Merge branch 'main' of https://github.com/DeepLink-org/DeepLinkExt in…

…to wx/support_flash_attention_for_ascend

deeplink_ext/internlm_ops/rotary/__init__.py

@@ -1,11 +1,14 @@
 # Copyright (c) 2024, DeepLink.
 
 try:
-    from .deeplink import DeepLinkApplyRotaryEmbQKV_
+    from .deeplink import DeepLinkApplyRotaryEmbQKV_, DeepLinkApplyRotaryEmb
 except:
     print(
         "[deeplink_ext] rotary is not implemented in diopi. Falling back to the slower implementation.\n",
         end="",
     )
-    from .fallback import ApplyRotaryEmbQKV_ as DeepLinkApplyRotaryEmbQKV_
+    from .fallback import (
+        ApplyRotaryEmbQKV_ as DeepLinkApplyRotaryEmbQKV_,
+        ApplyRotaryEmb as DeepLinkApplyRotaryEmb,
+    )
 from . import fallback

deeplink_ext/internlm_ops/rotary/deeplink.py

@@ -69,3 +69,63 @@ def backward(ctx, dqkv):
             interleaved,
         )
         return dqkv, None, None, None, None, None
+
+
+class DeepLinkApplyRotaryEmb(torch.autograd.Function):
+    @staticmethod
+    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.
+        """
+        batch, seqlen, nheads, headdim = x.shape
+        rotary_seqlen, rotary_dim = cos.shape
+        rotary_dim *= 2
+        assert rotary_dim <= headdim
+        assert seqlen <= rotary_seqlen
+        assert sin.shape == (rotary_seqlen, rotary_dim // 2)
+        x_ro = x[..., :rotary_dim]
+        out = torch.empty_like(x) if not inplace else x
+        out_ro = out[..., :rotary_dim]
+
+        ext.apply_rotary(
+            out_ro,
+            x_ro,
+            rearrange(cos[:seqlen], "s d -> s 1 d"),
+            rearrange(sin[:seqlen], "s d -> s 1 d"),
+            False,
+            interleaved,
+        )
+
+        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
+
+    @staticmethod
+    def backward(ctx, do):
+        cos, sin = ctx.saved_tensors
+        _, seqlen, _, headdim = do.shape
+        rotary_dim = cos.shape[-1]
+        rotary_dim *= 2
+        inplace = ctx.inplace
+        do_ro = do[..., :rotary_dim]
+        dx = torch.empty_like(do) if not inplace else do
+        dx_ro = dx[..., :rotary_dim]
+        ext.apply_rotary(
+            dx_ro,
+            do_ro,
+            rearrange(cos[:seqlen], "s d -> s 1 d"),
+            rearrange(sin[:seqlen], "s d -> s 1 d"),
+            True,
+            ctx.interleaved,
+        )
+        if not inplace and rotary_dim < headdim:
+            dx[..., rotary_dim:].copy_(do[..., rotary_dim:])
+        return dx, None, None, None, None

deeplink_ext/internlm_ops/rotary/fallback/__init__.py

@@ -1,3 +1,3 @@
 # Copyright (c) 2024, DeepLink.
 
-from .fallback import ApplyRotaryEmbQKV_
+from .fallback import ApplyRotaryEmbQKV_, ApplyRotaryEmb

deeplink_ext/internlm_ops/rotary/fallback/fallback.py

@@ -2,7 +2,6 @@
 
 import torch
 from einops import rearrange
-import deeplink_ext.cpp_extensions as ext
 
 
 # Rotary_emb
@@ -114,3 +113,9 @@ def backward(ctx, dqkv):
         )
         dqkv[:, :, 1, :, :rotary_dim] = torch.cat((dk1, dk2), dim=-1)
         return dqkv, None, None, None, None, None
+
+
+class ApplyRotaryEmb:
+    @staticmethod
+    def apply(*args, **kwargs):
+        raise NotImplementedError("fallback.ApplyRotaryEmb is not implemented yet")

deeplink_ext/patch_internlm.py

@@ -48,11 +48,6 @@ def CrossEntropyLossProxy(reduction, **_):
     def _patch_ops():
         import internlm.model.embedding  # type: ignore
 
-        def NotImplementedLegacyRotaryEmb(*args, **kwargs):
-            raise NotImplementedError(
-                "we assume that legacy_apply_rotary_embed is not used in internlm"
-            )
-
         class NonLegacyRotaryEmbQKV_(torch.autograd.Function):
             """the first 2 dims of qkv has been squeezed"""
 
@@ -74,7 +69,7 @@ def backward(ctx, dqkv: torch.Tensor, *args, **kwargs):
 
         internlm.model.embedding.apply_rotary_emb_qkv_ = NonLegacyRotaryEmbQKV_.apply
         internlm.model.embedding.legacy_apply_rotary_embed = (
-            NotImplementedLegacyRotaryEmb
+            ext.rotary.DeepLinkApplyRotaryEmb.apply
         )
         internlm.model.embedding.legacy_apply_rotary_embed_qkv = (
             ext.rotary.DeepLinkApplyRotaryEmbQKV_.apply