[Rotary] Implement GPT-J style (interleaved) rotary

csrc/rotary/rotary.cpp

@@ -1,3 +1,7 @@
+/******************************************************************************
+ * Copyright (c) 2023, Tri Dao.
+ ******************************************************************************/
+
 #include <torch/extension.h>
 #include <c10/cuda/CUDAGuard.h>
 

csrc/rotary/rotary_cuda.cu

@@ -1,3 +1,7 @@
+/******************************************************************************
+ * Copyright (c) 2023, Tri Dao.
+ ******************************************************************************/
+
 #include <torch/python.h>
 #include <ATen/native/TensorIterator.h>
 #include <ATen/native/cuda/Loops.cuh>

flash_attn/layers/rotary.py

@@ -1,4 +1,4 @@
-# Inspired by https://github.com/facebookresearch/xformers/blob/main/xformers/components/positional_embedding/rotary.py
+# Copyright (c) 2023, Tri Dao.
 
 from typing import Tuple
 import math
@@ -10,31 +10,37 @@ from einops import rearrange, repeat
 import rotary_emb
 
 
-def rotate_half(x):
-    x1, x2 = x.chunk(2, dim=-1)
-    return torch.cat((-x2, x1), dim=-1)
+def rotate_half(x, interleaved=False):
+    if not interleaved:
+        x1, x2 = x.chunk(2, dim=-1)
+        return torch.cat((-x2, x1), dim=-1)
+    else:
+        x1, x2 = x[..., ::2], x[..., 1::2]
+        return rearrange(torch.stack((-x2, x1), dim=-1), '... d two -> ... (d two)', two=2)
 
 
-def apply_rotary_emb_torch(x, cos, sin):
+def apply_rotary_emb_torch(x, cos, sin, interleaved=False):
     """
     x: (batch_size, seqlen, nheads, headdim)
     cos, sin: (seqlen, rotary_dim / 2)
     """
-    rotary_dim = cos.shape[-1] * 2
-    assert rotary_dim <= x.shape[-1]
+    ro_dim = cos.shape[-1] * 2
+    assert ro_dim <= x.shape[-1]
     cos = repeat(cos, 's d -> s 1 (2 d)')
     sin = repeat(sin, 's d -> s 1 (2 d)')
-    return torch.cat([x[..., :rotary_dim] * cos + rotate_half(x[..., :rotary_dim]) * sin,
-                      x[..., rotary_dim:]], dim=-1)
+    return torch.cat([x[..., :ro_dim] * cos + rotate_half(x[..., :ro_dim], interleaved) * sin,
+                      x[..., ro_dim:]], dim=-1)
 
 
 class ApplyRotaryEmb(torch.autograd.Function):
 
     @staticmethod
-    def forward(ctx, x, cos, sin, inplace=False):
+    def forward(ctx, x, cos, sin, interleaved=False, inplace=False):
         """
             x: (batch_size, seqlen, nheads, headdim)
             cos, sin: (seqlen, rotary_dim / 2)
+            interleaved: if True, rotate pairs of even and odd dimensions (GPT-J style) instead
+                of 1st half and 2nd half (GPT-NeoX style).
         rotary_dim must be <= headdim
         Apply rotary embedding to the first rotary_dim of x.
         """
@@ -44,14 +50,21 @@ class ApplyRotaryEmb(torch.autograd.Function):
         assert rotary_dim <= headdim
         assert seqlen <= rotary_seqlen
         assert sin.shape == (rotary_seqlen, rotary_dim // 2)
-        x1, x2 = x[..., :rotary_dim].chunk(2, dim=-1)
+        x_ro = x[..., :rotary_dim]
+        x1, x2 = x_ro.chunk(2, dim=-1) if not interleaved else (x_ro[..., ::2], x_ro[..., 1::2])
         out = torch.empty_like(x) if not inplace else x
-        o1, o2 = out[..., :rotary_dim].chunk(2, dim=-1) if not inplace else (x1, x2)
+        out_ro = out[..., :rotary_dim]
+        if inplace:
+            o1, o2 = x1, x2
+        else:
+            o1, o2 = (out_ro.chunk(2, dim=-1) if not interleaved
+                      else (out_ro[..., ::2], out_ro[..., 1::2]))
         rotary_emb.apply_rotary(x1, x2, rearrange(cos[:seqlen], 's d -> s 1 d'),
                                 rearrange(sin[:seqlen], 's d -> s 1 d'), o1, o2, False)
         if not inplace and rotary_dim < headdim:
             out[..., rotary_dim:].copy_(x[..., rotary_dim:])
         ctx.save_for_backward(cos, sin)
+        ctx.interleaved = interleaved
         ctx.inplace = inplace
         return out if not inplace else x
 
@@ -62,14 +75,21 @@ class ApplyRotaryEmb(torch.autograd.Function):
         rotary_dim = cos.shape[-1]
         rotary_dim *= 2
         inplace = ctx.inplace
-        do1, do2 = do[..., :rotary_dim].chunk(2, dim=-1)
+        do_ro = do[..., :rotary_dim]
+        do1, do2 = (do_ro.chunk(2, dim=-1) if not ctx.interleaved
+                    else (do_ro[..., ::2], do_ro[..., 1::2]))
         dx = torch.empty_like(do) if not inplace else do
-        dx1, dx2 = dx[..., :rotary_dim].chunk(2, dim=-1) if not inplace else (do1, do2)
+        if inplace:
+            dx1, dx2 = do1, do2
+        else:
+            dx_ro = dx[..., :rotary_dim]
+            dx1, dx2 = (dx_ro.chunk(2, dim=-1) if not ctx.interleaved
+                        else (dx_ro[..., ::2], dx_ro[..., 1::2]))
         rotary_emb.apply_rotary(do1, do2, rearrange(cos[:seqlen], 's d -> s 1 d'),
                                 rearrange(sin[:seqlen], 's d -> s 1 d'), dx1, dx2, True)
         if not inplace and rotary_dim < headdim:
             dx[..., rotary_dim:].copy_(do[..., rotary_dim:])
-        return dx, None, None, None
+        return dx, None, None, None, None
 
 
 apply_rotary_emb_func = ApplyRotaryEmb.apply
@@ -78,11 +98,13 @@ apply_rotary_emb_func = ApplyRotaryEmb.apply
 class ApplyRotaryEmbQKV_(torch.autograd.Function):
 
     @staticmethod
-    def forward(ctx, qkv, cos, sin, cos_k=None, sin_k=None):
+    def forward(ctx, qkv, cos, sin, cos_k=None, sin_k=None, interleaved=False):
         """
             qkv: (batch_size, seqlen, 3, nheads, headdim)
             cos, sin: (seqlen, rotary_dim / 2)
             cos_k, sin_k: (seqlen, rotary_dim / 2), optional
+            interleaved: if True, rotate pairs of even and odd dimensions (GPT-J style) instead of
+                1st half and 2nd half (GPT-NeoX style).
         rotary_dim must be <= headdim
         Apply rotary embedding *inplace* to the first rotary_dim of q and k.
         """
@@ -95,13 +117,16 @@ class ApplyRotaryEmbQKV_(torch.autograd.Function):
         cos_k = cos if cos_k is None else cos_k
         sin_k = sin if sin_k is None else sin_k
         assert sin.shape == cos_k.shape == sin_k.shape == (rotary_seqlen, rotary_dim // 2)
-        q1, q2 = qkv[:, :, 0, :, :rotary_dim].chunk(2, dim=-1)
+        q_ro = qkv[:, :, 0, :, :rotary_dim]
+        q1, q2 = q_ro.chunk(2, dim=-1) if not interleaved else (q_ro[..., ::2], q_ro[..., 1::2])
         rotary_emb.apply_rotary(q1, q2, rearrange(cos[:seqlen], 's d -> s 1 d'),
                                 rearrange(sin[:seqlen], 's d -> s 1 d'), q1, q2, False)
-        k1, k2 = qkv[:, :, 1, :, :rotary_dim].chunk(2, dim=-1)
+        k_ro = qkv[:, :, 1, :, :rotary_dim]
+        k1, k2 = k_ro.chunk(2, dim=-1) if not interleaved else (k_ro[..., ::2], k_ro[..., 1::2])
         rotary_emb.apply_rotary(k1, k2, rearrange(cos_k[:seqlen], 's d -> s 1 d'),
                                 rearrange(sin_k[:seqlen], 's d -> s 1 d'), k1, k2, False)
         ctx.save_for_backward(cos, sin, cos_k, sin_k)
+        ctx.interleaved = interleaved
         return qkv
 
     @staticmethod
@@ -110,13 +135,17 @@ class ApplyRotaryEmbQKV_(torch.autograd.Function):
         _, seqlen, _, _, headdim = dqkv.shape
         rotary_dim = cos.shape[-1]
         rotary_dim *= 2
-        dq1, dq2 = dqkv[:, :, 0, :, :rotary_dim].chunk(2, dim=-1)
+        dq_ro = dqkv[:, :, 0, :, :rotary_dim]
+        dq1, dq2 = (dq_ro.chunk(2, dim=-1) if not ctx.interleaved
+                    else (dq_ro[..., ::2], dq_ro[..., 1::2]))
         rotary_emb.apply_rotary(dq1, dq2, rearrange(cos[:seqlen], 's d -> s 1 d'),
                                 rearrange(sin[:seqlen], 's d -> s 1 d'), dq1, dq2, True)
-        dk1, dk2 = dqkv[:, :, 1, :, :rotary_dim].chunk(2, dim=-1)
+        dk_ro = dqkv[:, :, 1, :, :rotary_dim]
+        dk1, dk2 = (dk_ro.chunk(2, dim=-1) if not ctx.interleaved
+                    else (dk_ro[..., ::2], dk_ro[..., 1::2]))
         rotary_emb.apply_rotary(dk1, dk2, rearrange(cos_k[:seqlen], 's d -> s 1 d'),
                                 rearrange(sin_k[:seqlen], 's d -> s 1 d'), dk1, dk2, True)
-        return dqkv, None, None, None, None
+        return dqkv, None, None, None, None, None
 
 
 apply_rotary_emb_qkv_ = ApplyRotaryEmbQKV_.apply
@@ -135,22 +164,25 @@ class RotaryEmbedding(torch.nn.Module):
     .. _repo: https://github.com/ZhuiyiTechnology/roformer
     .. _GPT-NeoX: https://github.com/EleutherAI/gpt-neox
 
-    If scale_base > 0, this implements XPos (Sun et al., https://arxiv.org/abs/2212.10554).
+    If scale_base is not None, this implements XPos (Sun et al., https://arxiv.org/abs/2212.10554).
     A recommended value for scale_base is 512: https://github.com/HazyResearch/flash-attention/issues/96
     Reference: https://github.com/sunyt32/torchscale/blob/main/torchscale/component/xpos_relative_position.py
     """
 
-    def __init__(self, dim: int, base=10000, scale_base=0, device=None):
+    def __init__(self, dim: int, base=10000, interleaved=False, scale_base=None, device=None):
         """
+            interleaved: if True, rotate pairs of even and odd dimensions (GPT-J style) instead
+                of 1st half and 2nd half (GPT-NeoX style).
         """
         super().__init__()
         # Generate and save the inverse frequency buffer (non trainable)
         inv_freq = 1.0 / (base ** (torch.arange(0, dim, 2, device=device,
                                                 dtype=torch.float32) / dim))
         self.register_buffer("inv_freq", inv_freq)
+        self.interleaved = interleaved
         self.scale_base = scale_base
         scale = ((torch.arange(0, dim, 2, device=device, dtype=torch.float32) + 0.4 * dim)
-                 / (1.4 * dim) if scale_base > 0 else None)
+                 / (1.4 * dim) if scale_base is not None else None)
         self.register_buffer("scale", scale)
 
         self._seq_len_cached = 0
@@ -187,16 +219,19 @@ class RotaryEmbedding(torch.nn.Module):
 
     def forward(self, qkv: torch.Tensor, seqlen_offset: int = 0) -> Tuple[torch.Tensor, torch.Tensor]:
         """
+        qkv: (batch, seqlen, 3, nheads, headdim)
         seqlen_offset: can be used in generation where the qkv being passed in is only the last
         token in the batch.
         """
         self._update_cos_sin_cache(qkv, seqlen_offset)
         if self.scale is None:
             return apply_rotary_emb_qkv_(
-                qkv, self._cos_cached[seqlen_offset:], self._sin_cached[seqlen_offset:]
+                qkv, self._cos_cached[seqlen_offset:], self._sin_cached[seqlen_offset:],
+                None, None, self.interleaved
             )
         else:
             return apply_rotary_emb_qkv_(
                 qkv, self._cos_cached[seqlen_offset:], self._sin_cached[seqlen_offset:],
-                self._cos_k_cached[seqlen_offset:], self._sin_k_cached[seqlen_offset:]
+                self._cos_k_cached[seqlen_offset:], self._sin_k_cached[seqlen_offset:],
+                self.interleaved
             )

tests/layers/test_rotary.py

@@ -0,0 +1,114 @@
+# Copyright (c) 2023, Tri Dao.
+
+import math
+
+import torch
+import torch.nn.functional as F
+import pytest
+
+from einops import rearrange
+
+from transformers.models.gpt_neox.modeling_gpt_neox import RotaryEmbedding as RotaryEmbeddingNeoX
+from transformers.models.gpt_neox.modeling_gpt_neox import apply_rotary_pos_emb as apply_rotary_pos_emb_neox
+from transformers.models.gptj.modeling_gptj import fixed_pos_embedding
+from transformers.models.gptj.modeling_gptj import apply_rotary_pos_emb as apply_rotary_pos_emb_gptj
+
+from flash_attn.layers.rotary import apply_rotary_emb_func, apply_rotary_emb_qkv_
+from flash_attn.layers.rotary import RotaryEmbedding
+
+
+# NeoX-style rotary embedding
+@pytest.mark.parametrize('seqlen_offset', [0, 711])
+@pytest.mark.parametrize('rotary_emb_fraction', [0.5, 1.0])
+def test_rotary(rotary_emb_fraction, seqlen_offset):
+    device = 'cuda'
+    dtype = torch.float16
+    rtol, atol = (1e-3, 5e-3)
+    # set seed
+    torch.random.manual_seed(0)
+    batch_size = 8
+    seqlen_total = 2048
+    seqlen = seqlen_total - seqlen_offset
+    nheads = 16
+    headdim = 128
+    rotary_dim = int(headdim * rotary_emb_fraction)
+    qkv = torch.randn(batch_size, seqlen, 3, nheads, headdim, device=device, dtype=dtype,
+                      requires_grad=True)
+    qkv_og = qkv.clone().detach()  # Our implementation modifies qkv inplace
+    rotary = RotaryEmbedding(rotary_dim, device=device)
+    rotary_neox = RotaryEmbeddingNeoX(rotary_dim, seqlen_total, device=device)
+    # Doesn't matter what tensor we pass in, rotary_neox only uses the device of the tensor
+    cos_neox, sin_neox = rotary_neox(qkv, seq_len=seqlen_total)
+    cos_neox, sin_neox = cos_neox.to(dtype=dtype), sin_neox.to(dtype=dtype)
+    q_pt = rearrange(qkv[:, :, 0, :, :rotary_dim],
+                     'b s h d -> b h s d').detach().clone().requires_grad_(True)
+    k_pt = rearrange(qkv[:, :, 1, :, :rotary_dim],
+                     'b s h d -> b h s d').detach().clone().requires_grad_(True)
+    q_neox, k_neox = apply_rotary_pos_emb_neox(q_pt, k_pt, cos_neox, sin_neox, offset=seqlen_offset)
+    out = rotary(qkv, seqlen_offset=seqlen_offset)
+    assert torch.allclose(rotary._cos_cached, cos_neox[..., :rotary_dim // 2].to(dtype=dtype),
+                          rtol=rtol, atol=atol)
+    assert torch.allclose(rotary._sin_cached, sin_neox[..., :rotary_dim // 2].to(dtype=dtype),
+                          rtol=rtol, atol=atol)
+    assert torch.allclose(rearrange(q_neox, 'b h s d -> b s h d'), out[:, :, 0, :, :rotary_dim],
+                          rtol=rtol, atol=atol)
+    assert torch.allclose(rearrange(k_neox, 'b h s d -> b s h d'), out[:, :, 1, :, :rotary_dim],
+                          rtol=rtol, atol=atol)
+    assert torch.equal(out[:, :, 0:2, :, rotary_dim:], qkv_og[:, :, 0:2, :, rotary_dim:])
+    assert torch.equal(out[:, :, 2], qkv_og[:, :, 2])
+
+    g = torch.randn_like(out)
+    g_og = g.clone().detach()  # Our implementation modifies g inplace
+    out.backward(g)
+    q_neox.backward(rearrange(g_og[:, :, 0, :, :rotary_dim], 'b s h d -> b h s d'))
+    k_neox.backward(rearrange(g_og[:, :, 1, :, :rotary_dim], 'b s h d -> b h s d'))
+    assert torch.allclose(rearrange(q_pt.grad, 'b h s d -> b s h d'),
+                          qkv.grad[:, :, 0, :, :rotary_dim], rtol=rtol, atol=atol)
+    assert torch.allclose(rearrange(k_pt.grad, 'b h s d -> b s h d'),
+                          qkv.grad[:, :, 1, :, :rotary_dim], rtol=rtol, atol=atol)
+    assert torch.equal(qkv.grad[:, :, 0:2, :, rotary_dim:], g_og[:, :, 0:2, :, rotary_dim:])
+    assert torch.equal(qkv.grad[:, :, 2], g_og[:, :, 2])
+
+
+# GPT-J-style rotary embedding
+@pytest.mark.parametrize('seqlen_offset', [0, 711])
+@pytest.mark.parametrize('rotary_emb_fraction', [0.5, 1.0])
+def test_rotary_interleaved(rotary_emb_fraction, seqlen_offset):
+    device = 'cuda'
+    dtype = torch.float16
+    rtol, atol = (1e-3, 5e-3)
+    # set seed
+    torch.random.manual_seed(0)
+    batch_size = 8
+    seqlen_total = 2048
+    seqlen = seqlen_total - seqlen_offset
+    nheads = 16
+    headdim = 128
+    rotary_dim = int(headdim * rotary_emb_fraction)
+    qkv = torch.randn(batch_size, seqlen, 3, nheads, headdim, device=device, dtype=dtype,
+                      requires_grad=True)
+    qkv_og = qkv.clone().detach()  # Our implementation modifies qkv inplace
+    rotary = RotaryEmbedding(rotary_dim, interleaved=True, device=device)
+    sincos_gptj = fixed_pos_embedding(qkv[..., :rotary_dim], seq_dim=1, seq_len=seqlen_total)
+    sincos_gptj = tuple(x.to(dtype=dtype) for x in sincos_gptj)
+    q_pt = qkv[:, :, 0, :, :rotary_dim].detach().clone().requires_grad_(True)
+    k_pt = qkv[:, :, 1, :, :rotary_dim].detach().clone().requires_grad_(True)
+    q_gptj = apply_rotary_pos_emb_gptj(q_pt, sincos_gptj, offset=seqlen_offset)
+    k_gptj = apply_rotary_pos_emb_gptj(k_pt, sincos_gptj, offset=seqlen_offset)
+    out = rotary(qkv, seqlen_offset=seqlen_offset)
+    assert torch.allclose(rotary._cos_cached, sincos_gptj[1], rtol=rtol, atol=atol)
+    assert torch.allclose(rotary._sin_cached, sincos_gptj[0], rtol=rtol, atol=atol)
+    assert torch.allclose(q_gptj, out[:, :, 0, :, :rotary_dim], rtol=rtol, atol=atol)
+    assert torch.allclose(k_gptj, out[:, :, 1, :, :rotary_dim], rtol=rtol, atol=atol)
+    assert torch.equal(out[:, :, 0:2, :, rotary_dim:], qkv_og[:, :, 0:2, :, rotary_dim:])
+    assert torch.equal(out[:, :, 2], qkv_og[:, :, 2])
+
+    g = torch.randn_like(out)
+    g_og = g.clone().detach()  # Our implementation modifies g inplace
+    out.backward(g)
+    q_gptj.backward(g_og[:, :, 0, :, :rotary_dim])
+    k_gptj.backward(g_og[:, :, 1, :, :rotary_dim])
+    assert torch.allclose(q_pt.grad, qkv.grad[:, :, 0, :, :rotary_dim], rtol=rtol, atol=atol)
+    assert torch.allclose(k_pt.grad, qkv.grad[:, :, 1, :, :rotary_dim], rtol=rtol, atol=atol)
+    assert torch.equal(qkv.grad[:, :, 0:2, :, rotary_dim:], g_og[:, :, 0:2, :, rotary_dim:])
+    assert torch.equal(qkv.grad[:, :, 2], g_og[:, :, 2])