[BUGFIX][0.9.1] FIX ring_mla input ‘query_lens’ to cpu (#2170)

### What this PR does / why we need it?

[BUGFIX][0.9.1] FIX ring_mla input ‘query_lens’ to cpu

### Does this PR introduce _any_ user-facing change?

### How was this patch tested?


Signed-off-by: xuyexiong <[email protected]>

Author: JC-ut0

vllm_ascend/worker/mtp_proposer_v1.py

@@ -218,7 +218,7 @@ def propose(
         self.hidden_states[:num_tokens] = target_hidden_states
 
         if attn_metadata.prefill is not None:
-            attn_metadata.prefill.query_lens = query_lens
+            attn_metadata.prefill.query_lens = query_lens.cpu()
             attn_metadata.prefill.input_positions = target_positions
 
         if not self.runner.torchair_graph_enabled: