xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-10 09:01:40 +08:00
Code Issues Packages Projects Releases Wiki Activity

[model] make llama4 compatible with pure dense layers (#17315)

Browse Source 
Signed-off-by: Lucia Fang <fanglu@fb.com>
This commit is contained in:
Lucia Fang 2025-04-28 19:22:22 -07:00 committed by GitHub
parent e136000595
commit b4ac4fa04d
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 2 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
vllm/model_executor/models/llama4.py
View File

@ -273,8 +273,8 @@ class Llama4DecoderLayer(nn.Module):
cache_config=cache_config, cache_config=cache_config,
prefix=f"{prefix}.self_attn", prefix=f"{prefix}.self_attn",
) )
is_moe_layer = (self.layer_idx + is_moe_layer = config.interleave_moe_layer_step > 0 and (
1) % config.interleave_moe_layer_step == 0 self.layer_idx + 1) % config.interleave_moe_layer_step == 0
if is_moe_layer: if is_moe_layer:
self.feed_forward = Llama4MoE( self.feed_forward = Llama4MoE(
config=config, config=config,