[Spec] Split the capture width from num_tokens_per_req and gate replay on it (#31255)

2026-07-16 14:59:11 -07:00
parent 059ac7efe8
commit fc1e3797b7
9 changed files with 141 additions and 83 deletions
@@ -77,7 +77,7 @@ class TestEagleDraftCudaGraphRunner(CustomTestCase):
            dsa_seed_topk=None,
        )
        runner.capture_bs = [1, CAPTURE_BS]
-        runner.num_tokens_per_req = 1
+        runner.captured_req_width = 1
        runner.speculative_num_steps = NUM_STEPS
        runner.seq_len_fill_value = SEQ_LEN_FILL_VALUE
        runner.require_mlp_tp_gather = False