first commit

Author: zhoutianzi666

tests/layers/test_append_attention.py

@@ -628,7 +628,6 @@ def cmp_append_attention(self, naive_cache_k=None, naive_cache_v=None, attn_mask
             12,
             (self.q_num_head + 2 * self.kv_num_head) // self.kv_num_head,
             self.blocksize,
-            speculate_max_draft_token_num + 1,
         )
         if self.use_dynamic_quant:
             cache_quant_type = "block_wise_fp8"

tests/layers/test_append_attention_with_output.py

@@ -479,7 +479,6 @@ def cmp_append_attention(self, naive_cache_k=None, naive_cache_v=None, attn_mask
             12,
             (self.q_num_head + 2 * self.kv_num_head) // self.kv_num_head,
             self.blocksize,
-            speculate_max_draft_token_num + 1,
         )
 
         # Warm up

tests/operators/test_tree_mask.py

@@ -254,7 +254,6 @@ def run_append_c16_attention(
             decoder_block_shape_q,
             self.num_q_head // self.num_kv_head,
             self.block_size,
-            decoder_step_token_num,
         )
         s_time = 0
         for i in range(self.run_time + self.warm_up):