https://nvbugs/5590408: Fix the setting of mMaxSeqLenKv

Signed-off-by: ziyixiong-nv <219238287+ziyixiong-nv@users.noreply.github.com>

Author: ziyixiong-nv

tensorrt_llm/_torch/attention_backend/trtllm.py

@@ -839,6 +839,10 @@ def prepare(self) -> None:
             self.prepare_flash_mla()
         # number of tokens needed in the kv cache for each sequence after the next pass
         kv_lens = cached_token_lens + self.seq_lens_kv if cached_token_lens is not None else self.seq_lens_kv
+        # Store actual KV length (without extra tokens) for use in kv_lens_runtime.
+        # num_extra_kv_tokens are for internal cache management but should not be reported
+        # as actual past KV length in host_past_key_value_lengths.
+        self.kv_lens_actual = kv_lens.clone()
         # self.kv_lens is the valid kv cache length, while the self.kv_lens_cuda is
         # the sequence length including the cached tokens and the input tokens.
         self.kv_lens[:self.num_seqs].copy_(
@@ -881,7 +885,9 @@ def prepare(self) -> None:
             ) <= self.kv_cache_manager.max_seq_len, error_message
 
         self.kv_lens_cuda_runtime = self.kv_lens_cuda[:self.num_seqs]
-        self.kv_lens_runtime = self.kv_lens[:self.num_seqs]
+        # Use actual KV length (without extra tokens) for kv_lens_runtime,
+        # which becomes host_past_key_value_lengths and eventually mMaxSeqLenKv.
+        self.kv_lens_runtime = self.kv_lens_actual[:self.num_seqs]
         self.prompt_lens_cuda_runtime = self.prompt_lens_cuda[:self.num_seqs]
         self.prompt_lens_cpu_runtime = self.prompt_lens_cpu[:self.num_seqs]
         self.host_request_types_runtime = self.host_request_types[:self.