tmp: Test with reduced max token while keeping context length

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

Author: kthui

tests/fault_tolerance/cancellation/test_trtllm.py

@@ -35,7 +35,7 @@
     pytest.mark.gpu_1,
     pytest.mark.e2e,
     pytest.mark.model(FAULT_TOLERANCE_MODEL_NAME),
-    pytest.mark.post_merge,  # post_merge to pinpoint failure commit
+    pytest.mark.pre_merge,  # post_merge to pinpoint failure commit
     pytest.mark.parametrize("request_plane", ["nats", "tcp"], indirect=True),
 ]
 
@@ -428,7 +428,7 @@ def test_request_cancellation_trtllm_prefill_cancel(
                 )
 
 
-@pytest.mark.xfail(reason="Test fails only on CI", strict=False)
+# @pytest.mark.xfail(reason="Test fails only on CI", strict=False)
 @pytest.mark.timeout(195)  # 3x average
 def test_request_cancellation_trtllm_kv_transfer_cancel(
     request, runtime_services_dynamic_ports, predownload_models

tests/fault_tolerance/cancellation/utils.py

@@ -281,11 +281,11 @@ def send_cancellable_request(
         prompt += " Make sure it is" + " long" * 16000 + "!"
 
     if request_type == "completion":
-        return send_completion_request(prompt, 16384, frontend_port)
+        return send_completion_request(prompt, 16000, frontend_port)
     elif request_type == "chat_completion":
-        return send_chat_completion_request(prompt, 16384, frontend_port, stream=False)
+        return send_chat_completion_request(prompt, 16000, frontend_port, stream=False)
     elif request_type == "chat_completion_stream":
-        return send_chat_completion_request(prompt, 16384, frontend_port, stream=True)
+        return send_chat_completion_request(prompt, 16000, frontend_port, stream=True)
     else:
         raise ValueError(f"Unknown request type: {request_type}")