Skip to content

Fix full-attention layer id mapping for Hybrid models (e.g. Qwen3-Next/Qwen3.5) #249

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
wanzhenchn wants to merge 1 commit into
base: Qwen3.5_v0.5.9
Choose a base branch
from
Open

Fix full-attention layer id mapping for Hybrid models (e.g. Qwen3-Next/Qwen3.5) #249

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
13 changes: 12 additions & 1 deletion python/sglang/srt/layers/attention/aiter_backend.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -123,7 +123,18 @@ def __init__(
model_runner.model_config.num_attention_heads // get_attention_tp_size()
)
self.head_dim = model_runner.model_config.head_dim
self.v_head_dim = model_runner.token_to_kv_pool.get_value_buffer(0).shape[-1]
# Hybrid models (e.g. Qwen3-Next): KV pool only exists for full-attention layers,
# not layer 0. Match intel_amx_backend: pick any mapped full-attention layer id.
_vbuf_layer_id = 0
if hasattr(
model_runner.token_to_kv_pool, "full_attention_layer_id_mapping"
):
_vbuf_layer_id = next(
iter(model_runner.token_to_kv_pool.full_attention_layer_id_mapping)
)
self.v_head_dim = model_runner.token_to_kv_pool.get_value_buffer(
_vbuf_layer_id
).shape[-1]
self.num_kv_head = model_runner.model_config.get_num_kv_heads(
get_attention_tp_size()
)
Expand Down
Loading