[TRTLLM-13948][feat] DeepSeek V3 Lite TranceiverV2 migration

tests/integration/defs/accuracy/test_disaggregated_serving.py

@@ -1048,13 +1048,15 @@ def test_nixl_backend(self):
             "disable_overlap_scheduler": True,
             "cache_transceiver_config": {
                 "backend": "NIXL",
+                "transceiver_runtime": "PYTHON",
                 "max_tokens_in_buffer": 4096
             }
         }
         gen_server_config = {
             "disable_overlap_scheduler": True,
             "cache_transceiver_config": {
                 "backend": "NIXL",
+                "transceiver_runtime": "PYTHON",
                 "max_tokens_in_buffer": 4096
             }
         }

tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp1_gentp1_deepseek_v3_lite.yaml

@@ -9,10 +9,12 @@ context_servers:
   tensor_parallel_size: 1
   pipeline_parallel_size: 1
   cache_transceiver_config:
-    backend: DEFAULT
+    backend: NIXL
+    transceiver_runtime: PYTHON
 generation_servers:
   num_instances: 1
   tensor_parallel_size: 1
   pipeline_parallel_size: 1
   cache_transceiver_config:
-    backend: DEFAULT
+    backend: NIXL
+    transceiver_runtime: PYTHON

tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp1_gentp1_deepseek_v3_lite_one_mtp.yaml

@@ -13,11 +13,13 @@ context_servers:
   pipeline_parallel_size: 1
   enable_attention_dp: true
   cache_transceiver_config:
-    backend: DEFAULT
+    backend: NIXL
+    transceiver_runtime: PYTHON
 generation_servers:
   num_instances: 1
   tensor_parallel_size: 1
   pipeline_parallel_size: 1
   enable_attention_dp: false
   cache_transceiver_config:
-    backend: DEFAULT
+    backend: NIXL
+    transceiver_runtime: PYTHON

tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp1_gentp1_deepseek_v3_lite_one_mtp_attention_dp_overlap.yaml

@@ -13,12 +13,14 @@ context_servers:
   enable_attention_dp: true
   disable_overlap_scheduler: true
   cache_transceiver_config:
-    backend: DEFAULT
+    backend: NIXL
+    transceiver_runtime: PYTHON
 generation_servers:
   num_instances: 1
   tensor_parallel_size: 1
   pipeline_parallel_size: 1
   enable_attention_dp: true
   disable_overlap_scheduler: false
   cache_transceiver_config:
-    backend: DEFAULT
+    backend: NIXL
+    transceiver_runtime: PYTHON

tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp1_gentp1_deepseek_v3_lite_one_mtp_ctxpp2_gentp2.yaml

@@ -13,7 +13,8 @@ context_servers:
     decoding_type: MTP
     max_draft_len: 1
   cache_transceiver_config:
-    backend: DEFAULT
+    backend: NIXL
+    transceiver_runtime: PYTHON
 generation_servers:
   num_instances: 1
   tensor_parallel_size: 2
@@ -23,4 +24,5 @@ generation_servers:
     decoding_type: MTP
     max_draft_len: 1
   cache_transceiver_config:
-    backend: DEFAULT
+    backend: NIXL
+    transceiver_runtime: PYTHON

tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp1_gentp1_deepseek_v3_lite_two_mtp.yaml

@@ -13,11 +13,13 @@ context_servers:
   pipeline_parallel_size: 1
   enable_attention_dp: true
   cache_transceiver_config:
-    backend: DEFAULT
+    backend: NIXL
+    transceiver_runtime: PYTHON
 generation_servers:
   num_instances: 1
   tensor_parallel_size: 1
   pipeline_parallel_size: 1
   enable_attention_dp: false
   cache_transceiver_config:
-    backend: DEFAULT
+    backend: NIXL
+    transceiver_runtime: PYTHON

tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp2_gentp2_deepseek_v3_lite.yaml

@@ -9,10 +9,12 @@ context_servers:
   tensor_parallel_size: 2
   pipeline_parallel_size: 1
   cache_transceiver_config:
-    backend: DEFAULT
+    backend: NIXL
+    transceiver_runtime: PYTHON
 generation_servers:
   num_instances: 1
   tensor_parallel_size: 2
   pipeline_parallel_size: 1
   cache_transceiver_config:
-    backend: DEFAULT
+    backend: NIXL
+    transceiver_runtime: PYTHON

tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp2_gentp2_deepseek_v3_lite_attention_dp.yaml

@@ -10,11 +10,13 @@ context_servers:
   pipeline_parallel_size: 1
   enable_attention_dp: true
   cache_transceiver_config:
-    backend: DEFAULT
+    backend: NIXL
+    transceiver_runtime: PYTHON
 generation_servers:
   num_instances: 1
   tensor_parallel_size: 2
   pipeline_parallel_size: 1
   enable_attention_dp: true
   cache_transceiver_config:
-    backend: DEFAULT
+    backend: NIXL
+    transceiver_runtime: PYTHON

tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp2_gentp2_deepseek_v3_lite_attention_dp_one.yaml

@@ -10,11 +10,13 @@ context_servers:
   pipeline_parallel_size: 1
   enable_attention_dp: true
   cache_transceiver_config:
-    backend: DEFAULT
+    backend: NIXL
+    transceiver_runtime: PYTHON
 generation_servers:
   num_instances: 1
   tensor_parallel_size: 2
   pipeline_parallel_size: 1
   enable_attention_dp: false
   cache_transceiver_config:
-    backend: DEFAULT
+    backend: NIXL
+    transceiver_runtime: PYTHON

tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp2_gentp2_deepseek_v3_lite_attention_dp_one_mtp.yaml

@@ -13,11 +13,13 @@ context_servers:
   pipeline_parallel_size: 1
   enable_attention_dp: true
   cache_transceiver_config:
-    backend: DEFAULT
+    backend: NIXL
+    transceiver_runtime: PYTHON
 generation_servers:
   num_instances: 1
   tensor_parallel_size: 2
   pipeline_parallel_size: 1
   enable_attention_dp: false
   cache_transceiver_config:
-    backend: DEFAULT
+    backend: NIXL
+    transceiver_runtime: PYTHON

tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp2_gentp2_deepseek_v3_lite_attention_dp_overlap.yaml

@@ -10,12 +10,14 @@ context_servers:
   enable_attention_dp: true
   disable_overlap_scheduler: true
   cache_transceiver_config:
-    backend: DEFAULT
+    backend: NIXL
+    transceiver_runtime: PYTHON
 generation_servers:
   num_instances: 1
   tensor_parallel_size: 2
   pipeline_parallel_size: 1
   enable_attention_dp: true
   disable_overlap_scheduler: false
   cache_transceiver_config:
-    backend: DEFAULT
+    backend: NIXL
+    transceiver_runtime: PYTHON

tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp2_gentp2_deepseek_v3_lite_attention_dp_overlap_cuda_graph.yaml

@@ -9,7 +9,8 @@ context_servers:
   enable_attention_dp: true
   disable_overlap_scheduler: true
   cache_transceiver_config:
-    backend: DEFAULT
+    backend: NIXL
+    transceiver_runtime: PYTHON
 generation_servers:
   num_instances: 1
   tensor_parallel_size: 2
@@ -19,4 +20,5 @@ generation_servers:
     enable_padding: false
   disable_overlap_scheduler: false
   cache_transceiver_config:
-    backend: DEFAULT
+    backend: NIXL
+    transceiver_runtime: PYTHON

tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp2_gentp2_deepseek_v3_lite_nixl.yaml

@@ -9,9 +9,11 @@ context_servers:
   pipeline_parallel_size: 1
   cache_transceiver_config:
     backend: NIXL
+    transceiver_runtime: PYTHON
 generation_servers:
   num_instances: 1
   tensor_parallel_size: 2
   pipeline_parallel_size: 1
   cache_transceiver_config:
     backend: NIXL
+    transceiver_runtime: PYTHON

tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp2_gentp2_deepseek_v3_lite_overlap_cuda_graph.yaml

@@ -8,7 +8,8 @@ context_servers:
   pipeline_parallel_size: 1
   disable_overlap_scheduler: true
   cache_transceiver_config:
-    backend: DEFAULT
+    backend: NIXL
+    transceiver_runtime: PYTHON
 generation_servers:
   num_instances: 1
   tensor_parallel_size: 2
@@ -17,4 +18,5 @@ generation_servers:
     enable_padding: false
   disable_overlap_scheduler: false
   cache_transceiver_config:
-    backend: DEFAULT
+    backend: NIXL
+    transceiver_runtime: PYTHON

tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp2ep2pp2_gentp4_deepseek_v3_lite_one_mtp_block_reuse.yaml

@@ -16,7 +16,8 @@ context_servers:
   kv_cache_config:
     enable_block_reuse: true
   cache_transceiver_config:
-    backend: DEFAULT
+    backend: NIXL
+    transceiver_runtime: PYTHON
 generation_servers:
   num_instances: 1
   tensor_parallel_size: 4
@@ -29,4 +30,5 @@ generation_servers:
   kv_cache_config:
     enable_block_reuse: true
   cache_transceiver_config:
-    backend: DEFAULT
+    backend: NIXL
+    transceiver_runtime: PYTHON

tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp2ep2pp2_gentp4_deepseek_v3_lite_one_mtp_block_reuse_chunked.yaml

@@ -18,7 +18,8 @@ context_servers:
   kv_cache_config:
     enable_block_reuse: true
   cache_transceiver_config:
-    backend: DEFAULT
+    backend: NIXL
+    transceiver_runtime: PYTHON
 generation_servers:
   num_instances: 1
   tensor_parallel_size: 4
@@ -31,4 +32,5 @@ generation_servers:
   kv_cache_config:
     enable_block_reuse: true
   cache_transceiver_config:
-    backend: DEFAULT
+    backend: NIXL
+    transceiver_runtime: PYTHON

tests/integration/defs/disaggregated/test_configs/disagg_config_deepseek_v3_lite_empty_batch.yaml

@@ -17,7 +17,8 @@ context_servers:
     max_tokens: 512
   cache_transceiver_config:
     max_tokens_in_buffer: 8448
-    backend: DEFAULT
+    backend: NIXL
+    transceiver_runtime: PYTHON
 generation_servers:
   num_instances: 1
   tensor_parallel_size: 1
@@ -38,6 +39,7 @@ generation_servers:
     max_tokens: 2560
   cache_transceiver_config:
     max_tokens_in_buffer: 8448
-    backend: DEFAULT
+    backend: NIXL
+    transceiver_runtime: PYTHON
   stream_interval: 1
   num_postprocess_workers: 1

tests/integration/defs/disaggregated/test_configs/disagg_config_gentp2_deepseek_v3_lite_attention_dp_gen_only.yaml

@@ -14,7 +14,8 @@ generation_servers:
   pipeline_parallel_size: 1
   enable_attention_dp: True
   cache_transceiver_config:
-    backend: DEFAULT
+    backend: NIXL
+    transceiver_runtime: PYTHON
   urls:
       - "localhost:8002"
       - "localhost:8003"

tests/integration/test_lists/test-db/l0_gb200_multi_nodes_perf_sanity_ctx1_node1_gpu1_gen1_node1_gpu2.yml

@@ -18,3 +18,4 @@ l0_gb200_multi_nodes_perf_sanity_ctx1_node1_gpu1_gen1_node1_gpu2:
   - perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb200_gpt-oss-120b-fp4_1k1k_con512_ctx1_tp1_gen1_dep2_eplb0_mtp0_ccb-NIXL] TIMEOUT (90)
   - perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb200_gpt-oss-120b-fp4_8k1k_con512_ctx1_tp1_gen1_dep2_eplb0_mtp0_ccb-NIXL] TIMEOUT (90)
   - perf/test_perf_sanity.py::test_e2e[disagg_upload-e2e-gb200_gpt-oss-120b-fp4_1k1k_con2048_ctx1_tp1_gen1_dep2_eplb0_mtp0_ccb-NIXL] TIMEOUT (90)
+  - perf/test_perf_sanity.py::test_e2e[disagg_upload-e2e-gb200_deepseek-v3-lite-fp8_1k1k_con256_ctx1_tp1_gen1_tp2_eplb0_mtp0_ccb-NIXL] TIMEOUT (90)

tests/scripts/perf-sanity/disaggregated/gb200_deepseek-v3-lite-fp8_1k1k_con256_ctx1_tp1_gen1_tp2_eplb0_mtp0_ccb-NIXL.yaml

@@ -0,0 +1,86 @@
+metadata:
+  model_name: deepseek_v3_lite_fp8
+  precision: fp8
+  model_dir_name: DeepSeek-V3-Lite/fp8
+  supported_gpus:
+  - GB200
+  script_file: disaggr_torch.slurm
+  benchmark_type: 1k1k
+slurm:
+  script_file: disaggr_torch.slurm
+  partition: <partition>
+  account: <account>
+  job_time: 02:00:00
+  job_name: unified-benchmark
+  extra_args: --gres=gpu:4
+  numa_bind: true
+benchmark:
+  mode: e2e
+  use_nv_sa_benchmark: false
+  multi_round: 10
+  benchmark_ratio: 0.0
+  streaming: true
+  concurrency_list: '256'
+  input_length: 1024
+  output_length: 1024
+  dataset_file: <dataset_file>
+hardware:
+  gpus_per_node: 4
+  num_ctx_servers: 1
+  num_gen_servers: 1
+environment:
+  container_mount: <container_mount>
+  container_image: <container_image>
+  model_path: <model_path>
+  trtllm_repo: ''
+  build_wheel: false
+  work_dir: <full_path_to_work_dir>
+  worker_env_var: TLLM_LOG_LEVEL=INFO TRTLLM_SERVER_DISABLE_GC=1 TRTLLM_WORKER_DISABLE_GC=1 TRTLLM_ENABLE_PDL=1 ENROOT_ALLOW_DEV=yes
+  server_env_var: TRTLLM_SERVER_DISABLE_GC=1
+profiling:
+  nsys_on: false
+worker_config:
+  gen:
+    print_iter_log: true
+    max_batch_size: 512
+    max_num_tokens: 8192
+    tensor_parallel_size: 2
+    moe_expert_parallel_size: 2
+    pipeline_parallel_size: 1
+    context_parallel_size: 1
+    enable_attention_dp: false
+    cuda_graph_config:
+      enable_padding: true
+      max_batch_size: 512
+    kv_cache_config:
+      use_kv_cache_manager_v2: false
+      enable_block_reuse: false
+      free_gpu_memory_fraction: 0.9
+      dtype: fp8
+    cache_transceiver_config:
+      max_tokens_in_buffer: 4096
+      backend: NIXL
+      transceiver_runtime: PYTHON
+    disable_overlap_scheduler: false
+    num_postprocess_workers: 4
+    stream_interval: 20
+  ctx:
+    print_iter_log: true
+    max_batch_size: 64
+    max_num_tokens: 8192
+    tensor_parallel_size: 1
+    moe_expert_parallel_size: 1
+    pipeline_parallel_size: 1
+    context_parallel_size: 1
+    enable_attention_dp: false
+    cuda_graph_config: null
+    kv_cache_config:
+      use_kv_cache_manager_v2: false
+      enable_block_reuse: false
+      free_gpu_memory_fraction: 0.9
+      dtype: fp8
+    cache_transceiver_config:
+      max_tokens_in_buffer: 4096
+      backend: NIXL
+      transceiver_runtime: PYTHON
+    disable_overlap_scheduler: true