:, None] * K + o_k[None, :] ) b_h = tl.load(cache_ptr, mask=mask_h, other=0).to(tl.float32) b_q = tl.load(p_q, mask=mask_k, other=0).to(tl.float32) b_k = tl.load(p_k, mask=mask_k, other=0).to(tl.float32) b_v = tl.load(p_v, mask=mask_v, other=0).to(tl.float32) b_g = tl.load(p_g).to(tl.float32) if USE_QK_L2NORM_IN_KERNEL: b_q = b_q / (tl.sqrt(tl.sum(b_q * b_q) + 1e-6)) b_k = b_k / (tl.sqrt(tl.sum(b_k * b_k) + 1e-6)) b_q = b_q * scale # [BK, BV] b_h *= exp(b_g) # [BV] b_v -= tl.sum(b_h * b_k[None, :], 1) if IS_BETA_HEADWISE: b_beta = tl.load(p_beta, mask=mask_v, other=0).to(tl.float32) else: b_beta = tl.load(p_beta).to(tl.float32) b_v *= b_beta # [BV, BK] b_h += b_v[:, None] * b_k[None, :] # [BV] if not DISABLE_OUTPUT_CALCULATION: b_o = tl.sum(b_h * b_q[None, :], 1) # core attn output tl.store(p_o, b_o.to(p_o.dtype.element_ty), mask=mask_v) # store intermediate states if enabled if CACHE_INTERMEDIATE_STATES: if cache_idx >= 0: # Compute cache pointer for this step step_offset = step_idx * HV * K * V cache_ptr = ( intermediate_states_buffer + cache_idx * cache_steps * HV * K * V + step_offset + i_hv * K * V + o_v[:, None] * K + o_k[None, :] ) tl.store(cache_ptr, b_h.to(cache_ptr.dtype.element_ty), mask=mask_h) step_idx += 1 p_q += H * K p_k += H * K p_o += HV * V p_v += HV * V p_g += HV p_beta += HV * (V if IS_BETA_HEADWISE else 1) # Store final state back to h0_source with bounds checking # ssm states if not DISABLE_STATE_UPDATE: idx = tl.load(h0_indices + i_n) if idx >= 0: # Add bounds checking p_h0 = ( h0_source + idx * HV * K * V + i_hv * K * V + o_v[:, None] * K + o_k[None, :] ) tl.store(p_h0, b_h.to(p_h0.dtype.element_ty), mask=mask_h) def fused_recurrent_gated_delta_rule_update_fwd( q: torch.Tensor, k: torch.Tensor, v: torch.Tensor, g: torch.Tensor, beta: torch.Tensor, scale: float, initial_state_source: torch.Tensor, initial_state_indices: torch.Tensor, use_qk_l2norm_in_kernel: bool = False, cu_seqlens: Optional[torch.LongTensor] = None, disable_state_update: bool = False, disable_output_calculation: bool = False, intermediate_states_buffer: Optional[torch.Tensor] = None, intermediate_state_indices: Optional[torch.Tensor] = None, cache_steps: Optional[int] = None, retrieve_parent_token: Optional[torch.Tensor] = None, ) -> torch.Tensor: B, T, H, K, V = *k.shape, v.shape[-1] HV = v.shape[2] N = B if cu_seqlens is None else len(cu_seqlens) - 1 BK, BV = triton.next_power_of_2(K), min(triton.next_power_of_2(V), 8) NK, NV = triton.cdiv(K, BK), triton.cdiv(V, BV) assert NK == 1, "NK > 1 is not supported yet" num_stages = 3 num_warps = 1 if disable_output_calculation: # When output calculation is disabled, allocate minimal tensor o = q.new_empty(NK, 1, 1, 1, 1) # minimal allocation else: o = q.new_empty(NK, *v.shape) grid = (NK, NV, N * HV) # prepare retrieve next token buffer strides if provided if retrieve_parent_token is not None: stride_retrieve_parent_token_seq, stride_retrieve_parent_token_token = ( retrieve_parent_token.stride(0), retrieve_parent_token.stride(1), ) else: stride_retrieve_parent_token_seq = stride_retrieve_parent_token_token = 0 NP2_T = triton.next_power_of_2(T) fused_recurrent_gated_delta_rule_update_fwd_kernel[grid]( q=q, k=k, v=v, g=g, beta=beta, o=o, h0_source=initial_state_source, h0_indices=initial_state_indices, cu_seqlens=cu_seqlens, scale=scale, intermediate_states_buffer=intermediate_states_buffer, intermediate_state_indices=intermediate_state_indices, cache_steps=0 if cache_steps is None else cache_steps, retrieve_parent_token_ptr=retrieve_parent_token, stride_retrieve_parent_token_seq=stride_retrieve_parent_token_seq, stride_retrieve_parent_token_token=stride_retrieve_parent_token_token, T=T, NP2_T=NP2_T, B=B, H=H, HV=HV, K=K, V=V, BK=BK, BV=BV, USE_INITIAL_STATE=initial_state_source is not None, IS_VARLEN=cu_seqlens is not None, CACHE_INTERMEDIATE_STATES=intermediate_states_buffer is not None, HAS_EAGLE_TREE_CUSTOM_ATTN_MASK=retrieve_parent_token is not None, IS_BETA_HEADWISE=beta.ndim == v.ndim, USE_QK_L2NORM_IN_KERNEL=use_qk_l2norm_in_kernel, DISABLE_STATE_UPDATE=disable_state_update, DISABLE_OUTPUT_CALCULATION=disable_output_calculation, num_warps=num_warps, num_stages=num_stages, ) o = o.squeeze(0) return o class FusedRecurrentUpdateFunction(torch.autograd.Function): @staticmethod @input_guard def forward( ctx, q: torch.Tensor, k: torch.Tensor, v: torch.Tensor, g: torch.Tensor, beta: torch.Tensor, scale: float, initial_state_source: torch.Tensor, initial_state_indices: torch.Tensor, cu_seqlens: Optional[torch.LongTensor] = None, use_qk_l2norm_in_kernel: bool = False, disable_state_update: bool = False, disable_output_calculation: bool = False, intermediate_states_buffer: Optional[torch.Tensor] = None, intermediate_state_indices: Optional[torch.Tensor] = None, cache_steps: Optional[int] = None, retrieve_parent_token: Optional[torch.Tensor] = None, ): o = fused_recurrent_gated_delta_rule_update_fwd( q=q, k=k, v=v, g=g, beta=beta, scale=scale, initial_state_source=initial_state_source, initial_state_indices=initial_state_indices, use_qk_l2norm_in_kernel=use_qk_l2norm_in_kernel, cu_seqlens=cu_seqlens, disable_state_update=disable_state_update, disable_output_calculation=disable_output_calculation, intermediate_states_buffer=intermediate_states_buffer, intermediate_state_indices=intermediate_state_indices, cache_steps=cache_steps, retrieve_parent_token=retrieve_parent_token, ) return o @staticmethod @input_guard def backward(ctx, do, dht): raise NotImplementedError( "Backward pass is not implemented yet and we do not have plans to implement it " "because we haven't figured out how to compute dg without materializing the full " "hidden states for all time steps." ) def fused_recurrent_gated_delta_rule_update( q: torch.Tensor, k: torch.Tensor, v: torch.Tensor, g: torch.Tensor, beta: torch.Tensor = None, scale: float = None, initial_state_source: torch.Tensor = None, initial_state_indices: torch.Tensor = None, cu_seqlens: Optional[torch.LongTensor] = None, use_qk_l2norm_in_kernel: bool = False, disable_state_update: bool = False, disable_output_calculation: bool = False, intermediate_states_buffer: Optional[torch.Tensor] = None, intermediate_state_indices: Optional[torch.Tensor] = None, cache_steps: Optional[int] = None, retrieve_parent_token: Optional[torch.Tensor] = None, ) -> torch.Tensor: if cu_seqlens is not None: if q.shape[0] != 1: raise ValueError( f"The batch size is expected to be 1 rather than {q.shape[0]} when using `cu_seqlens`." f"Please flatten variable-length inputs before processing." ) if initial_state_source is not None: if initial_state_indices.shape[0] != len(cu_seqlens) - 1: raise ValueError( f"The number of initial states is expected to be equal to the number of input sequences, " f"i.e., {len(cu_seqlens) - 1} rather than {initial_state_indices.shape[0]}." ) if initial_state_indices.shape[0] != intermediate_state_indices.shape[0]: raise ValueError( f"The number of intermediate state indices is expected to be equal to the number of input sequences, " f"i.e., {initial_state_indices.shape[0]} != {intermediate_state_indices.shape[0]}." ) if scale is None: scale = k.shape[-1] ** -0.5 else: assert scale > 0, "scale must be positive" if beta is None: beta = torch.ones_like(q[..., 0]) o = FusedRecurrentUpdateFunction.apply( q, k, v, g, beta, scale, initial_state_source, initial_state_indices, cu_seqlens, use_qk_l2norm_in_kernel, disable_state_update, disable_output_calculation, intermediate_states_buffer, intermediate_state_indices, cache_steps, retrieve_parent_token, ) return o fused_recurrent_gdn = fused_recurrent_gated_delta_rule