fa3_rope (#4190)

2025-10-05 08:37:06 +08:00 · 2025-09-21 22:04:59 +08:00
parent 5223065d59
commit 9f1882d9a8
4 changed files with 17 additions and 7 deletions
--- a/fastdeploy/model_executor/layers/attention/flash_attn_backend.py
+++ b/fastdeploy/model_executor/layers/attention/flash_attn_backend.py
@@ -283,6 +283,7 @@ class FlashAttentionBackend(AttentionBackend):
            metadata.kv_token_num_cpu[0].item(),
            self.max_seq_len,
            getattr(layer, "cache_quant_type_str", "none"),
+            self.rope_3d,
        )

        res = self.flash_attn_func(