frozenleaves/vllm
1
0
Fork 0
mirror of https://github.com/vllm-project/vllm.git synced 2025-10-20 14:53:52 +08:00
Code Issues Packages Projects Releases Wiki Activity

[BugFix] Fix EXAONE4 rotary embeddings (#23918)

Browse Source 
Signed-off-by: lkm2835 <lkm2835@gmail.com>
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
Co-authored-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
This commit is contained in:
Kyungmin Lee
2025-09-02 23:40:55 +09:00
committed by GitHub
parent 0a74e9d0f2
commit 38ba061f6f
1 changed files with 3 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
vllm/model_executor/models/exaone4.py
View File

@ -164,8 +164,8 @@ class Exaone4Attention(nn.Module):
is_sliding = config.layer_types[layer_idx] == "sliding_attention"
self.sliding_window = config.sliding_window if is_sliding else None
# apply rotary embeddings to every layer
self.apply_all_layers = not is_sliding
# apply rotary embeddings to every layer in full attention models
self.apply_rope_all_layers = "sliding_attention" not in config.layer_types
self.rotary_emb = get_rope(
self.head_dim,
@ -201,7 +201,7 @@ class Exaone4Attention(nn.Module):
k = self.k_norm(k)
k = k.flatten(-2, -1)
if self.sliding_window or self.apply_all_layers:
if self.sliding_window or self.apply_rope_all_layers:
q, k = self.rotary_emb(positions, q, k)
attn_output = self.attn(q, k, v)
output, _ = self.o_proj(attn_output)