[TRTLLM-9247][infra] Upgrade NIXL to 0.7.1 (#9055)

Signed-off-by: Bo Deng <[email protected]>
Signed-off-by: jthomson04 <[email protected]>
Co-authored-by: jthomson04 <[email protected]>

Author: bo-nv

docker/common/install_nixl.sh

@@ -4,7 +4,7 @@ set -ex
 GITHUB_URL="https://github.com"
 UCX_INSTALL_PATH="/usr/local/ucx/"
 CUDA_PATH="/usr/local/cuda"
-NIXL_VERSION="0.5.0"
+NIXL_VERSION="0.7.1"
 NIXL_REPO="https://github.com/ai-dynamo/nixl.git"
 OLD_LD_LIBRARY_PATH=$LD_LIBRARY_PATH
 

jenkins/current_image_tags.properties

@@ -13,7 +13,7 @@
 #     images are adopted from PostMerge pipelines, the abbreviated commit hash is used instead.
 IMAGE_NAME=urm.nvidia.com/sw-tensorrt-docker/tensorrt-llm
 
-LLM_DOCKER_IMAGE=urm.nvidia.com/sw-tensorrt-docker/tensorrt-llm:pytorch-25.10-py3-x86_64-ubuntu24.04-trt10.13.3.9-skip-tritondevel-202511131803-8929
-LLM_SBSA_DOCKER_IMAGE=urm.nvidia.com/sw-tensorrt-docker/tensorrt-llm:pytorch-25.10-py3-aarch64-ubuntu24.04-trt10.13.3.9-skip-tritondevel-202511131803-8929
-LLM_ROCKYLINUX8_PY310_DOCKER_IMAGE=urm.nvidia.com/sw-tensorrt-docker/tensorrt-llm:cuda-13.0.2-devel-rocky8-x86_64-rocky8-py310-trt10.13.3.9-skip-tritondevel-202511131803-8929
-LLM_ROCKYLINUX8_PY312_DOCKER_IMAGE=urm.nvidia.com/sw-tensorrt-docker/tensorrt-llm:cuda-13.0.2-devel-rocky8-x86_64-rocky8-py312-trt10.13.3.9-skip-tritondevel-202511131803-8929
+LLM_DOCKER_IMAGE=urm.nvidia.com/sw-tensorrt-docker/tensorrt-llm:pytorch-25.10-py3-x86_64-ubuntu24.04-trt10.13.3.9-skip-tritondevel-202511200955-9055
+LLM_SBSA_DOCKER_IMAGE=urm.nvidia.com/sw-tensorrt-docker/tensorrt-llm:pytorch-25.10-py3-aarch64-ubuntu24.04-trt10.13.3.9-skip-tritondevel-202511200955-9055
+LLM_ROCKYLINUX8_PY310_DOCKER_IMAGE=urm.nvidia.com/sw-tensorrt-docker/tensorrt-llm:cuda-13.0.2-devel-rocky8-x86_64-rocky8-py310-trt10.13.3.9-skip-tritondevel-202511200955-9055
+LLM_ROCKYLINUX8_PY312_DOCKER_IMAGE=urm.nvidia.com/sw-tensorrt-docker/tensorrt-llm:cuda-13.0.2-devel-rocky8-x86_64-rocky8-py312-trt10.13.3.9-skip-tritondevel-202511200955-9055

tests/integration/defs/llmapi/test_llm_api_connector.py

@@ -356,11 +356,15 @@ def test_connector_disagg_prefill(enforce_single_worker, model_with_connector,
                                   save_async):
     model_fn, scheduler, worker = model_with_connector
 
-    model = model_fn(
+    prefill_worker = model_fn(
         disable_overlap_scheduler=True,
         cache_transceiver_config=CacheTransceiverConfig(backend="DEFAULT"))
 
-    sampling_params = SamplingParams(ignore_eos=True)
+    decode_worker = model_fn(
+        cache_transceiver_config=CacheTransceiverConfig(backend="DEFAULT"),
+        kv_connector_config=None)
+
+    sampling_params = SamplingParams(ignore_eos=True, max_tokens=16)
 
     disaggregated_params = DisaggregatedParams(request_type="context_only")
 
@@ -375,9 +379,16 @@ def test_connector_disagg_prefill(enforce_single_worker, model_with_connector,
         scheduler.request_finished.return_value = False
         worker.get_finished.return_value = [], []
 
-    model.generate([0] * 48,
-                   sampling_params=sampling_params,
-                   disaggregated_params=disaggregated_params)
+    result = prefill_worker.generate([0] * 48,
+                                     sampling_params=sampling_params,
+                                     disaggregated_params=disaggregated_params)
+
+    gen_disagg_params = result.disaggregated_params
+    gen_disagg_params.request_type = "generation_only"
+
+    result = decode_worker.generate([0] * 48,
+                                    sampling_params=sampling_params,
+                                    disaggregated_params=gen_disagg_params)
 
     assert scheduler.build_connector_meta.call_count == 1