docker run -td --name bug_repro --rm --device /dev/kfd \
    -e PYTORCH_ROCM_ARCH=gfx942 \
    -e PYTORCH_TUNABLEOP_ENABLED=0 \
    --device /dev/dri --security-opt seccomp=unconfined \
    --shm-size=128g --net host --pid host \
    -v $(pwd)/hf_cache:/hf_cache \
    rocm/vllm-dev:nightly_main_20260413 /bin/bash

docker exec -it bug_repro /bin/bash

---

import glob
import os

import huggingface_hub
import torch
import triton
from safetensors import safe_open

from vllm.model_executor.layers.quantization.utils.nvfp4_emulation_utils import (
    break_fp4_bytes,
    kE2M1ToFloat_handle,
    ref_nvfp4_quant,
)

# ---- Workaround env var ----
_SYNCHRONIZE_LAYER = bool(int(os.environ.get("SYNCHRONIZE_LAYER", "0")))

# ---- Qwen3-30B-A3B-FP4 dimensions ----
E = 128         # num experts
K = 2048        # hidden_dim
N = 768         # intermediate_size_per_partition (1536/2 for gated)
TOP_K = 8
NUM_LAYERS = 10  # higher number of layers = higher likelihood of issue
ATTN_DIM = 4096  # attention intermediate dim
BLOCK_SIZE = 16


# ---- 3D-capable dequantize (not on main branch) ----
def dequantize_to_dtype(
    tensor_fp4: torch.Tensor,
    tensor_sf: torch.Tensor,
    global_scale: torch.Tensor,
    dtype: torch.dtype,
):
    """Dequantize the fp4 tensor back to high precision.

    Supports both 2D and 3D inputs:
    - 2D: [m, packed_k] -> [m, k]
    - 3D: [dim0, m, packed_k] -> [dim0, m, k]
    """
    assert tensor_fp4.dtype == torch.uint8

    is_3d = tensor_fp4.ndim == 3

    if is_3d:
        dim0, m, packed_k = tensor_fp4.shape
        tensor_fp4 = tensor_fp4.reshape(-1, packed_k)
        tensor_sf = tensor_sf.reshape(-1, tensor_sf.shape[-1])
        global_scale = global_scale[:, None, None]
    else:
        m, packed_k = tensor_fp4.shape

    k = packed_k * 2
    tensor_f32 = break_fp4_bytes(tensor_fp4, torch.float32)
    tensor_f32 = tensor_f32.reshape(-1, k // BLOCK_SIZE, BLOCK_SIZE)
    tensor_sf = tensor_sf.view(torch.float8_e4m3fn)

    if is_3d:
        tensor_sf = tensor_sf.reshape(dim0, m, k // BLOCK_SIZE)
    tensor_sf_dtype = tensor_sf.to(torch.float32) * global_scale

    if is_3d:
        tensor_f32 = tensor_f32.reshape(dim0, m, -1, BLOCK_SIZE)

    out = tensor_f32 * tensor_sf_dtype.unsqueeze(-1)
    out = out.reshape(*out.shape[:-2], -1)

    return out.to(dtype)


# ---- Activation quantize-dequantize (inline, no dependency on branch code) ----
def ref_nvfp4_quant_dequant(x, global_scale):
    """Quantize to FP4 and immediately dequantize back (activation QDQ)."""
    fp4, scale = ref_nvfp4_quant(x, global_scale, BLOCK_SIZE)
    m, n = x.shape
    fp4 = fp4.reshape(m, n // BLOCK_SIZE, BLOCK_SIZE)
    scale = scale.unsqueeze(-1) / global_scale
    x_dq = (fp4 * scale).reshape(m, n).to(x.dtype)
    return x_dq, None


MODEL_ID = "nvidia/Qwen3-30B-A3B-NVFP4"


def load_checkpoint(device):
    """Download model and load all safetensors into a flat dict on device."""
    model_path = huggingface_hub.snapshot_download(MODEL_ID)
    shard_files = sorted(glob.glob(os.path.join(model_path, "*.safetensors")))
    print(f"Found {len(shard_files)} safetensors shard(s) in {model_path}")

    tensors = {}
    for sf_path in shard_files:
        with safe_open(sf_path, framework="pt", device=str(device)) as f:
            for k in f.keys():
                tensors[k] = f.get_tensor(k)
    return tensors


def load_layer_weights(tensors, layer_idx, device):
    """Extract MoE and attention weights for a single layer from the flat dict.

    Returns:
        (w1_packed, w1_blockscale, w1_global_scale,
         w2_packed, w2_blockscale, w2_global_scale,
         act_global_scale,
         lin_qkv, lin_o)
    """
    prefix = f"model.layers.{layer_idx}"

    # ---- MoE expert weights ----
    all_w1_packed, all_w1_blockscale = [], []
    all_w2_packed, all_w2_blockscale = [], []
    g1_alphas, g2_alphas = [], []
    a1_gscales = []

    for e in range(E):
        ep = f"{prefix}.mlp.experts.{e}"

        # gate_proj + up_proj -> w1 (stacked along out_dim)
        gate_w = tensors[f"{ep}.gate_proj.weight"]       # [N, K//2] uint8
        up_w = tensors[f"{ep}.up_proj.weight"]            # [N, K//2] uint8
        w1_packed = torch.cat([gate_w, up_w], dim=0)      # [2*N, K//2]
        all_w1_packed.append(w1_packed)

        gate_bs = tensors[f"{ep}.gate_proj.weight_scale"]  # [N, K//BLOCK_SIZE] fp8
        up_bs = tensors[f"{ep}.up_proj.weight_scale"]      # [N, K//BLOCK_SIZE] fp8
        w1_bs = torch.cat([gate_bs, up_bs], dim=0)
        all_w1_blockscale.append(w1_bs.view(torch.uint8))

        # Use gate_proj's global scale for w1 (gate and up have same value)
        g1_alphas.append(tensors[f"{ep}.gate_proj.weight_scale_2"])

        # down_proj -> w2
        all_w2_packed.append(tensors[f"{ep}.down_proj.weight"])
        down_bs = tensors[f"{ep}.down_proj.weight_scale"]
        all_w2_blockscale.append(down_bs.view(torch.uint8))
        g2_alphas.append(tensors[f"{ep}.down_proj.weight_scale_2"])

        # Activation input scale (same for gate/up)
        a1_gscales.append(tensors[f"{ep}.gate_proj.input_scale"])

    w1_packed = torch.stack(all_w1_packed)          # [E, 2*N, K//2]
    w1_blockscale = torch.stack(all_w1_blockscale)  # [E, 2*N, K//BLOCK_SIZE]
    w1_global_scale = torch.stack(g1_alphas)        # [E]
    w2_packed = torch.stack(all_w2_packed)           # [E, K, N//2]
    w2_blockscale = torch.stack(all_w2_blockscale)   # [E, K, N//BLOCK_SIZE]
    w2_global_scale = torch.stack(g2_alphas)        # [E]
    # Use the first expert's input_scale as representative
    act_global_scale = a1_gscales[0].unsqueeze(0)   # [1]

    # ---- Attention linear weights (use q_proj as lin_qkv, o_proj as lin_o) ----
    attn = f"{prefix}.self_attn"
    lin_global_scale = tensors[f"{attn}.q_proj.weight_scale_2"].unsqueeze(0)

    return (w1_packed, w1_blockscale, w1_global_scale,
            w2_packed, w2_blockscale, w2_global_scale,
            act_global_scale, lin_global_scale)


def main():
    device = torch.device("cuda:0")
    dtype = torch.bfloat16
    print(f"torch={torch.__version__}")
    print(f"triton={triton.__version__}")

    print(f"SYNCHRONIZE_LAYER={_SYNCHRONIZE_LAYER}")
    print(f"Layers={NUM_LAYERS}, E={E}, K={K}, N={N}, top_k={TOP_K}")

    kE2M1ToFloat_handle.val = kE2M1ToFloat_handle.val.to(device)

    # ---- Load checkpoint ----
    print(f"Downloading / loading {MODEL_ID}...", flush=True)
    tensors = load_checkpoint(device)
    print(f"Loaded {len(tensors)} tensors from checkpoint.", flush=True)

    # ---- Build all layers ----
    print("Building layers from checkpoint weights...", flush=True)
    layers = []
    for i in range(NUM_LAYERS):
        if (i + 1) % 10 == 0 or i == 0:
            print(f"  layer {i+1}/{NUM_LAYERS}", flush=True)

        (w1_packed, w1_blockscale, w1_global_scale,
         w2_packed, w2_blockscale, w2_global_scale,
         act_global_scale, lin_global_scale) = load_layer_weights(tensors, i, device)

        layers.append((w1_packed, w1_blockscale, w1_global_scale,
                        w2_packed, w2_blockscale, w2_global_scale,
                        act_global_scale, lin_global_scale))

    # Free the full checkpoint dict to save memory
    del tensors
    torch.cuda.synchronize()
    print("All layers built.\n", flush=True)

    # ---- Forward pass function (NO sync inside) ----
    def forward(hidden_states):
        M = hidden_states.shape[0]
        for (w1_packed, w1_scale_val, _g1_alphas,
             w2_packed, w2_scale_val, _g2_alphas,
             _a1_gscale, lin_global_scale) in layers:
            for _ in range(1):
                hidden_states, _ = ref_nvfp4_quant_dequant(hidden_states, lin_global_scale)

            # NOTE: no segfault/issue if commenting out the `dequantize_to_dtype` calls.

            # Dequantize w1 — result is a LOCAL variable
            w1_dequant = dequantize_to_dtype(
                tensor_fp4=w1_packed, tensor_sf=w1_scale_val,
                global_scale=_g1_alphas, dtype=hidden_states.dtype,
            )

            # Dequantize w2 — result is a LOCAL variable
            w2_dequant = dequantize_to_dtype(
                tensor_fp4=w2_packed, tensor_sf=w2_scale_val,
                global_scale=_g2_alphas, dtype=hidden_states.dtype,
            )

            # Activation QDQ
            hidden_states, _ = ref_nvfp4_quant_dequant(
                hidden_states, _a1_gscale
            )

        return hidden_states

    # ---- Prefill ----
    print("=== PREFILL (M=16) ===", flush=True)
    h = torch.randn(16, K, dtype=dtype, device=device)
    h = forward(h)
    torch.cuda.synchronize()
    print(f"  OK (norm={h.float().norm().item():.2f})\n", flush=True)

    # ---- Decode steps (M=1, no sync between steps) ----
    num_decode = 50
    print(f"=== DECODE ({num_decode} steps, M=1, no sync between steps) ===",
          flush=True)
    for step in range(num_decode):
        h = torch.randn(1, K, dtype=dtype, device=device)
        h = forward(h)
        # NO sync here — let GPU work queue build up across steps

    # Final sync to catch deferred errors
    torch.cuda.synchronize()
    print(f"  All {num_decode} decode steps completed.", flush=True)
    print("SUCCESS")


if __name__ == "__main__":
    main()

---

hf download nvidia/Qwen3-30B-A3B-NVFP4
CUDA_VISIBLE_DEVICES=0 python repro_nvfp4_moe_main.py`

---

torch=2.12.0.dev20260413+rocm7.2
triton=3.7.0
SYNCHRONIZE_LAYER=False
Layers=10, E=128, K=2048, N=768, top_k=8
Downloading / loading nvidia/Qwen3-30B-A3B-NVFP4...
Fetching 18 files: 100%|████████████████████████████████████████████████████████████████| 18/18 [00:00<00:00, 156633.76it/s]
Found 4 safetensors shard(s) in /scratch/huggingface_hub/models--nvidia--Qwen3-30B-A3B-NVFP4/snapshots/2538ded2a4edb247b4d2b4a8ba24e44bd4c017c3
Loaded 74835 tensors from checkpoint.
Building layers from checkpoint weights...
  layer 1/10
  layer 10/10
All layers built.

=== PREFILL (M=16) ===
WARNING 04-14 13:00:35 [fused_moe.py:1090] Using default MoE config. Performance might be sub-optimal! Config file not found at /felmarty/repos/vllm/vllm/model_executor/layers/fused_moe/configs/E=128,N=768,device_name=AMD_Instinct_MI325X.json
  OK (norm=0.00)

=== DECODE (50 steps, M=1, no sync between steps) ===
Segmentation fault (core dumped)

---

torch=2.12.0.dev20260413+rocm7.2
triton=3.7.0
SYNCHRONIZE_LAYER=False
Layers=10, E=128, K=2048, N=768, top_k=8
Downloading / loading nvidia/Qwen3-30B-A3B-NVFP4...
Fetching 18 files: 100%|████████████████████████████████████████████████████████████████| 18/18 [00:00<00:00, 246723.76it/s]
Found 4 safetensors shard(s) in /scratch/huggingface_hub/models--nvidia--Qwen3-30B-A3B-NVFP4/snapshots/2538ded2a4edb247b4d2b4a8ba24e44bd4c017c3
Loaded 74835 tensors from checkpoint.
Building layers from checkpoint weights...
  layer 1/10
  layer 10/10
All layers built.

=== PREFILL (M=16) ===
WARNING 04-14 13:01:19 [fused_moe.py:1090] Using default MoE config. Performance might be sub-optimal! Config file not found at /felmarty/repos/vllm/vllm/model_executor/layers/fused_moe/configs/E=128,N=768,device_name=AMD_Instinct_MI325X.json
  OK (norm=0.00)

=== DECODE (50 steps, M=1, no sync between steps) ===
  All 50 decode steps completed.
SUCCESS

---

pip uninstall torch
pip3 install --pre torch torchvision --index-url https://download.pytorch.org/whl/nightly/rocm7.2
pip uninstall vllm mori
git clone https://github.com/vllm-project/vllm.git && cd vllm
export PYTORCH_ROCM_ARCH=gfx942
python setup.py develop

---

FROM nvidia/cuda:13.0.2-cudnn-devel-ubuntu24.04

RUN apt update && apt install -y git wget nano curl

ENV PATH="/root/miniforge3/bin:${PATH}"
ARG PATH="/root/miniforge3/bin:${PATH}"

RUN wget https://github.com/conda-forge/miniforge/releases/latest/download/Miniforge3-Linux-x86_64.sh && \
    bash Miniforge3-Linux-x86_64.sh -b && \
    conda init

RUN pip install torch==2.11.0

ENV TORCH_CUDA_ARCH_LIST=9.0

RUN pip install accelerate datasets pytest nltk huggingface_hub

WORKDIR /workspace

---

VLLM_USE_PRECOMPILED=1 pip install -vvv --editable . --no-build-isolation

---

=== DECODE (50 steps, M=1, no sync between steps) ===

Thread 66 "python" received signal SIGSEGV, Segmentation fault.
[Switching to thread 66 (Thread 0x7ffdad9ff640 (LWP 3700660))]
0x00007ffec3a76921 in rocr::core::Runtime::AsyncEventsLoop(void*) ()
   from /usr/local/lib/python3.12/dist-packages/torch/lib/libhsa-runtime64.so
(gdb) thread apply all bt

Thread 76 (Thread 0x7fbd68ffc640 (LWP 3700672) "python"):
#0  0x00007ffff7ce7117 in ?? () from /lib/x86_64-linux-gnu/libc.so.6
#1  0x00007ffff7cf1f7e in ?? () from /lib/x86_64-linux-gnu/libc.so.6
#2  0x0000000000529af0 in PyThread_acquire_lock_timed ()
#3  0x000000000063b117 in ?? ()
#4  0x000000000066a252 in ?? ()
#5  0x0000000000570bab in ?? ()
#6  0x0000000000562386 in PyObject_Vectorcall ()
#7  0x0000000000549fe0 in _PyEval_EvalFrameDefault ()
#8  0x000000000059961d in ?? ()
#9  0x00000000005991e6 in ?? ()
#10 0x00000000006a7499 in ?? ()
#11 0x00000000006a7448 in ?? ()
#12 0x00007ffff7ceaac3 in ?? () from /lib/x86_64-linux-gnu/libc.so.6
#13 0x00007ffff7d7c8d0 in ?? () from /lib/x86_64-linux-gnu/libc.so.6

Thread 67 (Thread 0x7ffda7fff640 (LWP 3700661) "python"):
#0  0x00007ffff7d709cf in ioctl () from /lib/x86_64-linux-gnu/libc.so.6
#1  0x00007ffec3b222b0 in hsakmt_ioctl () from /usr/local/lib/python3.12/dist-packages/torch/lib/libhsa-runtime64.so
#2  0x00007ffec3b17f53 in hsaKmtWaitOnMultipleEvents_ExtCtx () from /usr/local/lib/python3.12/dist-packages/torch/lib/libhsa-runtime64.so
#3  0x00007ffec3a9b999 in rocr::core::Signal::WaitAnyExceptions(unsigned int, hsa_signal_s const*, hsa_signal_condition_t const*, long const*, long*) () from /usr/local/lib/python3.12/dist-packages/torch/lib/libhsa-runtime64.so
#4  0x00007ffec3a77248 in rocr::core::Runtime::AsyncEventsLoop(void*) () from /usr/local/lib/python3.12/dist-packages/torch/lib/libhsa-runtime64.so
#5  0x00007ffec3ad34dd in rocr::os::ThreadTrampoline(void*) () from /usr/local/lib/python3.12/dist-packages/torch/lib/libhsa-runtime64.so
#6  0x00007ffff7ceaac3 in ?? () from /lib/x86_64-linux-gnu/libc.so.6
#7  0x00007ffff7d7c8d0 in ?? () from /lib/x86_64-linux-gnu/libc.so.6

Thread 66 (Thread 0x7ffdad9ff640 (LWP 3700660) "python"):
#0  0x00007ffec3a76921 in rocr::core::Runtime::AsyncEventsLoop(void*) () from /usr/local/lib/python3.12/dist-packages/torch/lib/libhsa-runtime64.so

---

CMake version: version 3.31.10
Libc version: glibc-2.35

Python version: 3.12.13 (main, Mar  4 2026, 09:23:07) [GCC 11.4.0] (64-bit runtime)
Python platform: Linux-6.8.0-101-generic-x86_64-with-glibc2.35
Is CUDA available: True
CUDA runtime version: Could not collect
CUDA_MODULE_LOADING set to:
GPU models and configuration: AMD Instinct MI325X (gfx942:sramecc+:xnack-)
Nvidia driver version: Could not collect
cuDNN version: Could not collect
Is XPU available: False
HIP runtime version: 7.2.53211
MIOpen runtime version: 3.5.1
Is XNNPACK available: True
Caching allocator config: N/A

CPU:
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):                                  256
On-line CPU(s) list:                     0-255
Vendor ID:                               AuthenticAMD
Model name:                              AMD EPYC 9555 64-Core Processor
CPU family:                              26
Model:                                   2
Thread(s) per core:                      2
Core(s) per socket:                      64
Socket(s):                               2
Stepping:                                1
Frequency boost:                         enabled
CPU max MHz:                             4409.3750
CPU min MHz:                             1500.0000
BogoMIPS:                                6399.98
Flags:                                   fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ht syscall nx mmxext fxsr_opt pdpe1gb rdtscp lm constant_tsc rep_good amd_lbr_v2 nopl nonstop_tsc cpuid extd_apicid aperfmperf rapl pni pclmulqdq monitor ssse3 fma cx16 pcid sse4_1 sse4_2 x2apic movbe popcnt aes xsave avx f16c rdrand lahf_lm cmp_legacy svm extapic cr8_legacy abm sse4a misalignsse 3dnowprefetch osvw ibs skinit wdt tce topoext perfctr_core perfctr_nb bpext perfctr_llc mwaitx cpb cat_l3 cdp_l3 hw_pstate ssbd mba perfmon_v2 ibrs ibpb stibp ibrs_enhanced vmmcall fsgsbase tsc_adjust bmi1 avx2 smep bmi2 erms invpcid cqm rdt_a avx512f avx512dq rdseed adx smap avx512ifma clflushopt clwb avx512cd sha_ni avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_local user_shstk avx_vnni avx512_bf16 clzero irperf xsaveerptr rdpru wbnoinvd amd_ppin cppc amd_ibpb_ret arat npt lbrv svm_lock nrip_save tsc_scale vmcb_clean flushbyasid decodeassists pausefilter pfthreshold avic v_vmsave_vmload vgif x2avic v_spec_ctrl vnmi avx512vbmi umip pku ospke avx512_vbmi2 gfni vaes vpclmulqdq avx512_vnni avx512_bitalg avx512_vpopcntdq la57 rdpid bus_lock_detect movdiri movdir64b overflow_recov succor smca fsrm avx512_vp2intersect flush_l1d debug_swap
Virtualization:                          AMD-V
L1d cache:                               6 MiB (128 instances)
L1i cache:                               4 MiB (128 instances)
L2 cache:                                128 MiB (128 instances)
L3 cache:                                512 MiB (16 instances)
NUMA node(s):                            2
NUMA node0 CPU(s):                       0-63,128-191
NUMA node1 CPU(s):                       64-127,192-255
Vulnerability Gather data sampling:      Not affected
Vulnerability Indirect target selection: 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; STIBP always-on; PBRSB-eIBRS Not affected; BHI Not affected
Vulnerability Srbds:                     Not affected
Vulnerability Tsa:                       Not affected
Vulnerability Tsx async abort:           Not affected
Vulnerability Vmscape:                   Not affected

Versions of relevant libraries:
[pip3] conch-triton-kernels==1.2.1
[pip3] numpy==2.1.3
[pip3] onnx==1.19.0
[pip3] onnx-ir==0.2.0
[pip3] onnxscript==0.6.2
[pip3] onnxslim==0.1.91
[pip3] torch==2.12.0.dev20260413+rocm7.2
[pip3] torchaudio==2.9.0+eaa9e4e
[pip3] torchvision==0.24.1+d801a34
[pip3] triton==3.6.0
[pip3] triton_kernels==1.0.0
[pip3] triton-rocm==3.7.0+git282c8251
[conda] Could not collect

---

PyTorch version: 2.11.0+cu130
Is debug build: False
CUDA used to build PyTorch: 13.0
ROCM used to build PyTorch: N/A

OS: Ubuntu 24.04.3 LTS (x86_64)
GCC version: (Ubuntu 13.3.0-6ubuntu2~24.04) 13.3.0
Clang version: Could not collect
CMake version: Could not collect
Libc version: glibc-2.39

Python version: 3.13.12 | packaged by conda-forge | (main, Feb  5 2026, 05:53:46) [GCC 14.3.0] (64-bit runtime)
Python platform: Linux-6.8.0-107-generic-x86_64-with-glibc2.39
Is CUDA available: True
CUDA runtime version: 13.0.88
CUDA_MODULE_LOADING set to:
GPU models and configuration:
GPU 0: NVIDIA H100 80GB HBM3
GPU 1: NVIDIA H100 80GB HBM3
GPU 2: NVIDIA H100 80GB HBM3
GPU 3: NVIDIA H100 80GB HBM3
GPU 4: NVIDIA H100 80GB HBM3
GPU 5: NVIDIA H100 80GB HBM3
GPU 6: NVIDIA H100 80GB HBM3
GPU 7: NVIDIA H100 80GB HBM3

Nvidia driver version: 595.58.03
cuDNN version: Probably one of the following:
/usr/lib/x86_64-linux-gnu/libcudnn.so.9.14.0
/usr/lib/x86_64-linux-gnu/libcudnn_adv.so.9.14.0
/usr/lib/x86_64-linux-gnu/libcudnn_cnn.so.9.14.0
/usr/lib/x86_64-linux-gnu/libcudnn_engines_precompiled.so.9.14.0
/usr/lib/x86_64-linux-gnu/libcudnn_engines_runtime_compiled.so.9.14.0
/usr/lib/x86_64-linux-gnu/libcudnn_graph.so.9.14.0
/usr/lib/x86_64-linux-gnu/libcudnn_heuristic.so.9.14.0
/usr/lib/x86_64-linux-gnu/libcudnn_ops.so.9.14.0
Is XPU available: False
HIP runtime version: N/A
MIOpen runtime version: N/A
Is XNNPACK available: True
Caching allocator config: N/A

CPU:
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):                                  384
On-line CPU(s) list:                     0-383
Vendor ID:                               AuthenticAMD
Model name:                              AMD EPYC 9654 96-Core Processor
CPU family:                              25
Model:                                   17
Thread(s) per core:                      2
Core(s) per socket:                      96
Socket(s):                               2
Stepping:                                1
Frequency boost:                         enabled
CPU(s) scaling MHz:                      50%
CPU max MHz:                             3709.3569
CPU min MHz:                             1500.0000
BogoMIPS:                                4799.99
Flags:                                   fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ht syscall nx mmxext fxsr_opt pdpe1gb rdtscp lm constant_tsc rep_good amd_lbr_v2 nopl nonstop_tsc cpuid extd_apicid aperfmperf rapl pni pclmulqdq monitor ssse3 fma cx16 pcid sse4_1 sse4_2 x2apic movbe popcnt aes xsave avx f16c rdrand lahf_lm cmp_legacy svm extapic cr8_legacy abm sse4a misalignsse 3dnowprefetch osvw ibs skinit wdt tce topoext perfctr_core perfctr_nb bpext perfctr_llc mwaitx cpb cat_l3 cdp_l3 hw_pstate ssbd mba perfmon_v2 ibrs ibpb stibp ibrs_enhanced vmmcall fsgsbase bmi1 avx2 smep bmi2 erms invpcid cqm rdt_a avx512f avx512dq rdseed adx smap avx512ifma clflushopt clwb avx512cd sha_ni avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_local user_shstk avx512_bf16 clzero irperf xsaveerptr rdpru wbnoinvd amd_ppin cppc amd_ibpb_ret arat npt lbrv svm_lock nrip_save tsc_scale vmcb_clean flushbyasid decodeassists pausefilter pfthreshold avic v_vmsave_vmload vgif x2avic v_spec_ctrl vnmi avx512vbmi umip pku ospke avx512_vbmi2 gfni vaes vpclmulqdq avx512_vnni avx512_bitalg avx512_vpopcntdq la57 rdpid overflow_recov succor smca fsrm flush_l1d debug_swap ibpb_exit_to_user
Virtualization:                          AMD-V
L1d cache:                               6 MiB (192 instances)
L1i cache:                               6 MiB (192 instances)
L2 cache:                                192 MiB (192 instances)
L3 cache:                                768 MiB (24 instances)
NUMA node(s):                            2
NUMA node0 CPU(s):                       0-95,192-287
NUMA node1 CPU(s):                       96-191,288-383
Vulnerability Gather data sampling:      Not affected
Vulnerability Indirect target selection: 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:      Mitigation; Safe RET
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; STIBP always-on; PBRSB-eIBRS Not affected; BHI Not affected
Vulnerability Srbds:                     Not affected
Vulnerability Tsa:                       Vulnerable: Clear CPU buffers attempted, no microcode
Vulnerability Tsx async abort:           Not affected
Vulnerability Vmscape:                   Mitigation; IBPB before exit to userspace

Versions of relevant libraries:
[pip3] numpy==2.2.6
[pip3] nvidia-cublas==13.1.0.3
[pip3] nvidia-cuda-cupti==13.0.85
[pip3] nvidia-cuda-nvrtc==13.0.88
[pip3] nvidia-cuda-runtime==13.0.96
[pip3] nvidia-cudnn-cu13==9.19.0.56
[pip3] nvidia-cudnn-frontend==1.18.0
[pip3] nvidia-cufft==12.0.0.61
[pip3] nvidia-curand==10.4.0.35
[pip3] nvidia-cusolver==12.0.4.66
[pip3] nvidia-cusparse==12.6.3.3
[pip3] nvidia-cusparselt-cu13==0.8.0
[pip3] nvidia-nccl-cu13==2.28.9
[pip3] nvidia-nvjitlink==13.0.88
[pip3] nvidia-nvtx==13.0.85
[pip3] torch==2.11.0
[pip3] torch_c_dlpack_ext==0.1.5
[pip3] torchaudio==2.11.0
[pip3] torchvision==0.26.0
[pip3] triton==3.6.0
[conda] numpy                                       2.2.6                                    pypi_0                pypi
[conda] nvidia-cublas                               13.1.0.3                                 pypi_0                pypi
[conda] nvidia-cuda-cupti                           13.0.85                                  pypi_0                pypi
[conda] nvidia-cuda-nvrtc                           13.0.88                                  pypi_0                pypi
[conda] nvidia-cuda-runtime                         13.0.96                                  pypi_0                pypi
[conda] nvidia-cudnn-cu13                           9.19.0.56                                pypi_0                pypi
[conda] nvidia-cudnn-frontend                       1.18.0                                   pypi_0                pypi
[conda] nvidia-cufft                                12.0.0.61                                pypi_0                pypi
[conda] nvidia-curand                               10.4.0.35                                pypi_0                pypi
[conda] nvidia-cusolver                             12.0.4.66                                pypi_0                pypi
[conda] nvidia-cusparse                             12.6.3.3                                 pypi_0                pypi
[conda] nvidia-cusparselt-cu13                      0.8.0                                    pypi_0                pypi
[conda] nvidia-nccl-cu13                            2.28.9                                   pypi_0                pypi
[conda] nvidia-nvjitlink                            13.0.88                                  pypi_0                pypi
[conda] nvidia-nvtx                                 13.0.85                                  pypi_0                pypi
[conda] torch                                       2.11.0                                   pypi_0                pypi
[conda] torch-c-dlpack-ext                          0.1.5                                    pypi_0                pypi
[conda] torchaudio                                  2.11.0                                   pypi_0                pypi
[conda] torchvision                                 0.26.0                                   pypi_0                pypi
[conda] triton                                      3.6.0                                    pypi_0                pypi

---

PyTorch version: 2.10.0+git8514f05
Is debug build: False
CUDA used to build PyTorch: N/A
ROCM used to build PyTorch: 7.2.53211

OS: Ubuntu 22.04.5 LTS (x86_64)
GCC version: (Ubuntu 11.4.0-1ubuntu1~22.04.3) 11.4.0
Clang version: 22.0.0git (https://github.com/RadeonOpenCompute/llvm-project roc-7.2.1 26084 f58b06dce1f9c15707c5f808fd002e18c2accf7e)
CMake version: version 3.31.10
Libc version: glibc-2.35

Python version: 3.12.13 (main, Mar  4 2026, 09:23:07) [GCC 11.4.0] (64-bit runtime)
Python platform: Linux-6.8.0-101-generic-x86_64-with-glibc2.35
Is CUDA available: True
CUDA runtime version: Could not collect
CUDA_MODULE_LOADING set to:
GPU models and configuration:  (gfx942:sramecc+:xnack-)
Nvidia driver version: Could not collect
cuDNN version: Could not collect
Is XPU available: False
HIP runtime version: 7.2.53211
MIOpen runtime version: 3.5.1
Is XNNPACK available: True
Caching allocator config: N/A

CPU:
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):                                  256
On-line CPU(s) list:                     0-255
Vendor ID:                               AuthenticAMD
Model name:                              AMD EPYC 9555 64-Core Processor
CPU family:                              26
Model:                                   2
Thread(s) per core:                      2
Core(s) per socket:                      64
Socket(s):                               2
Stepping:                                1
Frequency boost:                         enabled
CPU max MHz:                             4409.3750
CPU min MHz:                             1500.0000
BogoMIPS:                                6399.98
Flags:                                   fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ht syscall nx mmxext fxsr_opt pdpe1gb rdtscp lm constant_tsc rep_good amd_lbr_v2 nopl nonstop_tsc cpuid extd_apicid aperfmperf rapl pni pclmulqdq monitor ssse3 fma cx16 pcid sse4_1 sse4_2 x2apic movbe popcnt aes xsave avx f16c rdrand lahf_lm cmp_legacy svm extapic cr8_legacy abm sse4a misalignsse 3dnowprefetch osvw ibs skinit wdt tce topoext perfctr_core perfctr_nb bpext perfctr_llc mwaitx cpb cat_l3 cdp_l3 hw_pstate ssbd mba perfmon_v2 ibrs ibpb stibp ibrs_enhanced vmmcall fsgsbase tsc_adjust bmi1 avx2 smep bmi2 erms invpcid cqm rdt_a avx512f avx512dq rdseed adx smap avx512ifma clflushopt clwb avx512cd sha_ni avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_local user_shstk avx_vnni avx512_bf16 clzero irperf xsaveerptr rdpru wbnoinvd amd_ppin cppc amd_ibpb_ret arat npt lbrv svm_lock nrip_save tsc_scale vmcb_clean flushbyasid decodeassists pausefilter pfthreshold avic v_vmsave_vmload vgif x2avic v_spec_ctrl vnmi avx512vbmi umip pku ospke avx512_vbmi2 gfni vaes vpclmulqdq avx512_vnni avx512_bitalg avx512_vpopcntdq la57 rdpid bus_lock_detect movdiri movdir64b overflow_recov succor smca fsrm avx512_vp2intersect flush_l1d debug_swap
Virtualization:                          AMD-V
L1d cache:                               6 MiB (128 instances)
L1i cache:                               4 MiB (128 instances)
L2 cache:                                128 MiB (128 instances)
L3 cache:                                512 MiB (16 instances)
NUMA node(s):                            2
NUMA node0 CPU(s):                       0-63,128-191
NUMA node1 CPU(s):                       64-127,192-255
Vulnerability Gather data sampling:      Not affected
Vulnerability Indirect target selection: 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; STIBP always-on; PBRSB-eIBRS Not affected; BHI Not affected
Vulnerability Srbds:                     Not affected
Vulnerability Tsa:                       Not affected
Vulnerability Tsx async abort:           Not affected
Vulnerability Vmscape:                   Not affected

Versions of relevant libraries:
[pip3] conch-triton-kernels==1.2.1
[pip3] numpy==2.1.3
[pip3] onnx==1.19.0
[pip3] onnx-ir==0.2.0
[pip3] onnxscript==0.6.2
[pip3] onnxslim==0.1.91
[pip3] torch==2.10.0+git8514f05
[pip3] torchaudio==2.9.0+eaa9e4e
[pip3] torchvision==0.24.1+d801a34
[pip3] triton==3.6.0
[pip3] triton_kernels==1.0.0
[conda] Could not collect