NVIDIA / NVIDIA/TensorRT-LLM

[Bug]: trtllm-serve hangs when TP > 1 on l40s on aws g6e.12xlarge

Open
#9,680 2 comments 0 reactions 1 assignee View on GitHub

@QiJune is already working on this.

Since Dec 12, 2025.

bug Scale-out
Dominant language
Python
Stars
14.7k
Forks
2.8k
Avg merge
2d 23h
Merged PRs (30d)
489

Description

System Info

TensorRT-LLM Container: nvcr.io/nvidia/tensorrt-llm/release:1.2.0rc2
AWS Instance: g6e.12xlarge (4xl40s)
CPU Memory: 20GB
CPU Shared Memory: 64MB (the default). GPT indicated that this could cause the hang behavior

# nvidia-smi

Thu Nov 20 23:57:18 2025       
+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 580.95.05              Driver Version: 580.95.05      CUDA Version: 13.0     |
+-----------------------------------------+------------------------+----------------------+
| GPU  Name                 Persistence-M | Bus-Id          Disp.A | Volatile Uncorr. ECC |
| Fan  Temp   Perf          Pwr:Usage/Cap |           Memory-Usage | GPU-Util  Compute M. |
|                                         |                        |               MIG M. |
|=========================================+========================+======================|
|   0  NVIDIA L40S                    On  |   00000000:38:00.0 Off |                    0 |
| N/A   28C    P0             78W /  350W |     886MiB /  46068MiB |      0%      Default |
|                                         |                        |                  N/A |
+-----------------------------------------+------------------------+----------------------+
|   1  NVIDIA L40S                    On  |   00000000:3A:00.0 Off |                    0 |
| N/A   27C    P0             83W /  350W |    4333MiB /  46068MiB |      0%      Default |
|                                         |                        |                  N/A |
+-----------------------------------------+------------------------+----------------------+
|   2  NVIDIA L40S                    On  |   00000000:3C:00.0 Off |                    0 |
| N/A   28C    P0             78W /  350W |     435MiB /  46068MiB |      0%      Default |
|                                         |                        |                  N/A |
+-----------------------------------------+------------------------+----------------------+
|   3  NVIDIA L40S                    On  |   00000000:3E:00.0 Off |                    0 |
| N/A   28C    P0             79W /  350W |     435MiB /  46068MiB |      0%      Default |
|                                         |                        |                  N/A |
+-----------------------------------------+------------------------+----------------------+

+-----------------------------------------------------------------------------------------+
| Processes:                                                                              |
|  GPU   GI   CI              PID   Type   Process name                        GPU Memory |
|        ID   ID                                                               Usage      |
|=========================================================================================|
|    0   N/A  N/A             267      C   /usr/bin/python                         446MiB |
|    0   N/A  N/A             935      C   /usr/bin/python                         426MiB |
|    2   N/A  N/A             937      C   /usr/bin/python                         426MiB |
|    3   N/A  N/A             938      C   /usr/bin/python                         426MiB |
+-----------------------------------------------------------------------------------------+
Who can help?

No response

Information
  • The official example scripts
  • My own modified scripts
Tasks
  • An officially supported task in the examples folder (such as GLUE/SQuAD, ...)
  • My own task or dataset (give details below)
Reproduction

Run trtllm-serve with an tensorrt engine:

trtllm-serve /tmp/llama3-8b-instruct \

--tokenizer "meta-llama/Llama-3.1-8B-Instruct" \
--backend tensorrt \
--port 8002 \
--host 0.0.0.0 \
--max_batch_size 2 \
--max_num_tokens 8192 \
--kv_cache_free_gpu_memory_fraction 0.90 \
--max_beam_width 1 \
--tp_size 4 \
--pp_size 1 \
--otlp_traces_endpoint="http://172.17.0.1:4317/" \
--log_level verbose
Expected behavior

Startup to complete successfully.

actual behavior

The process did not return and did not log anymore (aka hang).

`/usr/local/lib/python3.12/dist-packages/torch/cuda/init.py:63: FutureWarning: The pynvml package is deprecated. Please install nvidia-ml-py instead. If you did not install pynvml directly, please report this to the maintainers of the package that installed pynvml for you.
import pynvml # type: ignore[import]
[1763682855.106417] [8b3b04d4aa39:267 :0] ucp_context.c:2339 UCX WARN UCP API version is incompatible: required >= 1.20, actual 1.19.0 (loaded from /usr/local/ucx//lib/libucp.so.0)
[1763682855.206629] [8b3b04d4aa39:267 :0] ucp_context.c:2339 UCX WARN UCP API version is incompatible: required >= 1.20, actual 1.19.0 (loaded from /usr/local/ucx//lib/libucp.so.0)
[TensorRT-LLM] TensorRT LLM version: 1.2.0rc2
[1/2] /usr/local/cuda/bin/nvcc --generate-dependencies-with-compile --dependency-output moe_align_kernel.cuda.o.d -DTORCH_EXTENSION_NAME=moe_align_ext -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.12/dist-packages/torch/include -isystem /usr/local/lib/python3.12/dist-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/include/python3.12 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_100,code=sm_100 -gencode=arch=compute_120,code=compute_120 -gencode=arch=compute_120,code=sm_120 -gencode=arch=compute_75,code=sm_75 -gencode=arch=compute_80,code=sm_80 -gencode=arch=compute_86,code=sm_86 -gencode=arch=compute_90,code=sm_90 --compiler-options '-fPIC' -O3 --use_fast_math -U__CUDA_NO_HALF_OPERATORS__ -U__CUDA_NO_HALF_CONVERSIONS__ --expt-relaxed-constexpr -std=c++17 -c /usr/local/lib/python3.12/dist-packages/tensorrt_llm/_torch/auto_deploy/custom_ops/fused_moe/moe_align_kernel.cu -o moe_align_kernel.cuda.o
[2/2] c++ moe_align_kernel.cuda.o -shared -L/usr/local/lib/python3.12/dist-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o moe_align_ext.so
/usr/local/lib/python3.12/dist-packages/tensorrt_llm/serve/openai_protocol.py:94: UserWarning: Field name "schema" in "ResponseFormat" shadows an attribute in parent "OpenAIBaseModel"
class ResponseFormat(OpenAIBaseModel):
[11/20/2025-23:56:50] [TRT-LLM] [I] Using LLM with TensorRT backend
tokenizer_config.json: 100%|███████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 55.4k/55.4k [00:00<00:00, 27.9MB/s]
tokenizer.json: 100%|██████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 9.09M/9.09M [00:00<00:00, 37.5MB/s]
special_tokens_map.json: 100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 296/296 [00:00<00:00, 5.22MB/s]
[11/20/2025-23:56:51] [TRT-LLM] [W] Implicitly setting LLaMAConfig.fc_after_embed = False
[11/20/2025-23:56:51] [TRT-LLM] [W] Implicitly setting LLaMAConfig.use_input_layernorm_in_first_layer = True
[11/20/2025-23:56:51] [TRT-LLM] [W] Implicitly setting LLaMAConfig.use_last_layernorm = True
[11/20/2025-23:56:51] [TRT-LLM] [W] Implicitly setting LLaMAConfig.layer_idx_offset = 0
[11/20/2025-23:56:51] [TRT-LLM] [W] Implicitly setting LLaMAConfig.has_partial_lora_mask = False
[11/20/2025-23:56:51] [TRT-LLM] [W] Implicitly setting LLaMAConfig.tie_word_embeddings = False
[11/20/2025-23:56:51] [TRT-LLM] [W] Implicitly setting LLaMAConfig.speculative_model_dir = None
[11/20/2025-23:56:51] [TRT-LLM] [W] Implicitly setting LLaMAConfig.speculative_config = None
[11/20/2025-23:56:51] [TRT-LLM] [I] Set PluginConfig.dtype to bfloat16.
[11/20/2025-23:56:51] [TRT-LLM] [I] Set PluginConfig.bert_attention_plugin to auto.
[11/20/2025-23:56:51] [TRT-LLM] [I] Set PluginConfig.gpt_attention_plugin to auto.
[11/20/2025-23:56:51] [TRT-LLM] [I] Set PluginConfig.gemm_plugin to None.
[11/20/2025-23:56:51] [TRT-LLM] [I] Set PluginConfig.gemm_swiglu_plugin to None.
[11/20/2025-23:56:51] [TRT-LLM] [I] Set PluginConfig.fp8_rowwise_gemm_plugin to None.
[11/20/2025-23:56:51] [TRT-LLM] [I] Set PluginConfig.qserve_gemm_plugin to None.
[11/20/2025-23:56:51] [TRT-LLM] [I] Set PluginConfig.identity_plugin to None.
[11/20/2025-23:56:51] [TRT-LLM] [I] Set PluginConfig.nccl_plugin to bfloat16.
[11/20/2025-23:56:51] [TRT-LLM] [I] Set PluginConfig.lora_plugin to None.
[11/20/2025-23:56:51] [TRT-LLM] [I] Set PluginConfig.dora_plugin to False.
[11/20/2025-23:56:51] [TRT-LLM] [I] Set PluginConfig.weight_only_groupwise_quant_matmul_plugin to None.
[11/20/2025-23:56:51] [TRT-LLM] [I] Set PluginConfig.weight_only_quant_matmul_plugin to None.
[11/20/2025-23:56:51] [TRT-LLM] [I] Set PluginConfig.smooth_quant_plugins to True.
[11/20/2025-23:56:51] [TRT-LLM] [I] Set PluginConfig.smooth_quant_gemm_plugin to None.
[11/20/2025-23:56:51] [TRT-LLM] [I] Set PluginConfig.layernorm_quantization_plugin to None.
[11/20/2025-23:56:51] [TRT-LLM] [I] Set PluginConfig.rmsnorm_quantization_plugin to None.
[11/20/2025-23:56:51] [TRT-LLM] [I] Set PluginConfig.quantize_per_token_plugin to False.
[11/20/2025-23:56:51] [TRT-LLM] [I] Set PluginConfig.quantize_tensor_plugin to False.
[11/20/2025-23:56:51] [TRT-LLM] [I] Set PluginConfig.moe_plugin to auto.
[11/20/2025-23:56:51] [TRT-LLM] [I] Set PluginConfig.mamba_conv1d_plugin to auto.
[11/20/2025-23:56:51] [TRT-LLM] [I] Set PluginConfig.low_latency_gemm_plugin to None.
[11/20/2025-23:56:51] [TRT-LLM] [I] Set PluginConfig.low_latency_gemm_swiglu_plugin to None.
[11/20/2025-23:56:51] [TRT-LLM] [I] Set PluginConfig.gemm_allreduce_plugin to None.
[11/20/2025-23:56:51] [TRT-LLM] [I] Set PluginConfig.context_fmha to True.
[11/20/2025-23:56:51] [TRT-LLM] [I] Set PluginConfig.bert_context_fmha_fp32_acc to False.
[11/20/2025-23:56:51] [TRT-LLM] [I] Set PluginConfig.paged_kv_cache to True.
[11/20/2025-23:56:51] [TRT-LLM] [I] Set PluginConfig.remove_input_padding to True.
[11/20/2025-23:56:51] [TRT-LLM] [I] Set PluginConfig.norm_quant_fusion to False.
[11/20/2025-23:56:51] [TRT-LLM] [I] Set PluginConfig.reduce_fusion to False.
[11/20/2025-23:56:51] [TRT-LLM] [I] Set PluginConfig.user_buffer to False.
[11/20/2025-23:56:51] [TRT-LLM] [I] Set PluginConfig.tokens_per_block to 32.
[11/20/2025-23:56:51] [TRT-LLM] [I] Set PluginConfig.use_paged_context_fmha to True.
[11/20/2025-23:56:51] [TRT-LLM] [I] Set PluginConfig.use_fp8_context_fmha to False.
[11/20/2025-23:56:51] [TRT-LLM] [I] Set PluginConfig.fuse_fp4_quant to False.
[11/20/2025-23:56:51] [TRT-LLM] [I] Set PluginConfig.multiple_profiles to False.
[11/20/2025-23:56:51] [TRT-LLM] [I] Set PluginConfig.paged_state to False.
[11/20/2025-23:56:51] [TRT-LLM] [I] Set PluginConfig.streamingllm to False.
[11/20/2025-23:56:51] [TRT-LLM] [I] Set PluginConfig.manage_weights to False.
[11/20/2025-23:56:51] [TRT-LLM] [I] Set PluginConfig.use_fused_mlp to True.
[11/20/2025-23:56:51] [TRT-LLM] [I] Set PluginConfig.pp_reduce_scatter to False.
[11/20/2025-23:56:51] [TRT-LLM] [W] The build_config is ignored for model format of TLLM_ENGINE.
[11/20/2025-23:56:51] [TRT-LLM] [W] Implicitly setting LLaMAConfig.fc_after_embed = False
[11/20/2025-23:56:51] [TRT-LLM] [W] Implicitly setting LLaMAConfig.use_input_layernorm_in_first_layer = True
[11/20/2025-23:56:51] [TRT-LLM] [W] Implicitly setting LLaMAConfig.use_last_layernorm = True
[11/20/2025-23:56:51] [TRT-LLM] [W] Implicitly setting LLaMAConfig.layer_idx_offset = 0
[11/20/2025-23:56:51] [TRT-LLM] [W] Implicitly setting LLaMAConfig.has_partial_lora_mask = False
[11/20/2025-23:56:51] [TRT-LLM] [W] Implicitly setting LLaMAConfig.tie_word_embeddings = False
[11/20/2025-23:56:51] [TRT-LLM] [W] Implicitly setting LLaMAConfig.speculative_model_dir = None
[11/20/2025-23:56:51] [TRT-LLM] [W] Implicitly setting LLaMAConfig.speculative_config = None
[11/20/2025-23:56:51] [TRT-LLM] [I] Set PluginConfig.dtype to bfloat16.
[11/20/2025-23:56:51] [TRT-LLM] [I] Set PluginConfig.bert_attention_plugin to auto.
[11/20/2025-23:56:51] [TRT-LLM] [I] Set PluginConfig.gpt_attention_plugin to auto.
[11/20/2025-23:56:51] [TRT-LLM] [I] Set PluginConfig.gemm_plugin to None.
[11/20/2025-23:56:51] [TRT-LLM] [I] Set PluginConfig.gemm_swiglu_plugin to None.
[11/20/2025-23:56:51] [TRT-LLM] [I] Set PluginConfig.fp8_rowwise_gemm_plugin to None.
[11/20/2025-23:56:51] [TRT-LLM] [I] Set PluginConfig.qserve_gemm_plugin to None.
[11/20/2025-23:56:51] [TRT-LLM] [I] Set PluginConfig.identity_plugin to None.
[11/20/2025-23:56:51] [TRT-LLM] [I] Set PluginConfig.nccl_plugin to bfloat16.
[11/20/2025-23:56:51] [TRT-LLM] [I] Set PluginConfig.lora_plugin to None.
[11/20/2025-23:56:51] [TRT-LLM] [I] Set PluginConfig.dora_plugin to False.
[11/20/2025-23:56:51] [TRT-LLM] [I] Set PluginConfig.weight_only_groupwise_quant_matmul_plugin to None.
[11/20/2025-23:56:51] [TRT-LLM] [I] Set PluginConfig.weight_only_quant_matmul_plugin to None.
[11/20/2025-23:56:51] [TRT-LLM] [I] Set PluginConfig.smooth_quant_plugins to True.
[11/20/2025-23:56:51] [TRT-LLM] [I] Set PluginConfig.smooth_quant_gemm_plugin to None.
[11/20/2025-23:56:51] [TRT-LLM] [I] Set PluginConfig.layernorm_quantization_plugin to None.
[11/20/2025-23:56:51] [TRT-LLM] [I] Set PluginConfig.rmsnorm_quantization_plugin to None.
[11/20/2025-23:56:51] [TRT-LLM] [I] Set PluginConfig.quantize_per_token_plugin to False.
[11/20/2025-23:56:51] [TRT-LLM] [I] Set PluginConfig.quantize_tensor_plugin to False.
[11/20/2025-23:56:51] [TRT-LLM] [I] Set PluginConfig.moe_plugin to auto.
[11/20/2025-23:56:51] [TRT-LLM] [I] Set PluginConfig.mamba_conv1d_plugin to auto.
[11/20/2025-23:56:51] [TRT-LLM] [I] Set PluginConfig.low_latency_gemm_plugin to None.
[11/20/2025-23:56:51] [TRT-LLM] [I] Set PluginConfig.low_latency_gemm_swiglu_plugin to None.
[11/20/2025-23:56:51] [TRT-LLM] [I] Set PluginConfig.gemm_allreduce_plugin to None.
[11/20/2025-23:56:51] [TRT-LLM] [I] Set PluginConfig.context_fmha to True.
[11/20/2025-23:56:51] [TRT-LLM] [I] Set PluginConfig.bert_context_fmha_fp32_acc to False.
[11/20/2025-23:56:51] [TRT-LLM] [I] Set PluginConfig.paged_kv_cache to True.
[11/20/2025-23:56:51] [TRT-LLM] [I] Set PluginConfig.remove_input_padding to True.
[11/20/2025-23:56:51] [TRT-LLM] [I] Set PluginConfig.norm_quant_fusion to False.
[11/20/2025-23:56:51] [TRT-LLM] [I] Set PluginConfig.reduce_fusion to False.
[11/20/2025-23:56:51] [TRT-LLM] [I] Set PluginConfig.user_buffer to False.
[11/20/2025-23:56:51] [TRT-LLM] [I] Set PluginConfig.tokens_per_block to 32.
[11/20/2025-23:56:51] [TRT-LLM] [I] Set PluginConfig.use_paged_context_fmha to True.
[11/20/2025-23:56:51] [TRT-LLM] [I] Set PluginConfig.use_fp8_context_fmha to False.
[11/20/2025-23:56:51] [TRT-LLM] [I] Set PluginConfig.fuse_fp4_quant to False.
[11/20/2025-23:56:51] [TRT-LLM] [I] Set PluginConfig.multiple_profiles to False.
[11/20/2025-23:56:51] [TRT-LLM] [I] Set PluginConfig.paged_state to False.
[11/20/2025-23:56:51] [TRT-LLM] [I] Set PluginConfig.streamingllm to False.
[11/20/2025-23:56:51] [TRT-LLM] [I] Set PluginConfig.manage_weights to False.
[11/20/2025-23:56:51] [TRT-LLM] [I] Set PluginConfig.use_fused_mlp to True.
[11/20/2025-23:56:51] [TRT-LLM] [I] Set PluginConfig.pp_reduce_scatter to False.
[11/20/2025-23:56:51] [TRT-LLM] [V] LLM.args.mpi_session: None

[11/20/2025-23:56:51] [TRT-LLM] [I] start MpiSession with 4 workers
[11/20/2025-23:56:51] [TRT-LLM] [V] LLM create MpiPoolSession

The argument trust_remote_code is to be used with Auto classes. It has no effect here and is ignored.
[11/20/2025-23:56:51] [TRT-LLM] [V] Unable to load HF config from None: checkpoint_dir is None. Cannot load model config without a valid checkpoint directory.. Falling back.
[11/20/2025-23:56:51] [TRT-LLM] [W] Orchestrator is creating IPC executor
[11/20/2025-23:56:51] [TRT-LLM] [V] using external mpi session ...

rank 0 using MpiPoolSession to spawn MPI processes
[11/20/2025-23:56:51] [TRT-LLM] [V] Server [proxy_request_queue] bound to tcp://127.0.0.1:33681 in PAIR

[11/20/2025-23:56:51] [TRT-LLM] [I] Generating a new HMAC key for server proxy_request_queue
[11/20/2025-23:56:51] [TRT-LLM] [V] Server [worker_init_status_queue] bound to tcp://127.0.0.1:35295 in ROUTER

[11/20/2025-23:56:51] [TRT-LLM] [I] Generating a new HMAC key for server worker_init_status_queue
[11/20/2025-23:56:51] [TRT-LLM] [V] Server [proxy_result_queue] bound to tcp://127.0.0.1:45013 in PAIR

[11/20/2025-23:56:51] [TRT-LLM] [I] Generating a new HMAC key for server proxy_result_queue
[11/20/2025-23:56:51] [TRT-LLM] [V] Server [proxy_stats_queue] bound to tcp://127.0.0.1:42873 in PAIR

[11/20/2025-23:56:51] [TRT-LLM] [I] Generating a new HMAC key for server proxy_stats_queue
[11/20/2025-23:56:51] [TRT-LLM] [V] Server [proxy_kv_cache_events_queue] bound to tcp://127.0.0.1:33889 in PAIR

[11/20/2025-23:56:51] [TRT-LLM] [I] Generating a new HMAC key for server proxy_kv_cache_events_queue
[1763683011.281418] [8b3b04d4aa39:936 :0] ucp_context.c:2339 UCX WARN UCP API version is incompatible: required >= 1.20, actual 1.19.0 (loaded from /usr/local/ucx//lib/libucp.so.0)
[1763683011.281421] [8b3b04d4aa39:938 :0] ucp_context.c:2339 UCX WARN UCP API version is incompatible: required >= 1.20, actual 1.19.0 (loaded from /usr/local/ucx//lib/libucp.so.0)
[1763683011.281962] [8b3b04d4aa39:937 :0] ucp_context.c:2339 UCX WARN UCP API version is incompatible: required >= 1.20, actual 1.19.0 (loaded from /usr/local/ucx//lib/libucp.so.0)
[1763683011.282456] [8b3b04d4aa39:935 :0] ucp_context.c:2339 UCX WARN UCP API version is incompatible: required >= 1.20, actual 1.19.0 (loaded from /usr/local/ucx//lib/libucp.so.0)
[1763683011.724593] [8b3b04d4aa39:937 :0] ucp_context.c:2339 UCX WARN UCP API version is incompatible: required >= 1.20, actual 1.19.0 (loaded from /usr/local/ucx//lib/libucp.so.0)
[1763683011.724593] [8b3b04d4aa39:938 :0] ucp_context.c:2339 UCX WARN UCP API version is incompatible: required >= 1.20, actual 1.19.0 (loaded from /usr/local/ucx//lib/libucp.so.0)
[1763683011.724593] [8b3b04d4aa39:936 :0] ucp_context.c:2339 UCX WARN UCP API version is incompatible: required >= 1.20, actual 1.19.0 (loaded from /usr/local/ucx//lib/libucp.so.0)
[1763683011.724604] [8b3b04d4aa39:935 :0] ucp_context.c:2339 UCX WARN UCP API version is incompatible: required >= 1.20, actual 1.19.0 (loaded from /usr/local/ucx//lib/libucp.so.0)
[1763683011.740275] [8b3b04d4aa39:936 :0] ucp_context.c:2339 UCX WARN UCP API version is incompatible: required >= 1.20, actual 1.19.0 (loaded from /usr/local/ucx//lib/libucp.so.0)
[1763683011.740277] [8b3b04d4aa39:937 :0] ucp_context.c:2339 UCX WARN UCP API version is incompatible: required >= 1.20, actual 1.19.0 (loaded from /usr/local/ucx//lib/libucp.so.0)
[1763683011.740278] [8b3b04d4aa39:938 :0] ucp_context.c:2339 UCX WARN UCP API version is incompatible: required >= 1.20, actual 1.19.0 (loaded from /usr/local/ucx//lib/libucp.so.0)
[1763683011.740325] [8b3b04d4aa39:935 :0] ucp_context.c:2339 UCX WARN UCP API version is incompatible: required >= 1.20, actual 1.19.0 (loaded from /usr/local/ucx//lib/libucp.so.0)
/usr/local/lib/python3.12/dist-packages/torch/cuda/init.py:63: FutureWarning: The pynvml package is deprecated. Please install nvidia-ml-py instead. If you did not install pynvml directly, please report this to the maintainers of the package that installed pynvml for you.
import pynvml # type: ignore[import]
/usr/local/lib/python3.12/dist-packages/torch/cuda/init.py:63: FutureWarning: The pynvml package is deprecated. Please install nvidia-ml-py instead. If you did not install pynvml directly, please report this to the maintainers of the package that installed pynvml for you.
import pynvml # type: ignore[import]
/usr/local/lib/python3.12/dist-packages/torch/cuda/init.py:63: FutureWarning: The pynvml package is deprecated. Please install nvidia-ml-py instead. If you did not install pynvml directly, please report this to the maintainers of the package that installed pynvml for you.
import pynvml # type: ignore[import]
/usr/local/lib/python3.12/dist-packages/torch/cuda/init.py:63: FutureWarning: The pynvml package is deprecated. Please install nvidia-ml-py instead. If you did not install pynvml directly, please report this to the maintainers of the package that installed pynvml for you.
import pynvml # type: ignore[import]
Multiple distributions found for package optimum. Picked distribution: optimum
Multiple distributions found for package optimum. Picked distribution: optimum
Multiple distributions found for package optimum. Picked distribution: optimum
Multiple distributions found for package optimum. Picked distribution: optimum
[TensorRT-LLM] TensorRT LLM version: 1.2.0rc2
[TensorRT-LLM] TensorRT LLM version: 1.2.0rc2
[TensorRT-LLM] TensorRT LLM version: 1.2.0rc2
[TensorRT-LLM] TensorRT LLM version: 1.2.0rc2
ninja: no work to do.
/usr/local/lib/python3.12/dist-packages/tensorrt_llm/serve/openai_protocol.py:94: UserWarning: Field name "schema" in "ResponseFormat" shadows an attribute in parent "OpenAIBaseModel"
class ResponseFormat(OpenAIBaseModel):
/usr/local/lib/python3.12/dist-packages/tensorrt_llm/serve/openai_protocol.py:94: UserWarning: Field name "schema" in "ResponseFormat" shadows an attribute in parent "OpenAIBaseModel"
class ResponseFormat(OpenAIBaseModel):
/usr/local/lib/python3.12/dist-packages/tensorrt_llm/serve/openai_protocol.py:94: UserWarning: Field name "schema" in "ResponseFormat" shadows an attribute in parent "OpenAIBaseModel"
class ResponseFormat(OpenAIBaseModel):
/usr/local/lib/python3.12/dist-packages/tensorrt_llm/serve/openai_protocol.py:94: UserWarning: Field name "schema" in "ResponseFormat" shadows an attribute in parent "OpenAIBaseModel"
class ResponseFormat(OpenAIBaseModel):
[TensorRT-LLM][INFO] Refreshed the MPI local session
[TensorRT-LLM][INFO] Refreshed the MPI local session
[TensorRT-LLM][INFO] Refreshed the MPI local session
[TensorRT-LLM][INFO] Refreshed the MPI local session
[11/20/2025-23:57:04] [TRT-LLM] [V] Worker 0 ready to setup backend...

[TensorRT-LLM][INFO] Engine version 1.2.0rc2 found in the config file, assuming engine(s) built by new builder API.
[TensorRT-LLM][INFO] Engine version 1.2.0rc2 found in the config file, assuming engine(s) built by new builder API.
[TensorRT-LLM][INFO] Engine version 1.2.0rc2 found in the config file, assuming engine(s) built by new builder API.
[TensorRT-LLM][INFO] Engine version 1.2.0rc2 found in the config file, assuming engine(s) built by new builder API.
[TensorRT-LLM][INFO] Refreshed the MPI local session
[TensorRT-LLM][INFO] MPI size: 4, MPI local size: 4, rank: 2
[TensorRT-LLM][INFO] Refreshed the MPI local session
[TensorRT-LLM][INFO] MPI size: 4, MPI local size: 4, rank: 3
[TensorRT-LLM][INFO] Refreshed the MPI local session
[TensorRT-LLM][INFO] MPI size: 4, MPI local size: 4, rank: 1
[TensorRT-LLM][INFO] Refreshed the MPI local session
[TensorRT-LLM][INFO] MPI size: 4, MPI local size: 4, rank: 0
[TensorRT-LLM][INFO] Using user-specified devices: (0, 1, 2, 3)
[TensorRT-LLM][INFO] Rank 2 is using GPU 2
[TensorRT-LLM][INFO] Using user-specified devices: (0, 1, 2, 3)
[TensorRT-LLM][INFO] Rank 0 is using GPU 0
[TensorRT-LLM][INFO] Using user-specified devices: (0, 1, 2, 3)
[TensorRT-LLM][INFO] Rank 1 is using GPU 1
[TensorRT-LLM][INFO] Using user-specified devices: (0, 1, 2, 3)
[TensorRT-LLM][INFO] Rank 3 is using GPU 3
[TensorRT-LLM][INFO] TRTGptModel maxNumSequences: 2
[TensorRT-LLM][INFO] TRTGptModel maxBatchSize: 2
[TensorRT-LLM][INFO] TRTGptModel maxBeamWidth: 1
[TensorRT-LLM][INFO] TRTGptModel maxSequenceLen: 8192
[TensorRT-LLM][INFO] TRTGptModel maxDraftLen: 0
[TensorRT-LLM][INFO] TRTGptModel mMaxAttentionWindowSize: (8192) * 32
[TensorRT-LLM][INFO] TRTGptModel maxNumSequences: 2
[TensorRT-LLM][INFO] TRTGptModel maxBatchSize: 2
[TensorRT-LLM][INFO] TRTGptModel maxBeamWidth: 1
[TensorRT-LLM][INFO] TRTGptModel maxSequenceLen: 8192
[TensorRT-LLM][INFO] TRTGptModel maxDraftLen: 0
[TensorRT-LLM][INFO] TRTGptModel mMaxAttentionWindowSize: (8192) * 32
[TensorRT-LLM][INFO] TRTGptModel maxNumSequences: 2
[TensorRT-LLM][INFO] TRTGptModel maxBatchSize: 2
[TensorRT-LLM][INFO] TRTGptModel maxBeamWidth: 1
[TensorRT-LLM][INFO] TRTGptModel maxSequenceLen: 8192
[TensorRT-LLM][INFO] TRTGptModel maxDraftLen: 0
[TensorRT-LLM][INFO] TRTGptModel mMaxAttentionWindowSize: (8192) * 32
[TensorRT-LLM][INFO] TRTGptModel enableTrtOverlap: 0
[TensorRT-LLM][INFO] TRTGptModel normalizeLogProbs: 0
[TensorRT-LLM][INFO] TRTGptModel maxNumTokens: 8192
[TensorRT-LLM][INFO] TRTGptModel maxNumSequences: 2
[TensorRT-LLM][INFO] TRTGptModel maxBatchSize: 2
[TensorRT-LLM][INFO] TRTGptModel maxBeamWidth: 1
[TensorRT-LLM][INFO] TRTGptModel maxSequenceLen: 8192
[TensorRT-LLM][INFO] TRTGptModel maxDraftLen: 0
[TensorRT-LLM][INFO] TRTGptModel mMaxAttentionWindowSize: (8192) * 32
[TensorRT-LLM][INFO] TRTGptModel enableTrtOverlap: 0
[TensorRT-LLM][INFO] TRTGptModel normalizeLogProbs: 0
[TensorRT-LLM][INFO] TRTGptModel maxNumTokens: 8192
[TensorRT-LLM][INFO] TRTGptModel maxInputLen: 8191 = min(maxSequenceLen - 1, maxNumTokens) since context FMHA and usePackedInput are enabled
[TensorRT-LLM][INFO] TRTGptModel If model type is encoder, maxInputLen would be reset in trtEncoderModel to maxInputLen: min(maxSequenceLen, maxNumTokens).
[TensorRT-LLM][INFO] TRTGptModel enableTrtOverlap: 0
[TensorRT-LLM][INFO] TRTGptModel normalizeLogProbs: 0
[TensorRT-LLM][INFO] TRTGptModel maxNumTokens: 8192
[TensorRT-LLM][INFO] TRTGptModel maxInputLen: 8191 = min(maxSequenceLen - 1, maxNumTokens) since context FMHA and usePackedInput are enabled
[TensorRT-LLM][INFO] TRTGptModel If model type is encoder, maxInputLen would be reset in trtEncoderModel to maxInputLen: min(maxSequenceLen, maxNumTokens).
[TensorRT-LLM][INFO] Capacity Scheduler Policy: GUARANTEED_NO_EVICT
[TensorRT-LLM][INFO] Context Chunking Scheduler Policy: None
[TensorRT-LLM][INFO] TRTGptModel enableTrtOverlap: 0
[TensorRT-LLM][INFO] TRTGptModel normalizeLogProbs: 0
[TensorRT-LLM][INFO] TRTGptModel maxNumTokens: 8192
[TensorRT-LLM][INFO] TRTGptModel maxInputLen: 8191 = min(maxSequenceLen - 1, maxNumTokens) since context FMHA and usePackedInput are enabled
[TensorRT-LLM][INFO] TRTGptModel If model type is encoder, maxInputLen would be reset in trtEncoderModel to maxInputLen: min(maxSequenceLen, maxNumTokens).
[TensorRT-LLM][INFO] Capacity Scheduler Policy: GUARANTEED_NO_EVICT
[TensorRT-LLM][INFO] Context Chunking Scheduler Policy: None
[TensorRT-LLM][INFO] TRTGptModel maxInputLen: 8191 = min(maxSequenceLen - 1, maxNumTokens) since context FMHA and usePackedInput are enabled
[TensorRT-LLM][INFO] TRTGptModel If model type is encoder, maxInputLen would be reset in trtEncoderModel to maxInputLen: min(maxSequenceLen, maxNumTokens).
[TensorRT-LLM][INFO] Capacity Scheduler Policy: GUARANTEED_NO_EVICT
[TensorRT-LLM][INFO] Context Chunking Scheduler Policy: None
[TensorRT-LLM][INFO] Capacity Scheduler Policy: GUARANTEED_NO_EVICT
[TensorRT-LLM][INFO] Context Chunking Scheduler Policy: None
[TensorRT-LLM][INFO] Loaded engine size: 3906 MiB
[TensorRT-LLM][WARNING] Using an engine plan file across different models of devices is not supported and is likely to affect performance or even cause errors or deadlock.

Child job 2 terminated normally, but 1 process returned
a non-zero exit code. Per user-direction, the job has been aborted.

`

additional notes

GPU Processes had terminated as show in nvidia-smi.

`# nvidia-smi
Thu Nov 20 23:58:10 2025
+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 580.95.05 Driver Version: 580.95.05 CUDA Version: 13.0 |
+-----------------------------------------+------------------------+----------------------+
| GPU Name Persistence-M | Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap | Memory-Usage | GPU-Util Compute M. |
| | | MIG M. |
|=========================================+========================+======================|
| 0 NVIDIA L40S On | 00000000:38:00.0 Off | 0 |
| N/A 33C P0 80W / 350W | 455MiB / 46068MiB | 0% Default |
| | | N/A |
+-----------------------------------------+------------------------+----------------------+
| 1 NVIDIA L40S On | 00000000:3A:00.0 Off | 0 |
| N/A 27C P8 33W / 350W | 3MiB / 46068MiB | 0% Default |
| | | N/A |
+-----------------------------------------+------------------------+----------------------+
| 2 NVIDIA L40S On | 00000000:3C:00.0 Off | 0 |
| N/A 27C P8 31W / 350W | 3MiB / 46068MiB | 0% Default |
| | | N/A |
+-----------------------------------------+------------------------+----------------------+
| 3 NVIDIA L40S On | 00000000:3E:00.0 Off | 0 |
| N/A 28C P8 66W / 350W | 3MiB / 46068MiB | 0% Default |
| | | N/A |
+-----------------------------------------+------------------------+----------------------+

+-----------------------------------------------------------------------------------------+
| Processes: |
| GPU GI CI PID Type Process name GPU Memory |
| ID ID Usage |
|=========================================================================================|
| 0 N/A N/A 267 C /usr/bin/python 446MiB |
+-----------------------------------------------------------------------------------------+`

Before submitting a new issue...
  • Make sure you already searched for relevant issues, and checked the documentation and examples for answers to frequently asked questions.

Contributor guide

Open the contributing guide

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Assessment

This issue has not been assessed yet.

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.