Fix generation with large sequences (#2561)

* fixed a bug in generation when the sequence is larger than 2048 tokens * fixed the dynamic resizing of rotary embeddings
OpenNMT · Feb 22, 2024 · 0e72326 · 0e72326
1 parent 5deb20e
commit 0e72326
Showing 1 changed file with 31 additions and 30 deletions.
diff --git a/onmt/modules/multi_headed_attn.py b/onmt/modules/multi_headed_attn.py
@@ -18,14 +18,18 @@
 # are both < 2048 tokens.
 
 
-def rotaryembeddings(dim: int, maxseqlen=2048, base=10000):
+def rotaryembeddings(dim: int, maxseqlen=2048, base=10000, device=None):
     inv_freq = 1.0 / (base ** (torch.arange(0, dim, 2).float() / dim))
     tmax = torch.arange(maxseqlen, device=inv_freq.device)
     rope = torch.outer(tmax, inv_freq).float()
     # rope is now matrix [maxseqlen, dim/2]
     rope = torch.polar(torch.ones_like(rope), rope)
     rope = torch.cat((rope, rope), dim=1)
-    return rope
+    if device is not None:
+        rope = rope.to(device)
+    cos = rope[:, : rope.size(1) // 2].real.contiguous().half()
+    sin = rope[:, : rope.size(1) // 2].imag.contiguous().half()
+    return rope, cos, sin
 
 
 def rotate_half(x):
@@ -369,12 +373,8 @@ def __init__(
                     self.rotary_dim = self.dim_per_head
                 else:
                     self.rotary_dim = rotary_dim
-                self.rope = rotaryembeddings(self.rotary_dim, base=rotary_theta)
-                self.cos = (
-                    self.rope[:, : self.rope.size(1) // 2].real.contiguous().half()
-                )
-                self.sin = (
-                    self.rope[:, : self.rope.size(1) // 2].imag.contiguous().half()
+                self.rope, self.cos, self.sin = rotaryembeddings(
+                    self.rotary_dim, base=rotary_theta
                 )
                 self.rotary_interleave = rotary_interleave
                 self.rotary_theta = rotary_theta
@@ -465,11 +465,13 @@ def forward(
                 ):
                     if self.max_relative_positions == -1:  # Rotary Embeddings
                         if seqlen > self.rope.size(0):
-                            self.rope = rotaryembeddings(
+
+                            self.rope, _, _ = rotaryembeddings(
                                 self.rotary_dim,
                                 maxseqlen=(seqlen + 2048),
                                 base=self.rotary_theta,
-                            ).to(self.rope.device)
+                                device=self.rope.device,
+                            )
                         rope = self.rope[start_pos : start_pos + seqlen]
                         query, key = apply_rotary_emb(
                             query, key, rope, interleave=self.rotary_interleave
@@ -486,23 +488,6 @@ def forward(
                     self.layer_cache[1]["values"] = value
 
                 else:
-                    if self.max_relative_positions == -1:  # Rotary Embeddings
-                        if seqlen > self.rope.size(0):
-                            self.rope = rotaryembeddings(
-                                self.rotary_dim,
-                                maxseqlen=(seqlen + 2048),
-                                base=self.rotary_theta,
-                            ).to(self.rope.device)
-                            self.cos = (
-                                self.rope[:, : self.rope.size(1) // 2]
-                                .real.contiguous()
-                                .half()
-                            )
-                            self.sin = (
-                                self.rope[:, : self.rope.size(1) // 2]
-                                .imag.contiguous()
-                                .half()
-                            )
                     if start_pos >= self.layer_cache[1]["keys"].size(2):
                         self.layer_cache[1]["keys"] = torch.cat(
                             [
@@ -528,6 +513,20 @@ def forward(
                             ],
                             dim=-2,
                         )
+                        if (
+                            self.max_relative_positions == -1
+                            and start_pos + 32 >= self.rope.size(0)
+                        ):
+                            # Resize rotary embeddings.
+                            # We take a margin of 32 tokens as the kv_cache
+                            #   is incremented by 32 tokens every 32 tokens.
+                            self.rope, self.cos, self.sin = rotaryembeddings(
+                                self.rotary_dim,
+                                maxseqlen=(start_pos + 2048),
+                                base=self.rotary_theta,
+                                device=self.rope.device,
+                            )
+
                     if sliding_window > 0 and key.size(2) > sliding_window:
                         self.layer_cache[1]["keys"] = self.layer_cache[1]["keys"][
                             :, :, 1:, :
@@ -593,12 +592,14 @@ def forward(
                 start_pos = 0
                 seqlen = query.size(2)
                 if seqlen > self.rope.size(0):
-                    self.rope = rotaryembeddings(
+                    # Resize rotary embeddings.
+                    self.rope, self.cos, self.sin = rotaryembeddings(
                         self.rotary_dim,
                         maxseqlen=(seqlen + 2048),
                         base=self.rotary_theta,
-                    ).to(self.rope.device)
-                rope = self.rope[start_pos : start_pos + seqlen].to(query.device)
+                        device=query.device,
+                    )
+                rope = self.rope[start_pos : start_pos + seqlen]
                 query, key = apply_rotary_emb(
                     query, key, rope, interleave=self.rotary_interleave
                 )