fix error info

Signed-off-by: shikang-hangzhou <459956190@qq.com>

Author: shikang-hangzhou

vllm_ascend/multistream/ms_split.py

@@ -95,8 +95,10 @@ def model_input_split_v1_mla_attn(
     [num_prefills_pre, num_prefills_post
      ] = split_attn_int_type(attn_metadata.num_prefills,
                              max(0, seq_index - attn_metadata.num_decodes))
-    seq_lens = attn_metadata.prefill.seq_lens if attn_metadata.num_prefills > 0 else attn_metadata.decode.seq_lens
-    [seq_lens_pre, seq_lens_post] = split_attn_tensor_type(seq_lens, seq_index)
+    seq_lens = attn_metadata.seq_lens if attn_metadata.num_prefills > 0 else attn_metadata.decode.seq_lens
+    [seq_lens_pre, seq_lens_post
+     ] = split_attn_tensor_type(seq_lens, 
+                                max(0, seq_index - attn_metadata.num_decodes))
 
     query_start_loc_pre = query_start_loc_post = None
     if attn_metadata.query_start_loc is not None:
@@ -153,7 +155,7 @@ def model_input_split_v1_mla_attn(
                                                   attn_metadata.num_decodes:]
         ) - attn_metadata.prefill.query_start_loc[seq_index -
                                                   attn_metadata.num_decodes]
-        context_len_pre = seq_lens_pre[attn_metadata.num_decodes:]
+        context_len_pre = seq_lens_pre
         context_len_post = seq_lens_post
         prefill_max_query_len_pre = max(prefill_query_lens_pre)
         prefill_max_query_len_post = max(prefill_query_lens_post)