[Spec] Rename num_tokens_per_bs to num_tokens_per_req (#30977)

test/registered/attention/unittests/dense/test_tbo.py

@@ -170,8 +170,8 @@ class TestTboAttnDenseAttentionBackendCorrectness(CustomTestCase):
         )
 
         capture_bs = case.batch_size
-        num_tokens_per_bs = sum(case.extend_lens) // capture_bs
-        num_tokens = capture_bs * num_tokens_per_bs
+        num_tokens_per_req = sum(case.extend_lens) // capture_bs
+        num_tokens = capture_bs * num_tokens_per_req
         split_seq_index, split_token_index = (
             compute_split_indices_for_cuda_graph_replay(
                 forward_mode=batch.forward_mode,

test/registered/kv_canary/test_self_unit_capacities.py

@@ -59,7 +59,7 @@ class TestComputeLaunchCapacities(CustomTestCase):
         )
 
     def test_from_args_treats_missing_speculative_draft_tokens_as_zero(self) -> None:
-        """per_forward_write_entry_capacity is floored by max_prefill_tokens when batch * tokens_per_bs is smaller."""
+        """per_forward_write_entry_capacity is floored by max_prefill_tokens when batch * tokens_per_req is smaller."""
         server_args = self._make_server_args(max_bs=2)
         server_args.speculative_num_draft_tokens = None
 

test/registered/lora/test_chunked_sgmv_backend.py

@@ -1032,7 +1032,7 @@ class TestChunkedSGMV(unittest.TestCase):
         backend = ChunkedSgmvLoRABackend(
             max_loras_per_batch=5, device=self.device, server_args=mock_server_args
         )
-        backend.init_cuda_graph_batch_info(max_bs_in_cuda_graph=8, num_tokens_per_bs=1)
+        backend.init_cuda_graph_batch_info(max_bs_in_cuda_graph=8, num_tokens_per_req=1)
 
         lora_ranks = [8] * 5
         scalings = [1.0] * 5

test/registered/unit/spec/test_eagle_draft_cuda_graph_runner.py

@@ -77,7 +77,7 @@ class TestEagleDraftCudaGraphRunner(CustomTestCase):
             dsa_seed_topk=None,
         )
         runner.capture_bs = [1, CAPTURE_BS]
-        runner.num_tokens_per_bs = 1
+        runner.num_tokens_per_req = 1
         runner.speculative_num_steps = NUM_STEPS
         runner.seq_len_fill_value = SEQ_LEN_FILL_VALUE
         runner.require_mlp_tp_gather = False