Files

T

devbadxyz 48a20990d3 Improve compilation support

2026-03-15 13:04:09 +01:00

68 KiB

Czysty Wina Historia

Issue Report: vLLM Tensor Parallelism over RDMA on AMD Strix Halo

✅ RESOLVED (Feb 2, 2026) This issue is SOLVED. The root cause was indeed missing gfx1150 support in the upstream RCCL library.

I have patched and built a custom version of RCCL with native gfx1150 support. This patched library is now included in the toolbox container provided by this repository (kyuz0/vllm-therock-gfx1150).

See the RDMA Cluster Setup Guide for instructions on how to run the cluster using the fixed container.

TL;DR

I am attempting to run vLLM with Tensor Parallelism across two AMD Strix Halo (Ryzen AI MAX+ 395) nodes connected directly via Intel E810-CQDA1 RDMA (RoCE v2).

Current Status: RDMA communication is verified (low latency ~5us). Ray cluster is operational and can allocate tensors on both nodes.
Blocker: vLLM fails with HIP error: invalid kernel file when initializing the distributed environment.
Suspected Cause: Possible missing support for gfx1150 in the RCCL library included with the ROCm nightly build.
Goal: Solicit troubleshooting advice or confirmation if gfx1150 support is indeed missing/required in RCCL.

Context & Goal
System Setup
Verification Steps
- 3.1 RDMA/RoCE Link Test
- 3.2 Ray Cluster & Torch Verification
The Issue: Invalid Kernel File
Request for Help

1. Context & Goal

The objective is to enable Tensor Parallelism with vLLM distributed across two machines.

Hardware: Two AMD Strix Halo dev kits (Ryzen AI MAX+ 395).
Interconnect: Direct 100GbE connection using Intel E810 cards configured for RoCE v2.
Software Stack: Fedora 43, ROCm v7.12 (Nightlies), vLLM (Source build), Ray.

I have established a stable low-latency RDMA link and a functional Ray cluster over that link. However, the actual vLLM engine initialization crashes.

2. System Setup

Hardware Configuration

Nodes: 2x AMD Strix Halo APUs (Framework Laptop / Dev Kit).
CPU/GPU: AMD Ryzen AI MAX+ 395 w/ Radeon 8060S (128GB Unified Memory via GTT).
Interconnect: Direct connection via Intel Ethernet Controller E810-CQDA1.
Protocol: RoCE v2 (RDMA over Converged Ethernet).

Host Software (Fedora)

Node	Hostname	Kernel	OS	IP (RDMA Interface)
Node 1	`frmwk-dsk`	`6.18.5-200.fc43.x86_64`	Fedora Linux 43	`192.168.100.1/30`
Node 2	`fw2`	`6.18.6-200.fc43.x86_64`	Fedora Linux 43	`192.168.100.2/30`

Container Environment (Toolbox)

Both nodes run identical Toolboxes based on Fedora 43 with TheRock ROCm Nightlies.

Container Runtime: Podman 5.7.1 / Toolbox 0.3.
RDMA Device: rocep194s0 (State: PORT_ACTIVE, Link: Ethernet).
ROCm Version: 7.12.0 (TheRock Nightly 20260131).
Python: 3.13.11
Compiler: GCC 15.2.1

Key Packages:

Package	Version	Source
vLLM	`0.16.0rc1.dev55+g8980001c9.d20260131.rocm712`	Source Build (Nightly)
PyTorch	`2.11.0a0+rocm7.12.0a20260131`	TheRock Nightly
Triton	`3.6.0+git447f725f.rocm7.12.0a20260131`	TheRock Nightly
RCCL	`2.27.7` (Built-in)	TheRock Nightly

Toolbox Creation Command

The environment is created using toolbox (wrapping Podman) with specific flags to expose the Infiniband devices and unlock memory limits for RDMA:

toolbox create vllm \
  --image docker.io/kyuz0/vllm-therock-gfx1150:latest \
  -- \
  --device /dev/dri \
  --device /dev/kfd \
  --device /dev/infiniband \
  --group-add video \
  --group-add render \
  --group-add rdma \
  --ulimit memlock=-1 \
  --security-opt seccomp=unconfined

The following Dockerfile is used for the toolbox: https://github.com/kyuz0/amd-strix-halo-vllm-toolboxes/blob/main/Dockerfile

3. Verification Steps

3.1. RDMA/RoCE Link Test

Running the compare_eth_vs_rdma.sh shows indeed that this is working fine:

kyuz0@frmwk-dsk:~$ ./compare_eth_vs_rdma.sh 

=== Network Comparison ===

Path                 Latency      Bandwidth   
------------------------------------------------
Ethernet (1G LAN)    0.074 ms     0.94 Gbps   
Ethernet (RoCE NIC)  0.068 ms     55.70 Gbps  
RDMA (RoCE)          5.23 us      50.64 Gbps

3.2. Ray Cluster & Torch Verification

Before attempting distributed inference with vLLM, we established a Ray cluster to verify three things:

Control plane communication over the RDMA interface (192.168.100.x).
Hardware visibility (GPU) on both nodes under Ray's management.
Basic PyTorch tensor allocation capabilities on the Strix Halo APUs.

Head Node Configuration (Node 1)

We explicitly bind Ray's internal communication (Gloo) and the NCCL socket discovery to the high-speed interface. Crucially, we set RAY_EXPERIMENTAL_NOSET_ROCR_VISIBLE_DEVICES=1 to prevent Ray from interfering with the APU visibility, which is a known issue on ROCm 6.x+ with integrated graphics.

ray stop --force

# Detect RDMA interface
export RDMA_IFACE=$(ip -o addr show to 192.168.100.0/24 | awk '{print $2}' | head -n1)

# Bind Ray/Torch communication
export GLOO_SOCKET_IFNAME=$RDMA_IFACE
export NCCL_SOCKET_IFNAME=$RDMA_IFACE

# Prevent Ray from masking the APU (Strix Halo Requirement)
export RAY_DISABLE_METRICS=1
export RAY_EXPERIMENTAL_NOSET_ROCR_VISIBLE_DEVICES=1

# Start Head
ray start --head --port=6379 --node-ip-address=192.168.100.1 --num-gpus=1 --num-cpus=8 --disable-usage-stats

Worker Node Configuration (Node 2)

The worker joins using the same environment settings.

ray stop --force

export RDMA_IFACE=$(ip -o addr show to 192.168.100.0/24 | awk '{print $2}' | head -n1)
export GLOO_SOCKET_IFNAME=$RDMA_IFACE
export NCCL_SOCKET_IFNAME=$RDMA_IFACE
export RAY_DISABLE_METRICS=1
export RAY_EXPERIMENTAL_NOSET_ROCR_VISIBLE_DEVICES=1

ray start --address='192.168.100.1:6379' --num-gpus=1 --num-cpus=8 --disable-usage-stats

Cluster Status

Running ray status confirms both nodes are active, providing a total of 2 GPUs (1 per APU) to the pool.

======== Autoscaler status: 2026-01-31 14:20:05.449661 ========
Node status
---------------------------------------------------------------
Active:
 1 node_5a3b1f77fbfa9166383761707a12bb40f0a03d4f635e26acf703b0cc
 1 node_709eb6806677f4171ba13b3fc6aee20da41d68f3ef22acbce6fce967
Pending:
 (no pending nodes)
Recent failures:
 (no failures)

Resources
---------------------------------------------------------------
Total Usage:
 0.0/16.0 CPU
 0.0/2.0 GPU
 0.0/1.0 TPU
 0B/167.19GiB memory
 0B/71.65GiB object_store_memory

From request_resources:
 (none)
Pending Demands:
 (no resource demands)

3.4. Functional Verification Script

We ran a python script to submit remote tasks to both nodes, ensuring they can initialize the ROCm driver and allocate tensors.

Script (verify_cluster.py):

import ray
import torch
import socket
import os

print("Connecting to Ray Cluster...")
ray.init(address="192.168.100.1:6379")

@ray.remote(num_gpus=1)
def check_node_gpu():
    host = socket.gethostname()
    ip = socket.gethostbyname(host)
    try:
        gpu_name = torch.cuda.get_device_name(0)
        # Perform a tiny allocation to prove the driver is accepting commands
        t = torch.tensor([1.0], device="cuda")
        return f"SUCCESS: Node {host} ({ip}) - {gpu_name} - Tensor: {t}"
    except Exception as e:
        return f"FAILURE: Node {host} ({ip}) - Error: {e}"

print("Submitting tasks to both nodes...")
results = ray.get([check_node_gpu.remote() for _ in range(2)])

print("\n--- Cluster Status ---")
for res in results:
    print(r"  " + res)

Output:

Connecting to Ray Cluster...
2026-01-31 13:51:16,452	INFO worker.py:1821 -- Connecting to existing Ray cluster at address: 192.168.100.1:6379...
2026-01-31 13:51:16,461	INFO worker.py:2007 -- Connected to Ray cluster.
...
--- Cluster Status ---
  SUCCESS: Node toolbx (192.168.1.128) - Radeon 8060S Graphics - Tensor: tensor([1.], device='cuda:0')
  SUCCESS: Node toolbx (192.168.1.127) - Radeon 8060S Graphics - Tensor: tensor([1.], device='cuda:0')

4. The Issue: Invalid Kernel File

Now that the cluster is setup and demonstrated to work, I try to run vLLM with tensor-parallelism like this:

(venv) ⬢ [kyuz0@toolbx ~]$ # 1. Launcher Environment (Including RCCL/RDMA vars)
export VLLM_HOST_IP=192.168.100.1
export NCCL_SOCKET_IFNAME=$(ip -o addr show to 192.168.100.0/24 | awk '{print $2}' | head -n1)
export NCCL_IB_GID_INDEX=1
export NCCL_IB_DISABLE=0
export NCCL_NET_GDR_LEVEL=0
export RAY_EXPERIMENTAL_NOSET_ROCR_VISIBLE_DEVICES=1

# 2. Launch
vllm serve facebook/opt-125m \
  --tensor-parallel-size 2 \
  --distributed-executor-backend ray \
  --trust-remote-code \
  --enforce-eager

4.2. Error Logs

This results in an HIP error: invalid kernel file immediately upon engine initialization:

(APIServer pid=17743) INFO 01-31 14:20:54 [utils.py:346] 
(APIServer pid=17743) INFO 01-31 14:20:54 [utils.py:346]        █     █     █▄   ▄█
(APIServer pid=17743) INFO 01-31 14:20:54 [utils.py:346]  ▄▄ ▄█ █     █     █ ▀▄▀ █  version 0.16.0rc1.dev55+g8980001c9.d20260131
(APIServer pid=17743) INFO 01-31 14:20:54 [utils.py:346]   █▄█▀ █     █     █     █  model   facebook/opt-125m
(APIServer pid=17743) INFO 01-31 14:20:54 [utils.py:346]    ▀▀  ▀▀▀▀▀ ▀▀▀▀▀ ▀     ▀
(APIServer pid=17743) INFO 01-31 14:20:54 [utils.py:346] 
(APIServer pid=17743) INFO 01-31 14:20:54 [utils.py:282] non-default args: {'model_tag': 'facebook/opt-125m', 'api_server_count': 1, 'model': 'facebook/opt-125m', 'trust_remote_code': True, 'enforce_eager': True, 'distributed_executor_backend': 'ray', 'tensor_parallel_size': 2}
(APIServer pid=17743) The argument `trust_remote_code` is to be used with Auto classes. It has no effect here and is ignored.
(APIServer pid=17743) The argument `trust_remote_code` is to be used with Auto classes. It has no effect here and is ignored.
(APIServer pid=17743) INFO 01-31 14:20:55 [model.py:541] Resolved architecture: OPTForCausalLM
(APIServer pid=17743) INFO 01-31 14:20:55 [model.py:1559] Using max model len 2048
(APIServer pid=17743) INFO 01-31 14:20:55 [scheduler.py:226] Chunked prefill is enabled with max_num_batched_tokens=8192.
(APIServer pid=17743) WARNING 01-31 14:20:55 [vllm.py:651] Async scheduling will be disabled because it is not supported with the `ray` distributed executor backend (only `mp`, `uni`, and `external_launcher` are supported).
(APIServer pid=17743) INFO 01-31 14:20:55 [vllm.py:669] Asynchronous scheduling is disabled.
(APIServer pid=17743) WARNING 01-31 14:20:55 [vllm.py:707] Enforce eager set, overriding optimization level to -O0
(APIServer pid=17743) INFO 01-31 14:20:55 [vllm.py:807] Cudagraph is disabled under eager mode
(EngineCore_DP0 pid=17822) INFO 01-31 14:20:59 [core.py:96] Initializing a V1 LLM engine (v0.16.0rc1.dev55+g8980001c9.d20260131) with config: model='facebook/opt-125m', speculative_config=None, tokenizer='facebook/opt-125m', skip_tokenizer_init=False, tokenizer_mode=auto, revision=None, tokenizer_revision=None, trust_remote_code=True, dtype=torch.float16, max_seq_len=2048, download_dir=None, load_format=auto, tensor_parallel_size=2, pipeline_parallel_size=1, data_parallel_size=1, disable_custom_all_reduce=True, quantization=None, enforce_eager=True, enable_return_routed_experts=False, kv_cache_dtype=auto, device_config=cuda, structured_outputs_config=StructuredOutputsConfig(backend='auto', disable_fallback=False, disable_any_whitespace=False, disable_additional_properties=False, reasoning_parser='', reasoning_parser_plugin='', enable_in_reasoning=False), observability_config=ObservabilityConfig(show_hidden_metrics_for_version=None, otlp_traces_endpoint=None, collect_detailed_traces=None, kv_cache_metrics=False, kv_cache_metrics_sample=0.01, cudagraph_metrics=False, enable_layerwise_nvtx_tracing=False, enable_mfu_metrics=False, enable_mm_processor_stats=False, enable_logging_iteration_details=False), seed=0, served_model_name=facebook/opt-125m, enable_prefix_caching=True, enable_chunked_prefill=True, pooler_config=None, compilation_config={'level': None, 'mode': <CompilationMode.NONE: 0>, 'debug_dump_path': None, 'cache_dir': '', 'compile_cache_save_format': 'binary', 'backend': 'inductor', 'custom_ops': ['all', '+sparse_attn_indexer'], 'splitting_ops': [], 'compile_mm_encoder': False, 'compile_sizes': [], 'compile_ranges_split_points': [8192], 'inductor_compile_config': {'enable_auto_functionalized_v2': False, 'combo_kernels': True, 'benchmark_combo_kernel': True}, 'inductor_passes': {}, 'cudagraph_mode': <CUDAGraphMode.NONE: 0>, 'cudagraph_num_of_warmups': 0, 'cudagraph_capture_sizes': [], 'cudagraph_copy_inputs': False, 'cudagraph_specialize_lora': True, 'use_inductor_graph_partition': False, 'pass_config': {'fuse_norm_quant': False, 'fuse_act_quant': False, 'fuse_attn_quant': False, 'eliminate_noops': False, 'enable_sp': False, 'fuse_gemm_comms': False, 'fuse_allreduce_rms': False, 'fuse_act_padding': False}, 'max_cudagraph_capture_size': 0, 'dynamic_shapes_config': {'type': <DynamicShapesType.BACKED: 'backed'>, 'evaluate_guards': False, 'assume_32_bit_indexing': False}, 'local_cache_dir': None, 'static_all_moe_layers': []}
(EngineCore_DP0 pid=17822) 2026-01-31 14:20:59,483	INFO worker.py:1821 -- Connecting to existing Ray cluster at address: 192.168.100.1:6379...
(EngineCore_DP0 pid=17822) 2026-01-31 14:20:59,492	INFO worker.py:2007 -- Connected to Ray cluster.
(EngineCore_DP0 pid=17822) /opt/venv/lib64/python3.13/site-packages/ray/_private/worker.py:2046: FutureWarning: Tip: In future versions of Ray, Ray will no longer override accelerator visible devices env var if num_gpus=0 or num_gpus=None (default). To enable this behavior and turn off this error message, set RAY_ACCEL_ENV_VAR_OVERRIDE_ON_ZERO=0
(EngineCore_DP0 pid=17822)   warnings.warn(
(EngineCore_DP0 pid=17822) INFO 01-31 14:20:59 [ray_utils.py:402] No current placement group found. Creating a new placement group.
(EngineCore_DP0 pid=17822) WARNING 01-31 14:20:59 [ray_utils.py:213] tensor_parallel_size=2 is bigger than a reserved number of GPUs (1 GPUs) in a node 5a3b1f77fbfa9166383761707a12bb40f0a03d4f635e26acf703b0cc. Tensor parallel workers can be spread out to 2+ nodes which can degrade the performance unless you have fast interconnect across nodes, like Infiniband. To resolve this issue, make sure you have more than 2 GPUs available at each node.
(EngineCore_DP0 pid=17822) WARNING 01-31 14:20:59 [ray_utils.py:213] tensor_parallel_size=2 is bigger than a reserved number of GPUs (1 GPUs) in a node 709eb6806677f4171ba13b3fc6aee20da41d68f3ef22acbce6fce967. Tensor parallel workers can be spread out to 2+ nodes which can degrade the performance unless you have fast interconnect across nodes, like Infiniband. To resolve this issue, make sure you have more than 2 GPUs available at each node.
(EngineCore_DP0 pid=17822) INFO 01-31 14:21:02 [ray_env.py:66] RAY_NON_CARRY_OVER_ENV_VARS from config: set()
(EngineCore_DP0 pid=17822) INFO 01-31 14:21:02 [ray_env.py:69] Copying the following environment variables to workers: ['VLLM_ROCM_USE_AITER', 'VLLM_WORKER_MULTIPROC_METHOD', 'MAX_JOBS', 'LD_LIBRARY_PATH', 'VLLM_USE_TRITON_AWQ', 'VLLM_TARGET_DEVICE', 'VLLM_ROCM_USE_AITER_MOE']
(EngineCore_DP0 pid=17822) INFO 01-31 14:21:02 [ray_env.py:74] If certain env vars should NOT be copied, add them to /home/kyuz0/.config/vllm/ray_non_carry_over_env_vars.json file
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=9409, ip=192.168.100.2) WARNING 01-31 14:21:03 [worker_base.py:301] Missing `shared_worker_lock` argument from executor. This argument is needed for mm_processor_cache_type='shm'.
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=17931) INFO 01-31 14:21:03 [parallel_state.py:1234] world_size=2 rank=0 local_rank=0 distributed_init_method=tcp://192.168.100.1:49483 backend=nccl
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=17931) INFO 01-31 14:21:03 [pynccl.py:111] vLLM is using nccl==2.27.7
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946] EngineCore failed to start.
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946] Traceback (most recent call last):
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]   File "/opt/venv/lib64/python3.13/site-packages/vllm/v1/engine/core.py", line 937, in run_engine_core
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]     engine_core = EngineCoreProc(*args, engine_index=dp_rank, **kwargs)
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]   File "/opt/venv/lib64/python3.13/site-packages/vllm/v1/engine/core.py", line 691, in __init__
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]     super().__init__(
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]     ~~~~~~~~~~~~~~~~^
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]         vllm_config,
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]         ^^^^^^^^^^^^
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]     ...<3 lines>...
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]         internal_dp_balancing,
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]         ^^^^^^^^^^^^^^^^^^^^^^
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]     )
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]     ^
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]   File "/opt/venv/lib64/python3.13/site-packages/vllm/v1/engine/core.py", line 105, in __init__
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]     self.model_executor = executor_class(vllm_config)
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]                           ~~~~~~~~~~~~~~^^^^^^^^^^^^^
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]   File "/opt/venv/lib64/python3.13/site-packages/vllm/v1/executor/abstract.py", line 101, in __init__
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]     self._init_executor()
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]     ~~~~~~~~~~~~~~~~~~~^^
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]   File "/opt/venv/lib64/python3.13/site-packages/vllm/v1/executor/ray_executor.py", line 97, in _init_executor
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]     self._init_workers_ray(placement_group)
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]     ~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]   File "/opt/venv/lib64/python3.13/site-packages/vllm/v1/executor/ray_executor.py", line 366, in _init_workers_ray
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]     self.collective_rpc("init_device")
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]     ~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]   File "/opt/venv/lib64/python3.13/site-packages/vllm/v1/executor/ray_executor.py", line 489, in collective_rpc
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]     return ray.get(ray_worker_outputs, timeout=timeout)
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]            ~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]   File "/opt/venv/lib64/python3.13/site-packages/ray/_private/auto_init_hook.py", line 22, in auto_init_wrapper
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]     return fn(*args, **kwargs)
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]   File "/opt/venv/lib64/python3.13/site-packages/ray/_private/client_mode_hook.py", line 104, in wrapper
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]     return func(*args, **kwargs)
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]   File "/opt/venv/lib64/python3.13/site-packages/ray/_private/worker.py", line 2967, in get
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]     values, debugger_breakpoint = worker.get_objects(
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]                                   ~~~~~~~~~~~~~~~~~~^
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]         object_refs, timeout=timeout, _tensor_transport=_tensor_transport
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]     )
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]     ^
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]   File "/opt/venv/lib64/python3.13/site-packages/ray/_private/worker.py", line 1015, in get_objects
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]     raise value.as_instanceof_cause()
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946] ray.exceptions.RayTaskError(AcceleratorError): ray::RayWorkerWrapper.execute_method() (pid=17931, ip=192.168.100.1, actor_id=e2ed7f9c6fbd6274f30c736001000000, repr=<vllm.v1.executor.ray_utils.RayWorkerWrapper object at 0x7ff79ca5c590>)
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]   File "/opt/venv/lib64/python3.13/site-packages/vllm/v1/worker/worker_base.py", line 345, in execute_method
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]     raise e
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]   File "/opt/venv/lib64/python3.13/site-packages/vllm/v1/worker/worker_base.py", line 334, in execute_method
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]     return run_method(self, method, args, kwargs)
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]   File "/opt/venv/lib64/python3.13/site-packages/vllm/v1/serial_utils.py", line 459, in run_method
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]     return func(*args, **kwargs)
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]   File "/opt/venv/lib64/python3.13/site-packages/vllm/v1/worker/worker_base.py", line 326, in init_device
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]     self.worker.init_device()  # type: ignore
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]     ~~~~~~~~~~~~~~~~~~~~~~~^^
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]   File "/opt/venv/lib64/python3.13/site-packages/vllm/v1/worker/gpu_worker.py", line 221, in init_device
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]     init_worker_distributed_environment(
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]         self.vllm_config,
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]         ^^^^^^^^^^^^^^^^^
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]     ...<3 lines>...
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]         current_platform.dist_backend,
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]     )
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]     ^
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]   File "/opt/venv/lib64/python3.13/site-packages/vllm/v1/worker/gpu_worker.py", line 959, in init_worker_distributed_environment
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]     ensure_model_parallel_initialized(
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]         parallel_config.tensor_parallel_size,
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]     ...<2 lines>...
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]         parallel_config.decode_context_parallel_size,
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]     )
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]     ^
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]   File "/opt/venv/lib64/python3.13/site-packages/vllm/distributed/parallel_state.py", line 1472, in ensure_model_parallel_initialized
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]     initialize_model_parallel(
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]     ~~~~~~~~~~~~~~~~~~~~~~~~~^
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]         tensor_model_parallel_size,
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]         ^^^^^^^^^^^^^^^^^^^^^^^^^^^
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]     ...<3 lines>...
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]         backend,
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]         ^^^^^^^^
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]     )
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]     ^
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]   File "/opt/venv/lib64/python3.13/site-packages/vllm/distributed/parallel_state.py", line 1369, in initialize_model_parallel
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]     _TP = init_model_parallel_group(
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]         group_ranks,
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]     ...<3 lines>...
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]         group_name="tp",
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]     )
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]   File "/opt/venv/lib64/python3.13/site-packages/vllm/distributed/parallel_state.py", line 1089, in init_model_parallel_group
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]     return GroupCoordinator(
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]         group_ranks=group_ranks,
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]     ...<4 lines>...
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]         group_name=group_name,
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]     )
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]   File "/opt/venv/lib64/python3.13/site-packages/vllm/distributed/parallel_state.py", line 362, in __init__
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]     self.device_communicator = device_comm_cls(
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]                                ~~~~~~~~~~~~~~~^
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]         cpu_group=self.cpu_group,
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]         ^^^^^^^^^^^^^^^^^^^^^^^^^
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]     ...<2 lines>...
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]         unique_name=self.unique_name,
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]     )
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]     ^
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]   File "/opt/venv/lib64/python3.13/site-packages/vllm/distributed/device_communicators/cuda_communicator.py", line 58, in __init__
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]     self.pynccl_comm = PyNcclCommunicator(
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]                        ~~~~~~~~~~~~~~~~~~^
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]         group=self.cpu_group,
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]         ^^^^^^^^^^^^^^^^^^^^^
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]         device=self.device,
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]         ^^^^^^^^^^^^^^^^^^^
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]     )
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]     ^
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]   File "/opt/venv/lib64/python3.13/site-packages/vllm/distributed/device_communicators/pynccl.py", line 145, in __init__
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946]     data = torch.zeros(1, device=device)
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946] torch.AcceleratorError: HIP error: invalid kernel file
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946] Search for `hipErrorInvalidKernelFile' in https://rocm.docs.amd.com/projects/HIP/en/latest/index.html for more information.
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946] HIP kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect.
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946] For debugging consider passing AMD_SERIALIZE_KERNEL=3
(EngineCore_DP0 pid=17822) ERROR 01-31 14:21:03 [core.py:946] Device-side assertion tracking was not enabled by user.
(EngineCore_DP0 pid=17822) Process EngineCore_DP0:
(EngineCore_DP0 pid=17822) Traceback (most recent call last):
(EngineCore_DP0 pid=17822)   File "/usr/lib64/python3.13/multiprocessing/process.py", line 313, in _bootstrap
(EngineCore_DP0 pid=17822)     self.run()
(EngineCore_DP0 pid=17822)     ~~~~~~~~^^
(EngineCore_DP0 pid=17822)   File "/usr/lib64/python3.13/multiprocessing/process.py", line 108, in run
(EngineCore_DP0 pid=17822)     self._target(*self._args, **self._kwargs)
(EngineCore_DP0 pid=17822)     ~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(EngineCore_DP0 pid=17822)   File "/opt/venv/lib64/python3.13/site-packages/vllm/v1/engine/core.py", line 950, in run_engine_core
(EngineCore_DP0 pid=17822)     raise e
(EngineCore_DP0 pid=17822)   File "/opt/venv/lib64/python3.13/site-packages/vllm/v1/engine/core.py", line 937, in run_engine_core
(EngineCore_DP0 pid=17822)     engine_core = EngineCoreProc(*args, engine_index=dp_rank, **kwargs)
(EngineCore_DP0 pid=17822)   File "/opt/venv/lib64/python3.13/site-packages/vllm/v1/engine/core.py", line 691, in __init__
(EngineCore_DP0 pid=17822)     super().__init__(
(EngineCore_DP0 pid=17822)     ~~~~~~~~~~~~~~~~^
(EngineCore_DP0 pid=17822)         vllm_config,
(EngineCore_DP0 pid=17822)         ^^^^^^^^^^^^
(EngineCore_DP0 pid=17822)     ...<3 lines>...
(EngineCore_DP0 pid=17822)         internal_dp_balancing,
(EngineCore_DP0 pid=17822)         ^^^^^^^^^^^^^^^^^^^^^^
(EngineCore_DP0 pid=17822)     )
(EngineCore_DP0 pid=17822)     ^
(EngineCore_DP0 pid=17822)   File "/opt/venv/lib64/python3.13/site-packages/vllm/v1/engine/core.py", line 105, in __init__
(EngineCore_DP0 pid=17822)     self.model_executor = executor_class(vllm_config)
(EngineCore_DP0 pid=17822)                           ~~~~~~~~~~~~~~^^^^^^^^^^^^^
(EngineCore_DP0 pid=17822)   File "/opt/venv/lib64/python3.13/site-packages/vllm/v1/executor/abstract.py", line 101, in __init__
(EngineCore_DP0 pid=17822)     self._init_executor()
(EngineCore_DP0 pid=17822)     ~~~~~~~~~~~~~~~~~~~^^
(EngineCore_DP0 pid=17822)   File "/opt/venv/lib64/python3.13/site-packages/vllm/v1/executor/ray_executor.py", line 97, in _init_executor
(EngineCore_DP0 pid=17822)     self._init_workers_ray(placement_group)
(EngineCore_DP0 pid=17822)     ~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^
(EngineCore_DP0 pid=17822)   File "/opt/venv/lib64/python3.13/site-packages/vllm/v1/executor/ray_executor.py", line 366, in _init_workers_ray
(EngineCore_DP0 pid=17822)     self.collective_rpc("init_device")
(EngineCore_DP0 pid=17822)     ~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^
(EngineCore_DP0 pid=17822)   File "/opt/venv/lib64/python3.13/site-packages/vllm/v1/executor/ray_executor.py", line 489, in collective_rpc
(EngineCore_DP0 pid=17822)     return ray.get(ray_worker_outputs, timeout=timeout)
(EngineCore_DP0 pid=17822)            ~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(EngineCore_DP0 pid=17822)   File "/opt/venv/lib64/python3.13/site-packages/ray/_private/auto_init_hook.py", line 22, in auto_init_wrapper
(EngineCore_DP0 pid=17822)     return fn(*args, **kwargs)
(EngineCore_DP0 pid=17822)   File "/opt/venv/lib64/python3.13/site-packages/ray/_private/client_mode_hook.py", line 104, in wrapper
(EngineCore_DP0 pid=17822)     return func(*args, **kwargs)
(EngineCore_DP0 pid=17822)   File "/opt/venv/lib64/python3.13/site-packages/ray/_private/worker.py", line 2967, in get
(EngineCore_DP0 pid=17822)     values, debugger_breakpoint = worker.get_objects(
(EngineCore_DP0 pid=17822)                                   ~~~~~~~~~~~~~~~~~~^
(EngineCore_DP0 pid=17822)         object_refs, timeout=timeout, _tensor_transport=_tensor_transport
(EngineCore_DP0 pid=17822)         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(EngineCore_DP0 pid=17822)     )
(EngineCore_DP0 pid=17822)     ^
(EngineCore_DP0 pid=17822)   File "/opt/venv/lib64/python3.13/site-packages/ray/_private/worker.py", line 1015, in get_objects
(EngineCore_DP0 pid=17822)     raise value.as_instanceof_cause()
(EngineCore_DP0 pid=17822) ray.exceptions.RayTaskError(AcceleratorError): ray::RayWorkerWrapper.execute_method() (pid=17931, ip=192.168.100.1, actor_id=e2ed7f9c6fbd6274f30c736001000000, repr=<vllm.v1.executor.ray_utils.RayWorkerWrapper object at 0x7ff79ca5c590>)
(EngineCore_DP0 pid=17822)   File "/opt/venv/lib64/python3.13/site-packages/vllm/v1/worker/worker_base.py", line 345, in execute_method
(EngineCore_DP0 pid=17822)     raise e
(EngineCore_DP0 pid=17822)   File "/opt/venv/lib64/python3.13/site-packages/vllm/v1/worker/worker_base.py", line 334, in execute_method
(EngineCore_DP0 pid=17822)     return run_method(self, method, args, kwargs)
(EngineCore_DP0 pid=17822)   File "/opt/venv/lib64/python3.13/site-packages/vllm/v1/serial_utils.py", line 459, in run_method
(EngineCore_DP0 pid=17822)     return func(*args, **kwargs)
(EngineCore_DP0 pid=17822)   File "/opt/venv/lib64/python3.13/site-packages/vllm/v1/worker/worker_base.py", line 326, in init_device
(EngineCore_DP0 pid=17822)     self.worker.init_device()  # type: ignore
(EngineCore_DP0 pid=17822)     ~~~~~~~~~~~~~~~~~~~~~~~^^
(EngineCore_DP0 pid=17822)   File "/opt/venv/lib64/python3.13/site-packages/vllm/v1/worker/gpu_worker.py", line 221, in init_device
(EngineCore_DP0 pid=17822)     init_worker_distributed_environment(
(EngineCore_DP0 pid=17822)     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^
(EngineCore_DP0 pid=17822)         self.vllm_config,
(EngineCore_DP0 pid=17822)         ^^^^^^^^^^^^^^^^^
(EngineCore_DP0 pid=17822)     ...<3 lines>...
(EngineCore_DP0 pid=17822)         current_platform.dist_backend,
(EngineCore_DP0 pid=17822)         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(EngineCore_DP0 pid=17822)     )
(EngineCore_DP0 pid=17822)     ^
(EngineCore_DP0 pid=17822)   File "/opt/venv/lib64/python3.13/site-packages/vllm/v1/worker/gpu_worker.py", line 959, in init_worker_distributed_environment
(EngineCore_DP0 pid=17822)     ensure_model_parallel_initialized(
(EngineCore_DP0 pid=17822)     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^
(EngineCore_DP0 pid=17822)         parallel_config.tensor_parallel_size,
(EngineCore_DP0 pid=17822)         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(EngineCore_DP0 pid=17822)     ...<2 lines>...
(EngineCore_DP0 pid=17822)         parallel_config.decode_context_parallel_size,
(EngineCore_DP0 pid=17822)         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(EngineCore_DP0 pid=17822)     )
(EngineCore_DP0 pid=17822)     ^
(EngineCore_DP0 pid=17822)   File "/opt/venv/lib64/python3.13/site-packages/vllm/distributed/parallel_state.py", line 1472, in ensure_model_parallel_initialized
(EngineCore_DP0 pid=17822)     initialize_model_parallel(
(EngineCore_DP0 pid=17822)     ~~~~~~~~~~~~~~~~~~~~~~~~~^
(EngineCore_DP0 pid=17822)         tensor_model_parallel_size,
(EngineCore_DP0 pid=17822)         ^^^^^^^^^^^^^^^^^^^^^^^^^^^
(EngineCore_DP0 pid=17822)     ...<3 lines>...
(EngineCore_DP0 pid=17822)         backend,
(EngineCore_DP0 pid=17822)         ^^^^^^^^
(EngineCore_DP0 pid=17822)     )
(EngineCore_DP0 pid=17822)     ^
(EngineCore_DP0 pid=17822)   File "/opt/venv/lib64/python3.13/site-packages/vllm/distributed/parallel_state.py", line 1369, in initialize_model_parallel
(EngineCore_DP0 pid=17822)     _TP = init_model_parallel_group(
(EngineCore_DP0 pid=17822)         group_ranks,
(EngineCore_DP0 pid=17822)     ...<3 lines>...
(EngineCore_DP0 pid=17822)         group_name="tp",
(EngineCore_DP0 pid=17822)     )
(EngineCore_DP0 pid=17822)   File "/opt/venv/lib64/python3.13/site-packages/vllm/distributed/parallel_state.py", line 1089, in init_model_parallel_group
(EngineCore_DP0 pid=17822)     return GroupCoordinator(
(EngineCore_DP0 pid=17822)         group_ranks=group_ranks,
(EngineCore_DP0 pid=17822)     ...<4 lines>...
(EngineCore_DP0 pid=17822)         group_name=group_name,
(EngineCore_DP0 pid=17822)     )
(EngineCore_DP0 pid=17822)   File "/opt/venv/lib64/python3.13/site-packages/vllm/distributed/parallel_state.py", line 362, in __init__
(EngineCore_DP0 pid=17822)     self.device_communicator = device_comm_cls(
(EngineCore_DP0 pid=17822)                                ~~~~~~~~~~~~~~~^
(EngineCore_DP0 pid=17822)         cpu_group=self.cpu_group,
(EngineCore_DP0 pid=17822)         ^^^^^^^^^^^^^^^^^^^^^^^^^
(EngineCore_DP0 pid=17822)     ...<2 lines>...
(EngineCore_DP0 pid=17822)         unique_name=self.unique_name,
(EngineCore_DP0 pid=17822)         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(EngineCore_DP0 pid=17822)     )
(EngineCore_DP0 pid=17822)     ^
(EngineCore_DP0 pid=17822)   File "/opt/venv/lib64/python3.13/site-packages/vllm/distributed/device_communicators/cuda_communicator.py", line 58, in __init__
(EngineCore_DP0 pid=17822)     self.pynccl_comm = PyNcclCommunicator(
(EngineCore_DP0 pid=17822)                        ~~~~~~~~~~~~~~~~~~^
(EngineCore_DP0 pid=17822)         group=self.cpu_group,
(EngineCore_DP0 pid=17822)         ^^^^^^^^^^^^^^^^^^^^^
(EngineCore_DP0 pid=17822)         device=self.device,
(EngineCore_DP0 pid=17822)         ^^^^^^^^^^^^^^^^^^^
(EngineCore_DP0 pid=17822)     )
(EngineCore_DP0 pid=17822)     ^
(EngineCore_DP0 pid=17822)   File "/opt/venv/lib64/python3.13/site-packages/vllm/distributed/device_communicators/pynccl.py", line 145, in __init__
(EngineCore_DP0 pid=17822)     data = torch.zeros(1, device=device)
(EngineCore_DP0 pid=17822) torch.AcceleratorError: HIP error: invalid kernel file
(EngineCore_DP0 pid=17822) Search for `hipErrorInvalidKernelFile' in https://rocm.docs.amd.com/projects/HIP/en/latest/index.html for more information.
(EngineCore_DP0 pid=17822) HIP kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect.
(EngineCore_DP0 pid=17822) For debugging consider passing AMD_SERIALIZE_KERNEL=3
(EngineCore_DP0 pid=17822) Device-side assertion tracking was not enabled by user.
(EngineCore_DP0 pid=17822) INFO 01-31 14:21:03 [ray_executor.py:120] Shutting down Ray distributed executor. If you see error log from logging.cc regarding SIGTERM received, please ignore because this is the expected termination process in Ray.
(EngineCore_DP0 pid=17822) 2026-01-31 14:21:03,366	ERROR worker.py:433 -- Unhandled error (suppress with 'RAY_IGNORE_UNHANDLED_ERRORS=1'): ray::RayWorkerWrapper.execute_method() (pid=9409, ip=192.168.100.2, actor_id=4ec5a2c97a621462a00d303801000000, repr=<vllm.v1.executor.ray_utils.RayWorkerWrapper object at 0x7fc7d70fc590>)
(EngineCore_DP0 pid=17822)   File "/opt/venv/lib64/python3.13/site-packages/vllm/v1/worker/worker_base.py", line 345, in execute_method
(EngineCore_DP0 pid=17822)     raise e
(EngineCore_DP0 pid=17822)   File "/opt/venv/lib64/python3.13/site-packages/vllm/v1/worker/worker_base.py", line 334, in execute_method
(EngineCore_DP0 pid=17822)     return run_method(self, method, args, kwargs)
(EngineCore_DP0 pid=17822)   File "/opt/venv/lib64/python3.13/site-packages/vllm/v1/serial_utils.py", line 459, in run_method
(EngineCore_DP0 pid=17822)     return func(*args, **kwargs)
(EngineCore_DP0 pid=17822)   File "/opt/venv/lib64/python3.13/site-packages/vllm/v1/worker/worker_base.py", line 326, in init_device
(EngineCore_DP0 pid=17822)     self.worker.init_device()  # type: ignore
(EngineCore_DP0 pid=17822)     ~~~~~~~~~~~~~~~~~~~~~~~^^
(EngineCore_DP0 pid=17822)   File "/opt/venv/lib64/python3.13/site-packages/vllm/v1/worker/gpu_worker.py", line 221, in init_device
(EngineCore_DP0 pid=17822)     init_worker_distributed_environment(
(EngineCore_DP0 pid=17822)     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^
(EngineCore_DP0 pid=17822)         self.vllm_config,
(EngineCore_DP0 pid=17822)         ^^^^^^^^^^^^^^^^^
(EngineCore_DP0 pid=17822)     ...<3 lines>...
(EngineCore_DP0 pid=17822)         current_platform.dist_backend,
(EngineCore_DP0 pid=17822)         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(EngineCore_DP0 pid=17822)     )
(EngineCore_DP0 pid=17822)     ^
(EngineCore_DP0 pid=17822)   File "/opt/venv/lib64/python3.13/site-packages/vllm/v1/worker/gpu_worker.py", line 959, in init_worker_distributed_environment
(EngineCore_DP0 pid=17822)     ensure_model_parallel_initialized(
(EngineCore_DP0 pid=17822)     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^
(EngineCore_DP0 pid=17822)         parallel_config.tensor_parallel_size,
(EngineCore_DP0 pid=17822)         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(EngineCore_DP0 pid=17822)     ...<2 lines>...
(EngineCore_DP0 pid=17822)         parallel_config.decode_context_parallel_size,
(EngineCore_DP0 pid=17822)         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(EngineCore_DP0 pid=17822)     )
(EngineCore_DP0 pid=17822)     ^
(EngineCore_DP0 pid=17822)   File "/opt/venv/lib64/python3.13/site-packages/vllm/distributed/parallel_state.py", line 1472, in ensure_model_parallel_initialized
(EngineCore_DP0 pid=17822)     initialize_model_parallel(
(EngineCore_DP0 pid=17822)     ~~~~~~~~~~~~~~~~~~~~~~~~~^
(EngineCore_DP0 pid=17822)         tensor_model_parallel_size,
(EngineCore_DP0 pid=17822)         ^^^^^^^^^^^^^^^^^^^^^^^^^^^
(EngineCore_DP0 pid=17822)     ...<3 lines>...
(EngineCore_DP0 pid=17822)         backend,
(EngineCore_DP0 pid=17822)         ^^^^^^^^
(EngineCore_DP0 pid=17822)     )
(EngineCore_DP0 pid=17822)     ^
(EngineCore_DP0 pid=17822)   File "/opt/venv/lib64/python3.13/site-packages/vllm/distributed/parallel_state.py", line 1369, in initialize_model_parallel
(EngineCore_DP0 pid=17822)     _TP = init_model_parallel_group(
(EngineCore_DP0 pid=17822)         group_ranks,
(EngineCore_DP0 pid=17822)     ...<3 lines>...
(EngineCore_DP0 pid=17822)         group_name="tp",
(EngineCore_DP0 pid=17822)     )
(EngineCore_DP0 pid=17822)   File "/opt/venv/lib64/python3.13/site-packages/vllm/distributed/parallel_state.py", line 1089, in init_model_parallel_group
(EngineCore_DP0 pid=17822)     return GroupCoordinator(
(EngineCore_DP0 pid=17822)         group_ranks=group_ranks,
(EngineCore_DP0 pid=17822)     ...<4 lines>...
(EngineCore_DP0 pid=17822)         group_name=group_name,
(EngineCore_DP0 pid=17822)     )
(EngineCore_DP0 pid=17822)   File "/opt/venv/lib64/python3.13/site-packages/vllm/distributed/parallel_state.py", line 362, in __init__
(EngineCore_DP0 pid=17822)     self.device_communicator = device_comm_cls(
(EngineCore_DP0 pid=17822)                                ~~~~~~~~~~~~~~~^
(EngineCore_DP0 pid=17822)         cpu_group=self.cpu_group,
(EngineCore_DP0 pid=17822)         ^^^^^^^^^^^^^^^^^^^^^^^^^
(EngineCore_DP0 pid=17822)     ...<2 lines>...
(EngineCore_DP0 pid=17822)         unique_name=self.unique_name,
(EngineCore_DP0 pid=17822)         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(EngineCore_DP0 pid=17822)     )
(EngineCore_DP0 pid=17822)     ^
(EngineCore_DP0 pid=17822)   File "/opt/venv/lib64/python3.13/site-packages/vllm/distributed/device_communicators/cuda_communicator.py", line 58, in __init__
(EngineCore_DP0 pid=17822)     self.pynccl_comm = PyNcclCommunicator(
(EngineCore_DP0 pid=17822)                        ~~~~~~~~~~~~~~~~~~^
(EngineCore_DP0 pid=17822)         group=self.cpu_group,
(EngineCore_DP0 pid=17822)         ^^^^^^^^^^^^^^^^^^^^^
(EngineCore_DP0 pid=17822)         device=self.device,
(EngineCore_DP0 pid=17822)         ^^^^^^^^^^^^^^^^^^^
(EngineCore_DP0 pid=17822)     )
(EngineCore_DP0 pid=17822)     ^
(EngineCore_DP0 pid=17822)   File "/opt/venv/lib64/python3.13/site-packages/vllm/distributed/device_communicators/pynccl.py", line 145, in __init__
(EngineCore_DP0 pid=17822)     data = torch.zeros(1, device=device)
(EngineCore_DP0 pid=17822) torch.AcceleratorError: HIP error: invalid kernel file
(EngineCore_DP0 pid=17822) Search for `hipErrorInvalidKernelFile' in https://rocm.docs.amd.com/projects/HIP/en/latest/index.html for more information.
(EngineCore_DP0 pid=17822) HIP kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect.
(EngineCore_DP0 pid=17822) For debugging consider passing AMD_SERIALIZE_KERNEL=3
(EngineCore_DP0 pid=17822) Device-side assertion tracking was not enabled by user.
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=9409, ip=192.168.100.2) ERROR 01-31 14:21:03 [worker_base.py:344] Error executing method 'init_device'. This might cause deadlock in distributed execution.
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=9409, ip=192.168.100.2) ERROR 01-31 14:21:03 [worker_base.py:344] Traceback (most recent call last):
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=9409, ip=192.168.100.2) ERROR 01-31 14:21:03 [worker_base.py:344]   File "/opt/venv/lib64/python3.13/site-packages/vllm/v1/worker/worker_base.py", line 334, in execute_method
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=9409, ip=192.168.100.2) ERROR 01-31 14:21:03 [worker_base.py:344]     return run_method(self, method, args, kwargs)
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=9409, ip=192.168.100.2) ERROR 01-31 14:21:03 [worker_base.py:344]   File "/opt/venv/lib64/python3.13/site-packages/vllm/v1/serial_utils.py", line 459, in run_method
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=9409, ip=192.168.100.2) ERROR 01-31 14:21:03 [worker_base.py:344]     return func(*args, **kwargs)
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=9409, ip=192.168.100.2) ERROR 01-31 14:21:03 [worker_base.py:344]   File "/opt/venv/lib64/python3.13/site-packages/ray/util/tracing/tracing_helper.py", line 461, in _resume_span
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=9409, ip=192.168.100.2) ERROR 01-31 14:21:03 [worker_base.py:344]     return method(self, *_args, **_kwargs)
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=9409, ip=192.168.100.2) ERROR 01-31 14:21:03 [worker_base.py:344]   File "/opt/venv/lib64/python3.13/site-packages/vllm/v1/worker/worker_base.py", line 326, in init_device
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=9409, ip=192.168.100.2) ERROR 01-31 14:21:03 [worker_base.py:344]     self.worker.init_device()  # type: ignore
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=9409, ip=192.168.100.2) ERROR 01-31 14:21:03 [worker_base.py:344]     ~~~~~~~~~~~~~~~~~~~~~~~^^
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=9409, ip=192.168.100.2) ERROR 01-31 14:21:03 [worker_base.py:344]   File "/opt/venv/lib64/python3.13/site-packages/vllm/v1/worker/gpu_worker.py", line 221, in init_device
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=9409, ip=192.168.100.2) ERROR 01-31 14:21:03 [worker_base.py:344]     init_worker_distributed_environment(
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=9409, ip=192.168.100.2) ERROR 01-31 14:21:03 [worker_base.py:344]     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=9409, ip=192.168.100.2) ERROR 01-31 14:21:03 [worker_base.py:344]         self.vllm_config,
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=9409, ip=192.168.100.2) ERROR 01-31 14:21:03 [worker_base.py:344]         ^^^^^^^^^^^^^^^^^
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=9409, ip=192.168.100.2) ERROR 01-31 14:21:03 [worker_base.py:344]     ...<3 lines>...
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=9409, ip=192.168.100.2) ERROR 01-31 14:21:03 [worker_base.py:344]         current_platform.dist_backend,
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=9409, ip=192.168.100.2) ERROR 01-31 14:21:03 [worker_base.py:344]         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=9409, ip=192.168.100.2) ERROR 01-31 14:21:03 [worker_base.py:344]     )
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=9409, ip=192.168.100.2) ERROR 01-31 14:21:03 [worker_base.py:344]     ^
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=9409, ip=192.168.100.2) ERROR 01-31 14:21:03 [worker_base.py:344]   File "/opt/venv/lib64/python3.13/site-packages/vllm/v1/worker/gpu_worker.py", line 959, in init_worker_distributed_environment
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=9409, ip=192.168.100.2) ERROR 01-31 14:21:03 [worker_base.py:344]     ensure_model_parallel_initialized(
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=9409, ip=192.168.100.2) ERROR 01-31 14:21:03 [worker_base.py:344]     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=9409, ip=192.168.100.2) ERROR 01-31 14:21:03 [worker_base.py:344]         parallel_config.tensor_parallel_size,
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=9409, ip=192.168.100.2) ERROR 01-31 14:21:03 [worker_base.py:344]         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=9409, ip=192.168.100.2) ERROR 01-31 14:21:03 [worker_base.py:344]     ...<2 lines>...
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=9409, ip=192.168.100.2) ERROR 01-31 14:21:03 [worker_base.py:344]         parallel_config.decode_context_parallel_size,
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=9409, ip=192.168.100.2) ERROR 01-31 14:21:03 [worker_base.py:344]         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=9409, ip=192.168.100.2) ERROR 01-31 14:21:03 [worker_base.py:344]     )
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=9409, ip=192.168.100.2) ERROR 01-31 14:21:03 [worker_base.py:344]     ^
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=9409, ip=192.168.100.2) ERROR 01-31 14:21:03 [worker_base.py:344]   File "/opt/venv/lib64/python3.13/site-packages/vllm/distributed/parallel_state.py", line 1472, in ensure_model_parallel_initialized
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=9409, ip=192.168.100.2) ERROR 01-31 14:21:03 [worker_base.py:344]     initialize_model_parallel(
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=9409, ip=192.168.100.2) ERROR 01-31 14:21:03 [worker_base.py:344]     ~~~~~~~~~~~~~~~~~~~~~~~~~^
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=9409, ip=192.168.100.2) ERROR 01-31 14:21:03 [worker_base.py:344]         tensor_model_parallel_size,
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=9409, ip=192.168.100.2) ERROR 01-31 14:21:03 [worker_base.py:344]         ^^^^^^^^^^^^^^^^^^^^^^^^^^^
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=9409, ip=192.168.100.2) ERROR 01-31 14:21:03 [worker_base.py:344]     ...<3 lines>...
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=9409, ip=192.168.100.2) ERROR 01-31 14:21:03 [worker_base.py:344]         backend,
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=9409, ip=192.168.100.2) ERROR 01-31 14:21:03 [worker_base.py:344]         ^^^^^^^^
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=9409, ip=192.168.100.2) ERROR 01-31 14:21:03 [worker_base.py:344]     )
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=9409, ip=192.168.100.2) ERROR 01-31 14:21:03 [worker_base.py:344]     ^
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=9409, ip=192.168.100.2) ERROR 01-31 14:21:03 [worker_base.py:344]   File "/opt/venv/lib64/python3.13/site-packages/vllm/distributed/parallel_state.py", line 1369, in initialize_model_parallel
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=9409, ip=192.168.100.2) ERROR 01-31 14:21:03 [worker_base.py:344]     _TP = init_model_parallel_group(
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=9409, ip=192.168.100.2) ERROR 01-31 14:21:03 [worker_base.py:344]         group_ranks,
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=9409, ip=192.168.100.2) ERROR 01-31 14:21:03 [worker_base.py:344]     ...<3 lines>...
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=9409, ip=192.168.100.2) ERROR 01-31 14:21:03 [worker_base.py:344]         group_name="tp",
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=9409, ip=192.168.100.2) ERROR 01-31 14:21:03 [worker_base.py:344]     )
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=9409, ip=192.168.100.2) ERROR 01-31 14:21:03 [worker_base.py:344]   File "/opt/venv/lib64/python3.13/site-packages/vllm/distributed/parallel_state.py", line 1089, in init_model_parallel_group
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=9409, ip=192.168.100.2) ERROR 01-31 14:21:03 [worker_base.py:344]     return GroupCoordinator(
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=9409, ip=192.168.100.2) ERROR 01-31 14:21:03 [worker_base.py:344]         group_ranks=group_ranks,
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=9409, ip=192.168.100.2) ERROR 01-31 14:21:03 [worker_base.py:344]     ...<4 lines>...
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=9409, ip=192.168.100.2) ERROR 01-31 14:21:03 [worker_base.py:344]         group_name=group_name,
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=9409, ip=192.168.100.2) ERROR 01-31 14:21:03 [worker_base.py:344]     )
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=9409, ip=192.168.100.2) ERROR 01-31 14:21:03 [worker_base.py:344]   File "/opt/venv/lib64/python3.13/site-packages/vllm/distributed/parallel_state.py", line 362, in __init__
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=9409, ip=192.168.100.2) ERROR 01-31 14:21:03 [worker_base.py:344]     self.device_communicator = device_comm_cls(
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=9409, ip=192.168.100.2) ERROR 01-31 14:21:03 [worker_base.py:344]                                ~~~~~~~~~~~~~~~^
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=9409, ip=192.168.100.2) ERROR 01-31 14:21:03 [worker_base.py:344]         cpu_group=self.cpu_group,
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=9409, ip=192.168.100.2) ERROR 01-31 14:21:03 [worker_base.py:344]         ^^^^^^^^^^^^^^^^^^^^^^^^^
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=9409, ip=192.168.100.2) ERROR 01-31 14:21:03 [worker_base.py:344]     ...<2 lines>...
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=9409, ip=192.168.100.2) ERROR 01-31 14:21:03 [worker_base.py:344]         unique_name=self.unique_name,
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=9409, ip=192.168.100.2) ERROR 01-31 14:21:03 [worker_base.py:344]         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=9409, ip=192.168.100.2) ERROR 01-31 14:21:03 [worker_base.py:344]     )
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=9409, ip=192.168.100.2) ERROR 01-31 14:21:03 [worker_base.py:344]     ^
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=9409, ip=192.168.100.2) ERROR 01-31 14:21:03 [worker_base.py:344]   File "/opt/venv/lib64/python3.13/site-packages/vllm/distributed/device_communicators/cuda_communicator.py", line 58, in __init__
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=9409, ip=192.168.100.2) ERROR 01-31 14:21:03 [worker_base.py:344]     self.pynccl_comm = PyNcclCommunicator(
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=9409, ip=192.168.100.2) ERROR 01-31 14:21:03 [worker_base.py:344]                        ~~~~~~~~~~~~~~~~~~^
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=9409, ip=192.168.100.2) ERROR 01-31 14:21:03 [worker_base.py:344]         group=self.cpu_group,
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=9409, ip=192.168.100.2) ERROR 01-31 14:21:03 [worker_base.py:344]         ^^^^^^^^^^^^^^^^^^^^^
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=9409, ip=192.168.100.2) ERROR 01-31 14:21:03 [worker_base.py:344]         device=self.device,
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=9409, ip=192.168.100.2) ERROR 01-31 14:21:03 [worker_base.py:344]         ^^^^^^^^^^^^^^^^^^^
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=9409, ip=192.168.100.2) ERROR 01-31 14:21:03 [worker_base.py:344]     )
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=9409, ip=192.168.100.2) ERROR 01-31 14:21:03 [worker_base.py:344]     ^
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=9409, ip=192.168.100.2) ERROR 01-31 14:21:03 [worker_base.py:344]   File "/opt/venv/lib64/python3.13/site-packages/vllm/distributed/device_communicators/pynccl.py", line 145, in __init__
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=9409, ip=192.168.100.2) ERROR 01-31 14:21:03 [worker_base.py:344]     data = torch.zeros(1, device=device)
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=9409, ip=192.168.100.2) ERROR 01-31 14:21:03 [worker_base.py:344] torch.AcceleratorError: HIP error: invalid kernel file
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=9409, ip=192.168.100.2) ERROR 01-31 14:21:03 [worker_base.py:344] Search for `hipErrorInvalidKernelFile' in https://rocm.docs.amd.com/projects/HIP/en/latest/index.html for more information.
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=9409, ip=192.168.100.2) ERROR 01-31 14:21:03 [worker_base.py:344] HIP kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect.
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=9409, ip=192.168.100.2) ERROR 01-31 14:21:03 [worker_base.py:344] For debugging consider passing AMD_SERIALIZE_KERNEL=3
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=9409, ip=192.168.100.2) ERROR 01-31 14:21:03 [worker_base.py:344] Device-side assertion tracking was not enabled by user.
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=17931) WARNING 01-31 14:21:03 [worker_base.py:301] Missing `shared_worker_lock` argument from executor. This argument is needed for mm_processor_cache_type='shm'.
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=9409, ip=192.168.100.2) INFO 01-31 14:21:03 [parallel_state.py:1234] world_size=2 rank=1 local_rank=0 distributed_init_method=tcp://192.168.100.1:49483 backend=nccl
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=17931) ERROR 01-31 14:21:03 [worker_base.py:344] Error executing method 'init_device'. This might cause deadlock in distributed execution.
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=17931) ERROR 01-31 14:21:03 [worker_base.py:344] Traceback (most recent call last):
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=17931) ERROR 01-31 14:21:03 [worker_base.py:344]   File "/opt/venv/lib64/python3.13/site-packages/vllm/v1/worker/worker_base.py", line 334, in execute_method
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=17931) ERROR 01-31 14:21:03 [worker_base.py:344]     return run_method(self, method, args, kwargs)
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=17931) ERROR 01-31 14:21:03 [worker_base.py:344]   File "/opt/venv/lib64/python3.13/site-packages/vllm/v1/serial_utils.py", line 459, in run_method
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=17931) ERROR 01-31 14:21:03 [worker_base.py:344]     return func(*args, **kwargs)
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=17931) ERROR 01-31 14:21:03 [worker_base.py:344]   File "/opt/venv/lib64/python3.13/site-packages/ray/util/tracing/tracing_helper.py", line 461, in _resume_span
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=17931) ERROR 01-31 14:21:03 [worker_base.py:344]     return method(self, *_args, **_kwargs)
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=17931) ERROR 01-31 14:21:03 [worker_base.py:344]   File "/opt/venv/lib64/python3.13/site-packages/vllm/v1/worker/gpu_worker.py", line 221, in init_device [repeated 2x across cluster] (Ray deduplicates logs by default. Set RAY_DEDUP_LOGS=0 to disable log deduplication, or see https://docs.ray.io/en/master/ray-observability/user-guides/configure-logging.html#log-deduplication for more options.)
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=17931) ERROR 01-31 14:21:03 [worker_base.py:344]     self.worker.init_device()  # type: ignore
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=17931) ERROR 01-31 14:21:03 [worker_base.py:344]     ~~~~~~~~~~~~~~~~~~~~~~~^^
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=17931) ERROR 01-31 14:21:03 [worker_base.py:344]     init_worker_distributed_environment(
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=17931) ERROR 01-31 14:21:03 [worker_base.py:344]     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=17931) ERROR 01-31 14:21:03 [worker_base.py:344]         self.vllm_config,
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=17931) ERROR 01-31 14:21:03 [worker_base.py:344]         ^^^^^^^^^^^^^^^^^
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=17931) ERROR 01-31 14:21:03 [worker_base.py:344]     ...<2 lines>... [repeated 6x across cluster]
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=17931) ERROR 01-31 14:21:03 [worker_base.py:344]         current_platform.dist_backend,
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=17931) ERROR 01-31 14:21:03 [worker_base.py:344]         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=17931) ERROR 01-31 14:21:03 [worker_base.py:344]     ) [repeated 7x across cluster]
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=17931) ERROR 01-31 14:21:03 [worker_base.py:344]     ^ [repeated 5x across cluster]
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=17931) ERROR 01-31 14:21:03 [worker_base.py:344]   File "/opt/venv/lib64/python3.13/site-packages/vllm/v1/worker/gpu_worker.py", line 959, in init_worker_distributed_environment
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=17931) ERROR 01-31 14:21:03 [worker_base.py:344]     ensure_model_parallel_initialized(
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=17931) ERROR 01-31 14:21:03 [worker_base.py:344]     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=17931) ERROR 01-31 14:21:03 [worker_base.py:344]         parallel_config.tensor_parallel_size,
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=17931) ERROR 01-31 14:21:03 [worker_base.py:344]         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=17931) ERROR 01-31 14:21:03 [worker_base.py:344]         parallel_config.decode_context_parallel_size,
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=17931) ERROR 01-31 14:21:03 [worker_base.py:344]         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=17931) ERROR 01-31 14:21:03 [worker_base.py:344]   File "/opt/venv/lib64/python3.13/site-packages/vllm/distributed/parallel_state.py", line 1472, in ensure_model_parallel_initialized
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=17931) ERROR 01-31 14:21:03 [worker_base.py:344]     initialize_model_parallel(
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=17931) ERROR 01-31 14:21:03 [worker_base.py:344]     ~~~~~~~~~~~~~~~~~~~~~~~~~^
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=17931) ERROR 01-31 14:21:03 [worker_base.py:344]         tensor_model_parallel_size,
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=17931) ERROR 01-31 14:21:03 [worker_base.py:344]         ^^^^^^^^^^^^^^^^^^^^^^^^^^^
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=17931) ERROR 01-31 14:21:03 [worker_base.py:344]         backend,
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=17931) ERROR 01-31 14:21:03 [worker_base.py:344]         ^^^^^^^^
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=17931) ERROR 01-31 14:21:03 [worker_base.py:344]   File "/opt/venv/lib64/python3.13/site-packages/vllm/distributed/parallel_state.py", line 1369, in initialize_model_parallel
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=17931) ERROR 01-31 14:21:03 [worker_base.py:344]     _TP = init_model_parallel_group(
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=17931) ERROR 01-31 14:21:03 [worker_base.py:344]         group_ranks,
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=17931) ERROR 01-31 14:21:03 [worker_base.py:344]         group_name="tp",
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=17931) ERROR 01-31 14:21:03 [worker_base.py:344]   File "/opt/venv/lib64/python3.13/site-packages/vllm/distributed/parallel_state.py", line 1089, in init_model_parallel_group
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=17931) ERROR 01-31 14:21:03 [worker_base.py:344]     return GroupCoordinator(
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=17931) ERROR 01-31 14:21:03 [worker_base.py:344]         group_ranks=group_ranks,
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=17931) ERROR 01-31 14:21:03 [worker_base.py:344]         group_name=group_name,
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=17931) ERROR 01-31 14:21:03 [worker_base.py:344]   File "/opt/venv/lib64/python3.13/site-packages/vllm/distributed/device_communicators/pynccl.py", line 145, in __init__ [repeated 3x across cluster]
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=17931) ERROR 01-31 14:21:03 [worker_base.py:344]     self.device_communicator = device_comm_cls(
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=17931) ERROR 01-31 14:21:03 [worker_base.py:344]                                ~~~~~~~~~~~~~~~^
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=17931) ERROR 01-31 14:21:03 [worker_base.py:344]         cpu_group=self.cpu_group,
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=17931) ERROR 01-31 14:21:03 [worker_base.py:344]         ^^^^^^^^^^^^^^^^^^^^^^^^^
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=17931) ERROR 01-31 14:21:03 [worker_base.py:344]         unique_name=self.unique_name,
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=17931) ERROR 01-31 14:21:03 [worker_base.py:344]         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=17931) ERROR 01-31 14:21:03 [worker_base.py:344]     self.pynccl_comm = PyNcclCommunicator(
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=17931) ERROR 01-31 14:21:03 [worker_base.py:344]                        ~~~~~~~~~~~~~~~~~~^
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=17931) ERROR 01-31 14:21:03 [worker_base.py:344]         group=self.cpu_group,
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=17931) ERROR 01-31 14:21:03 [worker_base.py:344]         ^^^^^^^^^^^^^^^^^^^^^
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=17931) ERROR 01-31 14:21:03 [worker_base.py:344]         device=self.device,
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=17931) ERROR 01-31 14:21:03 [worker_base.py:344]         ^^^^^^^^^^^^^^^^^^^
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=17931) ERROR 01-31 14:21:03 [worker_base.py:344]     data = torch.zeros(1, device=device)
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=17931) ERROR 01-31 14:21:03 [worker_base.py:344] torch.AcceleratorError: HIP error: invalid kernel file
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=17931) ERROR 01-31 14:21:03 [worker_base.py:344] Search for `hipErrorInvalidKernelFile' in https://rocm.docs.amd.com/projects/HIP/en/latest/index.html for more information.
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=17931) ERROR 01-31 14:21:03 [worker_base.py:344] HIP kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect.
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=17931) ERROR 01-31 14:21:03 [worker_base.py:344] For debugging consider passing AMD_SERIALIZE_KERNEL=3
(EngineCore_DP0 pid=17822) (RayWorkerWrapper pid=17931) ERROR 01-31 14:21:03 [worker_base.py:344] Device-side assertion tracking was not enabled by user.

4.1 - Possible reasons

This invalid kernel file might be related to RCCL not supporting gfx1150. There was a PR that was never merged:

https://github.com/ROCm/rccl/pull/2075

Instead of being merged, this was closed when AMD moved ROCm to rocm-systems. Right now there is a feature request:

https://github.com/ROCm/rocm-systems/issues/2788

Again, I do not know if this is the issue or if the invalid kernel is a different issue altogether.

68 KiB Czysty Wina Historia