support qk norm (#3145)

2025-10-05 08:37:06 +08:00 · 2025-08-05 16:46:14 +08:00
parent 4a10e29804
commit 7ce00e597c
17 changed files with 791 additions and 201 deletions
--- a/fastdeploy/model_executor/layers/attention/append_attn_backend.py
+++ b/fastdeploy/model_executor/layers/attention/append_attn_backend.py
@@ -262,6 +262,9 @@ class AppendAttentionBackend(AttentionBackend):
            layer.linear_shift,
            layer.linear_smooth,
            metadata.kv_signal_data_list[layer.layer_id],
+            getattr(layer, "q_norm_weight", None),
+            getattr(layer, "k_norm_weight", None),
+            getattr(layer, "rms_norm_eps", 1e-6),
            metadata._fuse_kernel_compute_dtype,
            getattr(layer, "cache_quant_type_str", "none"),
            layer.use_neox_rotary_style,