Merge pull request #14 from lancerts/minor-fix

minor fix test

tests/test_flash_mla.py

@@ -7,7 +7,7 @@ import triton
 from flash_mla import get_mla_metadata, flash_mla_with_kvcache
 
 
-def scaled_dot_product_attention(query, key, value, is_causal=False):
+def scaled_dot_product_attention(query, key, value, h_q, h_kv, is_causal=False):
     query = query.float()
     key = key.float()
     value = value.float()
@@ -76,6 +76,8 @@ def test_flash_mla(b, s_q, mean_sk, h_q, h_kv, d, dv, causal, varlen):
                 q[i].transpose(0, 1),
                 blocked_k.view(-1, h_kv, d)[begin:end].transpose(0, 1),
                 blocked_v.view(-1, h_kv, dv)[begin:end].transpose(0, 1),
+                h_q=h_q,
+                h_kv=h_kv,
                 is_causal=causal,
             )
             out[i] = O.transpose(0, 1)