Run make fix-copies.

Author: abdullahselek

src/transformers/models/aria/modeling_aria.py

@@ -460,6 +460,12 @@ def apply_rotary_pos_emb(q, k, cos, sin, position_ids=None, unsqueeze_dim=1):
     """
     cos = cos.unsqueeze(unsqueeze_dim)
     sin = sin.unsqueeze(unsqueeze_dim)
+
+    # Adjust the rotary embedding dimensions if they don't match q's last dimension.
+    if cos.shape[-1] != q.shape[-1]:
+        cos = cos[..., : q.shape[-1]]
+        sin = sin[..., : q.shape[-1]]
+
     q_embed = (q * cos) + (rotate_half(q) * sin)
     k_embed = (k * cos) + (rotate_half(k) * sin)
     return q_embed, k_embed

src/transformers/models/chameleon/modeling_chameleon.py

@@ -176,6 +176,12 @@ def apply_rotary_pos_emb(q, k, cos, sin, position_ids=None, unsqueeze_dim=1):
     """
     cos = cos.unsqueeze(unsqueeze_dim)
     sin = sin.unsqueeze(unsqueeze_dim)
+
+    # Adjust the rotary embedding dimensions if they don't match q's last dimension.
+    if cos.shape[-1] != q.shape[-1]:
+        cos = cos[..., : q.shape[-1]]
+        sin = sin[..., : q.shape[-1]]
+
     q_embed = (q * cos) + (rotate_half(q) * sin)
     k_embed = (k * cos) + (rotate_half(k) * sin)
     return q_embed, k_embed

src/transformers/models/dbrx/modeling_dbrx.py

@@ -107,6 +107,12 @@ def apply_rotary_pos_emb(q, k, cos, sin, position_ids=None, unsqueeze_dim=1):
     """
     cos = cos.unsqueeze(unsqueeze_dim)
     sin = sin.unsqueeze(unsqueeze_dim)
+
+    # Adjust the rotary embedding dimensions if they don't match q's last dimension.
+    if cos.shape[-1] != q.shape[-1]:
+        cos = cos[..., : q.shape[-1]]
+        sin = sin[..., : q.shape[-1]]
+
     q_embed = (q * cos) + (rotate_half(q) * sin)
     k_embed = (k * cos) + (rotate_half(k) * sin)
     return q_embed, k_embed

src/transformers/models/diffllama/modeling_diffllama.py

@@ -106,6 +106,12 @@ def apply_rotary_pos_emb(q, k, cos, sin, position_ids=None, unsqueeze_dim=1):
     """
     cos = cos.unsqueeze(unsqueeze_dim)
     sin = sin.unsqueeze(unsqueeze_dim)
+
+    # Adjust the rotary embedding dimensions if they don't match q's last dimension.
+    if cos.shape[-1] != q.shape[-1]:
+        cos = cos[..., : q.shape[-1]]
+        sin = sin[..., : q.shape[-1]]
+
     q_embed = (q * cos) + (rotate_half(q) * sin)
     k_embed = (k * cos) + (rotate_half(k) * sin)
     return q_embed, k_embed

src/transformers/models/emu3/modeling_emu3.py

@@ -119,6 +119,12 @@ def apply_rotary_pos_emb(q, k, cos, sin, position_ids=None, unsqueeze_dim=1):
     """
     cos = cos.unsqueeze(unsqueeze_dim)
     sin = sin.unsqueeze(unsqueeze_dim)
+
+    # Adjust the rotary embedding dimensions if they don't match q's last dimension.
+    if cos.shape[-1] != q.shape[-1]:
+        cos = cos[..., : q.shape[-1]]
+        sin = sin[..., : q.shape[-1]]
+
     q_embed = (q * cos) + (rotate_half(q) * sin)
     k_embed = (k * cos) + (rotate_half(k) * sin)
     return q_embed, k_embed

src/transformers/models/falcon/modeling_falcon.py

@@ -104,6 +104,12 @@ def apply_rotary_pos_emb(q, k, cos, sin, position_ids=None, unsqueeze_dim=1):
     """
     cos = cos.unsqueeze(unsqueeze_dim)
     sin = sin.unsqueeze(unsqueeze_dim)
+
+    # Adjust the rotary embedding dimensions if they don't match q's last dimension.
+    if cos.shape[-1] != q.shape[-1]:
+        cos = cos[..., : q.shape[-1]]
+        sin = sin[..., : q.shape[-1]]
+
     q_embed = (q * cos) + (rotate_half(q) * sin)
     k_embed = (k * cos) + (rotate_half(k) * sin)
     return q_embed, k_embed

src/transformers/models/gemma/modeling_gemma.py

@@ -182,6 +182,12 @@ def apply_rotary_pos_emb(q, k, cos, sin, position_ids=None, unsqueeze_dim=1):
     """
     cos = cos.unsqueeze(unsqueeze_dim)
     sin = sin.unsqueeze(unsqueeze_dim)
+
+    # Adjust the rotary embedding dimensions if they don't match q's last dimension.
+    if cos.shape[-1] != q.shape[-1]:
+        cos = cos[..., : q.shape[-1]]
+        sin = sin[..., : q.shape[-1]]
+
     q_embed = (q * cos) + (rotate_half(q) * sin)
     k_embed = (k * cos) + (rotate_half(k) * sin)
     return q_embed, k_embed

src/transformers/models/gemma2/modeling_gemma2.py

@@ -121,6 +121,12 @@ def apply_rotary_pos_emb(q, k, cos, sin, position_ids=None, unsqueeze_dim=1):
     """
     cos = cos.unsqueeze(unsqueeze_dim)
     sin = sin.unsqueeze(unsqueeze_dim)
+
+    # Adjust the rotary embedding dimensions if they don't match q's last dimension.
+    if cos.shape[-1] != q.shape[-1]:
+        cos = cos[..., : q.shape[-1]]
+        sin = sin[..., : q.shape[-1]]
+
     q_embed = (q * cos) + (rotate_half(q) * sin)
     k_embed = (k * cos) + (rotate_half(k) * sin)
     return q_embed, k_embed

src/transformers/models/gpt_neox_japanese/modeling_gpt_neox_japanese.py

@@ -314,6 +314,12 @@ def apply_rotary_pos_emb(q, k, cos, sin, position_ids=None, unsqueeze_dim=1):
     """
     cos = cos.unsqueeze(unsqueeze_dim)
     sin = sin.unsqueeze(unsqueeze_dim)
+
+    # Adjust the rotary embedding dimensions if they don't match q's last dimension.
+    if cos.shape[-1] != q.shape[-1]:
+        cos = cos[..., : q.shape[-1]]
+        sin = sin[..., : q.shape[-1]]
+
     q_embed = (q * cos) + (rotate_half(q) * sin)
     k_embed = (k * cos) + (rotate_half(k) * sin)
     return q_embed, k_embed

src/transformers/models/granite/modeling_granite.py

@@ -77,6 +77,12 @@ def apply_rotary_pos_emb(q, k, cos, sin, position_ids=None, unsqueeze_dim=1):
     """
     cos = cos.unsqueeze(unsqueeze_dim)
     sin = sin.unsqueeze(unsqueeze_dim)
+
+    # Adjust the rotary embedding dimensions if they don't match q's last dimension.
+    if cos.shape[-1] != q.shape[-1]:
+        cos = cos[..., : q.shape[-1]]
+        sin = sin[..., : q.shape[-1]]
+
     q_embed = (q * cos) + (rotate_half(q) * sin)
     k_embed = (k * cos) + (rotate_half(k) * sin)
     return q_embed, k_embed

src/transformers/models/jetmoe/modeling_jetmoe.py

@@ -479,6 +479,12 @@ def apply_rotary_pos_emb(q, k, cos, sin, position_ids=None, unsqueeze_dim=1):
     """
     cos = cos.unsqueeze(unsqueeze_dim)
     sin = sin.unsqueeze(unsqueeze_dim)
+
+    # Adjust the rotary embedding dimensions if they don't match q's last dimension.
+    if cos.shape[-1] != q.shape[-1]:
+        cos = cos[..., : q.shape[-1]]
+        sin = sin[..., : q.shape[-1]]
+
     q_embed = (q * cos) + (rotate_half(q) * sin)
     k_embed = (k * cos) + (rotate_half(k) * sin)
     return q_embed, k_embed

src/transformers/models/mimi/modeling_mimi.py

@@ -456,6 +456,12 @@ def apply_rotary_pos_emb(q, k, cos, sin, position_ids=None, unsqueeze_dim=1):
     """
     cos = cos.unsqueeze(unsqueeze_dim)
     sin = sin.unsqueeze(unsqueeze_dim)
+
+    # Adjust the rotary embedding dimensions if they don't match q's last dimension.
+    if cos.shape[-1] != q.shape[-1]:
+        cos = cos[..., : q.shape[-1]]
+        sin = sin[..., : q.shape[-1]]
+
     q_embed = (q * cos) + (rotate_half(q) * sin)
     k_embed = (k * cos) + (rotate_half(k) * sin)
     return q_embed, k_embed

src/transformers/models/mistral/modeling_mistral.py

@@ -87,6 +87,12 @@ def apply_rotary_pos_emb(q, k, cos, sin, position_ids=None, unsqueeze_dim=1):
     """
     cos = cos.unsqueeze(unsqueeze_dim)
     sin = sin.unsqueeze(unsqueeze_dim)
+
+    # Adjust the rotary embedding dimensions if they don't match q's last dimension.
+    if cos.shape[-1] != q.shape[-1]:
+        cos = cos[..., : q.shape[-1]]
+        sin = sin[..., : q.shape[-1]]
+
     q_embed = (q * cos) + (rotate_half(q) * sin)
     k_embed = (k * cos) + (rotate_half(k) * sin)
     return q_embed, k_embed

src/transformers/models/mixtral/modeling_mixtral.py

@@ -200,6 +200,12 @@ def apply_rotary_pos_emb(q, k, cos, sin, position_ids=None, unsqueeze_dim=1):
     """
     cos = cos.unsqueeze(unsqueeze_dim)
     sin = sin.unsqueeze(unsqueeze_dim)
+
+    # Adjust the rotary embedding dimensions if they don't match q's last dimension.
+    if cos.shape[-1] != q.shape[-1]:
+        cos = cos[..., : q.shape[-1]]
+        sin = sin[..., : q.shape[-1]]
+
     q_embed = (q * cos) + (rotate_half(q) * sin)
     k_embed = (k * cos) + (rotate_half(k) * sin)
     return q_embed, k_embed

src/transformers/models/mllama/modeling_mllama.py

@@ -641,6 +641,12 @@ def apply_rotary_pos_emb(q, k, cos, sin, position_ids=None, unsqueeze_dim=1):
     """
     cos = cos.unsqueeze(unsqueeze_dim)
     sin = sin.unsqueeze(unsqueeze_dim)
+
+    # Adjust the rotary embedding dimensions if they don't match q's last dimension.
+    if cos.shape[-1] != q.shape[-1]:
+        cos = cos[..., : q.shape[-1]]
+        sin = sin[..., : q.shape[-1]]
+
     q_embed = (q * cos) + (rotate_half(q) * sin)
     k_embed = (k * cos) + (rotate_half(k) * sin)
     return q_embed, k_embed

src/transformers/models/modernbert/modeling_modernbert.py

@@ -330,6 +330,12 @@ def apply_rotary_pos_emb(q, k, cos, sin, position_ids=None, unsqueeze_dim=1):
     """
     cos = cos.unsqueeze(unsqueeze_dim)
     sin = sin.unsqueeze(unsqueeze_dim)
+
+    # Adjust the rotary embedding dimensions if they don't match q's last dimension.
+    if cos.shape[-1] != q.shape[-1]:
+        cos = cos[..., : q.shape[-1]]
+        sin = sin[..., : q.shape[-1]]
+
     q_embed = (q * cos) + (rotate_half(q) * sin)
     k_embed = (k * cos) + (rotate_half(k) * sin)
     return q_embed, k_embed

src/transformers/models/moshi/modeling_moshi.py

@@ -400,6 +400,12 @@ def apply_rotary_pos_emb(q, k, cos, sin, position_ids=None, unsqueeze_dim=1):
     """
     cos = cos.unsqueeze(unsqueeze_dim)
     sin = sin.unsqueeze(unsqueeze_dim)
+
+    # Adjust the rotary embedding dimensions if they don't match q's last dimension.
+    if cos.shape[-1] != q.shape[-1]:
+        cos = cos[..., : q.shape[-1]]
+        sin = sin[..., : q.shape[-1]]
+
     q_embed = (q * cos) + (rotate_half(q) * sin)
     k_embed = (k * cos) + (rotate_half(k) * sin)
     return q_embed, k_embed

src/transformers/models/olmo/modeling_olmo.py

@@ -93,6 +93,12 @@ def apply_rotary_pos_emb(q, k, cos, sin, position_ids=None, unsqueeze_dim=1):
     """
     cos = cos.unsqueeze(unsqueeze_dim)
     sin = sin.unsqueeze(unsqueeze_dim)
+
+    # Adjust the rotary embedding dimensions if they don't match q's last dimension.
+    if cos.shape[-1] != q.shape[-1]:
+        cos = cos[..., : q.shape[-1]]
+        sin = sin[..., : q.shape[-1]]
+
     q_embed = (q * cos) + (rotate_half(q) * sin)
     k_embed = (k * cos) + (rotate_half(k) * sin)
     return q_embed, k_embed

src/transformers/models/olmo2/modeling_olmo2.py

@@ -82,6 +82,12 @@ def apply_rotary_pos_emb(q, k, cos, sin, position_ids=None, unsqueeze_dim=1):
     """
     cos = cos.unsqueeze(unsqueeze_dim)
     sin = sin.unsqueeze(unsqueeze_dim)
+
+    # Adjust the rotary embedding dimensions if they don't match q's last dimension.
+    if cos.shape[-1] != q.shape[-1]:
+        cos = cos[..., : q.shape[-1]]
+        sin = sin[..., : q.shape[-1]]
+
     q_embed = (q * cos) + (rotate_half(q) * sin)
     k_embed = (k * cos) + (rotate_half(k) * sin)
     return q_embed, k_embed

src/transformers/models/olmoe/modeling_olmoe.py

@@ -250,6 +250,12 @@ def apply_rotary_pos_emb(q, k, cos, sin, position_ids=None, unsqueeze_dim=1):
     """
     cos = cos.unsqueeze(unsqueeze_dim)
     sin = sin.unsqueeze(unsqueeze_dim)
+
+    # Adjust the rotary embedding dimensions if they don't match q's last dimension.
+    if cos.shape[-1] != q.shape[-1]:
+        cos = cos[..., : q.shape[-1]]
+        sin = sin[..., : q.shape[-1]]
+
     q_embed = (q * cos) + (rotate_half(q) * sin)
     k_embed = (k * cos) + (rotate_half(k) * sin)
     return q_embed, k_embed

src/transformers/models/persimmon/modeling_persimmon.py

@@ -148,6 +148,12 @@ def apply_rotary_pos_emb(q, k, cos, sin, position_ids=None, unsqueeze_dim=1):
     """
     cos = cos.unsqueeze(unsqueeze_dim)
     sin = sin.unsqueeze(unsqueeze_dim)
+
+    # Adjust the rotary embedding dimensions if they don't match q's last dimension.
+    if cos.shape[-1] != q.shape[-1]:
+        cos = cos[..., : q.shape[-1]]
+        sin = sin[..., : q.shape[-1]]
+
     q_embed = (q * cos) + (rotate_half(q) * sin)
     k_embed = (k * cos) + (rotate_half(k) * sin)
     return q_embed, k_embed

src/transformers/models/phi/modeling_phi.py

@@ -70,6 +70,12 @@ def apply_rotary_pos_emb(q, k, cos, sin, position_ids=None, unsqueeze_dim=1):
     """
     cos = cos.unsqueeze(unsqueeze_dim)
     sin = sin.unsqueeze(unsqueeze_dim)
+
+    # Adjust the rotary embedding dimensions if they don't match q's last dimension.
+    if cos.shape[-1] != q.shape[-1]:
+        cos = cos[..., : q.shape[-1]]
+        sin = sin[..., : q.shape[-1]]
+
     q_embed = (q * cos) + (rotate_half(q) * sin)
     k_embed = (k * cos) + (rotate_half(k) * sin)
     return q_embed, k_embed

src/transformers/models/phi3/modeling_phi3.py

@@ -104,6 +104,12 @@ def apply_rotary_pos_emb(q, k, cos, sin, position_ids=None, unsqueeze_dim=1):
     """
     cos = cos.unsqueeze(unsqueeze_dim)
     sin = sin.unsqueeze(unsqueeze_dim)
+
+    # Adjust the rotary embedding dimensions if they don't match q's last dimension.
+    if cos.shape[-1] != q.shape[-1]:
+        cos = cos[..., : q.shape[-1]]
+        sin = sin[..., : q.shape[-1]]
+
     q_embed = (q * cos) + (rotate_half(q) * sin)
     k_embed = (k * cos) + (rotate_half(k) * sin)
     return q_embed, k_embed

src/transformers/models/qwen2/modeling_qwen2.py

@@ -87,6 +87,12 @@ def apply_rotary_pos_emb(q, k, cos, sin, position_ids=None, unsqueeze_dim=1):
     """
     cos = cos.unsqueeze(unsqueeze_dim)
     sin = sin.unsqueeze(unsqueeze_dim)
+
+    # Adjust the rotary embedding dimensions if they don't match q's last dimension.
+    if cos.shape[-1] != q.shape[-1]:
+        cos = cos[..., : q.shape[-1]]
+        sin = sin[..., : q.shape[-1]]
+
     q_embed = (q * cos) + (rotate_half(q) * sin)
     k_embed = (k * cos) + (rotate_half(k) * sin)
     return q_embed, k_embed

src/transformers/models/qwen2_moe/modeling_qwen2_moe.py

@@ -259,6 +259,12 @@ def apply_rotary_pos_emb(q, k, cos, sin, position_ids=None, unsqueeze_dim=1):
     """
     cos = cos.unsqueeze(unsqueeze_dim)
     sin = sin.unsqueeze(unsqueeze_dim)
+
+    # Adjust the rotary embedding dimensions if they don't match q's last dimension.
+    if cos.shape[-1] != q.shape[-1]:
+        cos = cos[..., : q.shape[-1]]
+        sin = sin[..., : q.shape[-1]]
+
     q_embed = (q * cos) + (rotate_half(q) * sin)
     k_embed = (k * cos) + (rotate_half(k) * sin)
     return q_embed, k_embed

src/transformers/models/recurrent_gemma/modeling_recurrent_gemma.py

@@ -124,6 +124,12 @@ def apply_rotary_pos_emb(q, k, cos, sin, position_ids=None, unsqueeze_dim=1):
     """
     cos = cos.unsqueeze(unsqueeze_dim)
     sin = sin.unsqueeze(unsqueeze_dim)
+
+    # Adjust the rotary embedding dimensions if they don't match q's last dimension.
+    if cos.shape[-1] != q.shape[-1]:
+        cos = cos[..., : q.shape[-1]]
+        sin = sin[..., : q.shape[-1]]
+
     q_embed = (q * cos) + (rotate_half(q) * sin)
     k_embed = (k * cos) + (rotate_half(k) * sin)
     return q_embed, k_embed

src/transformers/models/stablelm/modeling_stablelm.py

@@ -154,6 +154,12 @@ def apply_rotary_pos_emb(q, k, cos, sin, position_ids=None, unsqueeze_dim=1):
     """
     cos = cos.unsqueeze(unsqueeze_dim)
     sin = sin.unsqueeze(unsqueeze_dim)
+
+    # Adjust the rotary embedding dimensions if they don't match q's last dimension.
+    if cos.shape[-1] != q.shape[-1]:
+        cos = cos[..., : q.shape[-1]]
+        sin = sin[..., : q.shape[-1]]
+
     q_embed = (q * cos) + (rotate_half(q) * sin)
     k_embed = (k * cos) + (rotate_half(k) * sin)
     return q_embed, k_embed

src/transformers/models/starcoder2/modeling_starcoder2.py

@@ -107,6 +107,12 @@ def apply_rotary_pos_emb(q, k, cos, sin, position_ids=None, unsqueeze_dim=1):
     """
     cos = cos.unsqueeze(unsqueeze_dim)
     sin = sin.unsqueeze(unsqueeze_dim)
+
+    # Adjust the rotary embedding dimensions if they don't match q's last dimension.
+    if cos.shape[-1] != q.shape[-1]:
+        cos = cos[..., : q.shape[-1]]
+        sin = sin[..., : q.shape[-1]]
+
     q_embed = (q * cos) + (rotate_half(q) * sin)
     k_embed = (k * cos) + (rotate_half(k) * sin)
     return q_embed, k_embed

src/transformers/models/zamba2/modeling_zamba2.py

@@ -350,6 +350,12 @@ def apply_rotary_pos_emb(q, k, cos, sin, position_ids=None, unsqueeze_dim=1):
     """
     cos = cos.unsqueeze(unsqueeze_dim)
     sin = sin.unsqueeze(unsqueeze_dim)
+
+    # Adjust the rotary embedding dimensions if they don't match q's last dimension.
+    if cos.shape[-1] != q.shape[-1]:
+        cos = cos[..., : q.shape[-1]]
+        sin = sin[..., : q.shape[-1]]
+
     q_embed = (q * cos) + (rotate_half(q) * sin)
     k_embed = (k * cos) + (rotate_half(k) * sin)
     return q_embed, k_embed