[Bug]: graph.eliminate_dead_code() break the fx graph with enable_fi_allreduce_fusion when TP == 2 #23091
Description
Your current environment
The output of python collect_env.py
Collecting environment information...
==============================
System Info
==============================
OS : Ubuntu 24.04.2 LTS (x86_64)
GCC version : (Ubuntu 13.3.0-6ubuntu2~24.04) 13.3.0
Clang version : Could not collect
CMake version : version 3.24.0
Libc version : glibc-2.39
==============================
PyTorch Info
==============================
PyTorch version : 2.7.1+cu128
Is debug build : False
CUDA used to build PyTorch : 12.8
ROCM used to build PyTorch : N/A
==============================
Python Environment
==============================
Python version : 3.12.3 (main, Feb 4 2025, 14:48:35) [GCC 13.3.0] (64-bit runtime)
Python platform : Linux-6.8.0-57-generic-x86_64-with-glibc2.39
==============================
CUDA / GPU Info
==============================
Is CUDA available : True
CUDA runtime version : 12.9.86
CUDA_MODULE_LOADING set to : LAZY
GPU models and configuration :
GPU 0: NVIDIA B200
GPU 1: NVIDIA B200
Nvidia driver version : 580.65.01
cuDNN version : Probably one of the following:
/usr/lib/x86_64-linux-gnu/libcudnn.so.9.10.2
/usr/lib/x86_64-linux-gnu/libcudnn_adv.so.9.10.2
/usr/lib/x86_64-linux-gnu/libcudnn_cnn.so.9.10.2
/usr/lib/x86_64-linux-gnu/libcudnn_engines_precompiled.so.9.10.2
/usr/lib/x86_64-linux-gnu/libcudnn_engines_runtime_compiled.so.9.10.2
/usr/lib/x86_64-linux-gnu/libcudnn_graph.so.9.10.2
/usr/lib/x86_64-linux-gnu/libcudnn_heuristic.so.9.10.2
/usr/lib/x86_64-linux-gnu/libcudnn_ops.so.9.10.2
HIP runtime version : N/A
MIOpen runtime version : N/A
Is XNNPACK available : True
==============================
CPU Info
==============================
Architecture: x86_64
CPU op-mode(s): 32-bit, 64-bit
Address sizes: 52 bits physical, 57 bits virtual
Byte Order: Little Endian
CPU(s): 224
On-line CPU(s) list: 0-223
Vendor ID: GenuineIntel
Model name: INTEL(R) XEON(R) PLATINUM 8570
CPU family: 6
Model: 207
Thread(s) per core: 2
Core(s) per socket: 56
Socket(s): 2
Stepping: 2
CPU(s) scaling MHz: 28%
CPU max MHz: 4000.0000
CPU min MHz: 800.0000
BogoMIPS: 4200.00
Flags: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe syscall nx pdpe1gb rdtscp lm constant_tsc art arch_perfmon pebs bts rep_good nopl xtopology nonstop_tsc cpuid aperfmperf tsc_known_freq pni pclmulqdq dtes64 monitor ds_cpl vmx smx est tm2 ssse3 sdbg fma cx16 xtpr pdcm pcid dca sse4_1 sse4_2 x2apic movbe popcnt tsc_deadline_timer aes xsave avx f16c rdrand lahf_lm abm 3dnowprefetch cpuid_fault epb cat_l3 cat_l2 cdp_l3 intel_ppin cdp_l2 ssbd mba ibrs ibpb stibp ibrs_enhanced tpr_shadow flexpriority ept vpid ept_ad fsgsbase tsc_adjust bmi1 avx2 smep bmi2 erms invpcid cqm rdt_a avx512f avx512dq rdseed adx smap avx512ifma clflushopt clwb intel_pt avx512cd sha_ni avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_local split_lock_detect user_shstk avx_vnni avx512_bf16 wbnoinvd dtherm ida arat pln pts hwp hwp_act_window hwp_epp hwp_pkg_req vnmi avx512vbmi umip pku ospke waitpkg avx512_vbmi2 gfni vaes vpclmulqdq avx512_vnni avx512_bitalg tme avx512_vpopcntdq la57 rdpid bus_lock_detect cldemote movdiri movdir64b enqcmd fsrm md_clear serialize tsxldtrk pconfig arch_lbr ibt amx_bf16 avx512_fp16 amx_tile amx_int8 flush_l1d arch_capabilities
Virtualization: VT-x
L1d cache: 5.3 MiB (112 instances)
L1i cache: 3.5 MiB (112 instances)
L2 cache: 224 MiB (112 instances)
L3 cache: 600 MiB (2 instances)
NUMA node(s): 2
NUMA node0 CPU(s): 0-55,112-167
NUMA node1 CPU(s): 56-111,168-223
Vulnerability Gather data sampling: Not affected
Vulnerability Itlb multihit: Not affected
Vulnerability L1tf: Not affected
Vulnerability Mds: Not affected
Vulnerability Meltdown: Not affected
Vulnerability Mmio stale data: Not affected
Vulnerability Reg file data sampling: Not affected
Vulnerability Retbleed: Not affected
Vulnerability Spec rstack overflow: Not affected
Vulnerability Spec store bypass: Mitigation; Speculative Store Bypass disabled via prctl
Vulnerability Spectre v1: Mitigation; usercopy/swapgs barriers and __user pointer sanitization
Vulnerability Spectre v2: Mitigation; Enhanced / Automatic IBRS; IBPB conditional; RSB filling; PBRSB-eIBRS SW sequence; BHI BHI_DIS_S
Vulnerability Srbds: Not affected
Vulnerability Tsx async abort: Not affected
==============================
Versions of relevant libraries
==============================
[pip3] numpy==2.2.6
[pip3] nvidia-cublas-cu12==12.8.3.14
[pip3] nvidia-cuda-cupti-cu12==12.8.57
[pip3] nvidia-cuda-nvrtc-cu12==12.8.61
[pip3] nvidia-cuda-runtime-cu12==12.8.57
[pip3] nvidia-cudnn-cu12==9.7.1.26
[pip3] nvidia-cudnn-frontend==1.14.0
[pip3] nvidia-cufft-cu12==11.3.3.41
[pip3] nvidia-cufile-cu12==1.13.0.11
[pip3] nvidia-curand-cu12==10.3.9.55
[pip3] nvidia-cusolver-cu12==11.7.2.55
[pip3] nvidia-cusparse-cu12==12.5.7.53
[pip3] nvidia-cusparselt-cu12==0.6.3
[pip3] nvidia-ml-py==12.575.51
[pip3] nvidia-nccl-cu12==2.26.2
[pip3] nvidia-nvjitlink-cu12==12.8.61
[pip3] nvidia-nvtx-cu12==12.8.55
[pip3] pynvml==12.0.0
[pip3] pyzmq==27.0.1
[pip3] torch==2.7.1+cu128
[pip3] torchaudio==2.7.1+cu128
[pip3] torchvision==0.22.1+cu128
[pip3] transformers==4.55.2
[pip3] triton==3.3.1
[conda] Could not collect
==============================
vLLM Info
==============================
ROCM Version : Could not collect
Neuron SDK Version : N/A
vLLM Version : 0.10.2.dev6+g89657a557.d20250818 (git sha: 89657a557, date: 20250818)
vLLM Build Flags:
CUDA Archs: Not Set; ROCm: Disabled; Neuron: Disabled
GPU Topology:
�[4mGPU0 GPU1 NIC0 NIC1 NIC2 NIC3 NIC4 NIC5 NIC6 NIC7 NIC8 NIC9 NIC10 NIC11 NIC12 NIC13 NIC14 NIC15 CPU Affinity NUMA Affinity GPU NUMA ID�[0m
GPU0 X NV18 NODE NODE NODE NODE PXB NODE NODE NODE NODE NODE NODE NODE NODE NODE NODE NODE 0-27,112-139 0 N/A
GPU1 NV18 X NODE NODE NODE NODE NODE NODE NODE NODE NODE PXB NODE NODE NODE NODE NODE NODE 0-27,112-139 0 N/A
NIC0 NODE NODE X PIX PIX PIX NODE NODE NODE NODE NODE NODE NODE NODE NODE NODE NODE NODE
NIC1 NODE NODE PIX X PIX PIX NODE NODE NODE NODE NODE NODE NODE NODE NODE NODE NODE NODE
NIC2 NODE NODE PIX PIX X PIX NODE NODE NODE NODE NODE NODE NODE NODE NODE NODE NODE NODE
NIC3 NODE NODE PIX PIX PIX X NODE NODE NODE NODE NODE NODE NODE NODE NODE NODE NODE NODE
NIC4 PXB NODE NODE NODE NODE NODE X NODE NODE NODE NODE NODE NODE NODE NODE NODE NODE NODE
NIC5 NODE NODE NODE NODE NODE NODE NODE X PIX NODE NODE NODE NODE NODE NODE NODE NODE NODE
NIC6 NODE NODE NODE NODE NODE NODE NODE PIX X NODE NODE NODE NODE NODE NODE NODE NODE NODE
NIC7 NODE NODE NODE NODE NODE NODE NODE NODE NODE X NODE NODE NODE NODE NODE NODE NODE NODE
NIC8 NODE NODE NODE NODE NODE NODE NODE NODE NODE NODE X NODE NODE NODE NODE NODE NODE NODE
NIC9 NODE PXB NODE NODE NODE NODE NODE NODE NODE NODE NODE X NODE NODE NODE NODE NODE NODE
NIC10 NODE NODE NODE NODE NODE NODE NODE NODE NODE NODE NODE NODE X SYS SYS SYS SYS SYS
NIC11 NODE NODE NODE NODE NODE NODE NODE NODE NODE NODE NODE NODE SYS X PIX SYS SYS SYS
NIC12 NODE NODE NODE NODE NODE NODE NODE NODE NODE NODE NODE NODE SYS PIX X SYS SYS SYS
NIC13 NODE NODE NODE NODE NODE NODE NODE NODE NODE NODE NODE NODE SYS SYS SYS X SYS SYS
NIC14 NODE NODE NODE NODE NODE NODE NODE NODE NODE NODE NODE NODE SYS SYS SYS SYS X SYS
NIC15 NODE NODE NODE NODE NODE NODE NODE NODE NODE NODE NODE NODE SYS SYS SYS SYS SYS X
Legend:
X = Self
SYS = Connection traversing PCIe as well as the SMP interconnect between NUMA nodes (e.g., QPI/UPI)
NODE = Connection traversing PCIe as well as the interconnect between PCIe Host Bridges within a NUMA node
PHB = Connection traversing PCIe as well as a PCIe Host Bridge (typically the CPU)
PXB = Connection traversing multiple PCIe bridges (without traversing the PCIe Host Bridge)
PIX = Connection traversing at most a single PCIe bridge
NV# = Connection traversing a bonded set of # NVLinks
NIC Legend:
NIC0: mlx5_0
NIC1: mlx5_1
NIC2: mlx5_2
NIC3: mlx5_3
NIC4: mlx5_4
NIC5: mlx5_5
NIC6: mlx5_6
NIC7: mlx5_7
NIC8: mlx5_8
NIC9: mlx5_9
NIC10: mlx5_10
NIC11: mlx5_11
NIC12: mlx5_12
NIC13: mlx5_13
NIC14: mlx5_14
NIC15: mlx5_15
==============================
Environment Variables
==============================
NVIDIA_VISIBLE_DEVICES=0,3
CUBLAS_VERSION=12.9.1.4
NVIDIA_REQUIRE_CUDA=cuda>=9.0
NCCL_VERSION=2.27.3
NVIDIA_DRIVER_CAPABILITIES=compute,utility,video
CUDA_ARCH_LIST=7.5 8.0 8.6 9.0 10.0 12.0
NVIDIA_PRODUCT_NAME=NVIDIA TensorRT
CUDA_VERSION=12.9.1.010
CUBLASMP_VERSION=0.4.0.789
CUDNN_FRONTEND_VERSION=1.12.0
CUDNN_VERSION=9.10.2.21
LD_LIBRARY_PATH=/usr/local/cuda/compat/lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64
NVIDIA_BUILD_ID=177567330
CUDA_DRIVER_VERSION=575.57.08
NVIDIA_TENSORRT_VERSION=25.06
NVIDIA_REQUIRE_JETPACK_HOST_MOUNTS=
NCCL_CUMEM_ENABLE=0
PYTORCH_NVML_BASED_CUDA_CHECK=1
TORCHINDUCTOR_COMPILE_THREADS=1
CUDA_MODULE_LOADING=LAZY
🐛 Describe the bug
We are trying to add graph.eliminate_dead_code() to the end of all the passes before fix_functionalization. i.e. the code here.
However, when setting compilation-config to '{"custom_ops":["+rms_norm","+quant_fp8"],"pass_config":{"enable_fi_allreduce_fusion":true}}', we encountered a fx graph error. Provided the reproduction steps and error back trace. Note that the issue is only reproducible on TP2 or maybe larger TP.
VLLM_SKIP_P2P_CHECK=1 \
VLLM_ATTENTION_BACKEND=FLASHINFER \
\
vllm serve \
nvidia/Llama-3.3-70B-Instruct-FP8 \
--host 0.0.0.0 --port 8080 \
--tokenizer nvidia/Llama-3.3-70B-Instruct-FP8 \
--dtype auto \
--kv-cache-dtype fp8 \
--tensor-parallel-size 2 \
--pipeline-parallel-size 1 \
--swap-space 16 \
--max-num-seqs 512 \
--max-model-len 9216 \
--max-num-batched-tokens 8192 \
--gpu-memory-utilization 0.9 \
--compilation-config '{"custom_ops":["+rms_norm","+quant_fp8"],"pass_config":{"enable_fi_allreduce_fusion":true}}' \
--trust-remote-code \
--enable-chunked-prefill \
--no-enable-prefix-caching \
--load-format dummy
Error log:
ERROR 08-18 07:54:17 [multiproc_executor.py:596] WorkerProc hit an exception.
ERROR 08-18 07:54:17 [multiproc_executor.py:596] Traceback (most recent call last):
ERROR 08-18 07:54:17 [multiproc_executor.py:596] File "/workspace/vllm/vllm/v1/executor/multiproc_executor.py", line 591, in worker_busy_loop
ERROR 08-18 07:54:17 [multiproc_executor.py:596] output = func(*args, **kwargs)
ERROR 08-18 07:54:17 [multiproc_executor.py:596] ^^^^^^^^^^^^^^^^^^^^^
ERROR 08-18 07:54:17 [multiproc_executor.py:596] File "/workspace/.venv/lib/python3.12/site-packages/torch/utils/_contextlib.py", line 116, in decorate_context
ERROR 08-18 07:54:17 [multiproc_executor.py:596] return func(*args, **kwargs)
ERROR 08-18 07:54:17 [multiproc_executor.py:596] ^^^^^^^^^^^^^^^^^^^^^
ERROR 08-18 07:54:17 [multiproc_executor.py:596] File "/workspace/vllm/vllm/v1/worker/gpu_worker.py", line 244, in determine_available_memory
ERROR 08-18 07:54:17 [multiproc_executor.py:596] self.model_runner.profile_run()
ERROR 08-18 07:54:17 [multiproc_executor.py:596] File "/workspace/vllm/vllm/v1/worker/gpu_model_runner.py", line 2621, in profile_run
ERROR 08-18 07:54:17 [multiproc_executor.py:596] = self._dummy_run(self.max_num_tokens, is_profile=True)
ERROR 08-18 07:54:17 [multiproc_executor.py:596] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
ERROR 08-18 07:54:17 [multiproc_executor.py:596] File "/workspace/.venv/lib/python3.12/site-packages/torch/utils/_contextlib.py", line 116, in decorate_context
ERROR 08-18 07:54:17 [multiproc_executor.py:596] return func(*args, **kwargs)
ERROR 08-18 07:54:17 [multiproc_executor.py:596] ^^^^^^^^^^^^^^^^^^^^^
ERROR 08-18 07:54:17 [multiproc_executor.py:596] File "/workspace/vllm/vllm/v1/worker/gpu_model_runner.py", line 2394, in _dummy_run
ERROR 08-18 07:54:17 [multiproc_executor.py:596] outputs = self.model(
ERROR 08-18 07:54:17 [multiproc_executor.py:596] ^^^^^^^^^^^
ERROR 08-18 07:54:17 [multiproc_executor.py:596] File "/workspace/.venv/lib/python3.12/site-packages/torch/nn/modules/module.py", line 1751, in _wrapped_call_impl
ERROR 08-18 07:54:17 [multiproc_executor.py:596] return self._call_impl(*args, **kwargs)
ERROR 08-18 07:54:17 [multiproc_executor.py:596] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
ERROR 08-18 07:54:17 [multiproc_executor.py:596] File "/workspace/.venv/lib/python3.12/site-packages/torch/nn/modules/module.py", line 1762, in _call_impl
ERROR 08-18 07:54:17 [multiproc_executor.py:596] return forward_call(*args, **kwargs)
ERROR 08-18 07:54:17 [multiproc_executor.py:596] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
ERROR 08-18 07:54:17 [multiproc_executor.py:596] File "/workspace/vllm/vllm/model_executor/models/llama.py", line 577, in forward
ERROR 08-18 07:54:17 [multiproc_executor.py:596] model_output = self.model(input_ids, positions, intermediate_tensors,
ERROR 08-18 07:54:17 [multiproc_executor.py:596] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
ERROR 08-18 07:54:17 [multiproc_executor.py:596] File "/workspace/vllm/vllm/compilation/decorators.py", line 272, in __call__
ERROR 08-18 07:54:17 [multiproc_executor.py:596] output = self.compiled_callable(*args, **kwargs)
ERROR 08-18 07:54:17 [multiproc_executor.py:596] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
ERROR 08-18 07:54:17 [multiproc_executor.py:596] File "/workspace/.venv/lib/python3.12/site-packages/torch/_dynamo/eval_frame.py", line 663, in _fn
ERROR 08-18 07:54:17 [multiproc_executor.py:596] raise e.remove_dynamo_frames() from None # see TORCHDYNAMO_VERBOSE=1
ERROR 08-18 07:54:17 [multiproc_executor.py:596] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
ERROR 08-18 07:54:17 [multiproc_executor.py:596] File "/workspace/.venv/lib/python3.12/site-packages/torch/_inductor/compile_fx.py", line 760, in _compile_fx_inner
ERROR 08-18 07:54:17 [multiproc_executor.py:596] raise InductorError(e, currentframe()).with_traceback(
ERROR 08-18 07:54:17 [multiproc_executor.py:596] File "/workspace/.venv/lib/python3.12/site-packages/torch/_inductor/compile_fx.py", line 745, in _compile_fx_inner
ERROR 08-18 07:54:17 [multiproc_executor.py:596] mb_compiled_graph = fx_codegen_and_compile(
ERROR 08-18 07:54:17 [multiproc_executor.py:596] ^^^^^^^^^^^^^^^^^^^^^^^
ERROR 08-18 07:54:17 [multiproc_executor.py:596] File "/workspace/.venv/lib/python3.12/site-packages/torch/_inductor/compile_fx.py", line 1295, in fx_codegen_and_compile
ERROR 08-18 07:54:17 [multiproc_executor.py:596] return scheme.codegen_and_compile(gm, example_inputs, inputs_to_check, graph_kwargs)
ERROR 08-18 07:54:17 [multiproc_executor.py:596] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
ERROR 08-18 07:54:17 [multiproc_executor.py:596] File "/workspace/.venv/lib/python3.12/site-packages/torch/_inductor/compile_fx.py", line 1006, in codegen_and_compile
ERROR 08-18 07:54:17 [multiproc_executor.py:596] _recursive_post_grad_passes(gm, is_inference=is_inference)
ERROR 08-18 07:54:17 [multiproc_executor.py:596] File "/workspace/.venv/lib/python3.12/site-packages/torch/_inductor/compile_fx.py", line 379, in _recursive_post_grad_passes
ERROR 08-18 07:54:17 [multiproc_executor.py:596] post_grad_passes(gm, is_inference)
ERROR 08-18 07:54:17 [multiproc_executor.py:596] File "/workspace/.venv/lib/python3.12/site-packages/torch/_inductor/fx_passes/post_grad.py", line 177, in post_grad_passes
ERROR 08-18 07:54:17 [multiproc_executor.py:596] GraphTransformObserver(gm, "post_grad_custom_post_pass").apply_graph_pass(
ERROR 08-18 07:54:17 [multiproc_executor.py:596] File "/workspace/.venv/lib/python3.12/site-packages/torch/fx/passes/graph_transform_observer.py", line 85, in apply_graph_pass
ERROR 08-18 07:54:17 [multiproc_executor.py:596] return pass_fn(self.gm.graph)
ERROR 08-18 07:54:17 [multiproc_executor.py:596] ^^^^^^^^^^^^^^^^^^^^^^
ERROR 08-18 07:54:17 [multiproc_executor.py:596] File "/workspace/vllm/vllm/compilation/pass_manager.py", line 51, in __call__
ERROR 08-18 07:54:17 [multiproc_executor.py:596] graph.eliminate_dead_code()
ERROR 08-18 07:54:17 [multiproc_executor.py:596] File "/workspace/.venv/lib/python3.12/site-packages/torch/fx/graph.py", line 1833, in eliminate_dead_code
ERROR 08-18 07:54:17 [multiproc_executor.py:596] self.lint()
ERROR 08-18 07:54:17 [multiproc_executor.py:596] File "/workspace/.venv/lib/python3.12/site-packages/torch/fx/graph.py", line 1720, in lint
ERROR 08-18 07:54:17 [multiproc_executor.py:596] check_arg(arg, node)
ERROR 08-18 07:54:17 [multiproc_executor.py:596] File "/workspace/.venv/lib/python3.12/site-packages/torch/fx/graph.py", line 1705, in check_arg
ERROR 08-18 07:54:17 [multiproc_executor.py:596] raise RuntimeError(
ERROR 08-18 07:54:17 [multiproc_executor.py:596] torch._inductor.exc.InductorError: RuntimeError: Argument 'empty_1' of Node 'auto_functionalized_3' was used before it has been defined! Please check that Nodes in the graph are topologically ordered
Before submitting a new issue...
- Make sure you already searched for relevant issues, and asked the chatbot living at the bottom right corner of the documentation page, which can answer lots of frequently asked questions.