Misc fixes (#432)

python/sglang/srt/managers/router/model_rpc.py

@@ -426,7 +426,9 @@ class ModelRpcServer:
             # Only transfer the selected logprobs of the next token to CPU to reduce overhead.
             if last_logprobs is not None:
                 last_token_logprobs = (
-                    last_logprobs[torch.arange(len(batch.reqs)), next_token_ids].tolist()
+                    last_logprobs[
+                        torch.arange(len(batch.reqs), device=next_token_ids.device),
+                        next_token_ids].tolist()
                 )
 
             next_token_ids = next_token_ids.tolist()
@@ -587,6 +589,7 @@ class ModelRpcServer:
                     - req.prompt_tokens,
                     "completion_tokens_wo_jump_forward": req.completion_tokens_wo_jump_forward,
                     "finish_reason": str(req.finish_reason),  # FIXME: convert to the correct string
+                    "hit_stop_str": req.hit_stop_str,
                 }
                 if req.return_logprob:
                     (

python/sglang/srt/managers/router/model_runner.py

@@ -110,8 +110,8 @@ class InputMetadata:
         self.kv_last_page_len = torch.ones(
             (self.batch_size,), dtype=torch.int32, device="cuda"
         )
-        req_pool_indices_cpu = self.req_pool_indices.cpu().tolist()
-        seq_lens_cpu = self.seq_lens.tolist()
+        req_pool_indices_cpu = self.req_pool_indices.cpu().numpy()
+        seq_lens_cpu = self.seq_lens.cpu().numpy()
         self.kv_indices = torch.cat(
             [
                 self.req_to_token_pool.req_to_token[