AttributeError: module 'torch.xpu' has no attribute 'has_2d_block_array'

[kdbeall]

Describe the bug

Hi! I'm running the latest 2.7 Pytorch+XPU and 2.7 IPEX
I'm using the Intel Arc B580 card.

I end up with the error

AttributeError: module 'torch.xpu' has no attribute 'has_2d_block_array' #831

import torch
from transformers import AutoTokenizer, AutoModel
import intel_extension_for_pytorch as ipex  # Triggers error internally

import torch.nn.functional as F

def mean_pooling(last_hidden_state, attention_mask):
    mask = attention_mask.unsqueeze(-1).expand(last_hidden_state.size()).float()
    return torch.sum(last_hidden_state * mask, 1) / torch.clamp(mask.sum(1), min=1e-9)

# Setup
model_name = "nomic-ai/modernbert-embed-base"
device = torch.device("xpu:1")

# Load tokenizer and model
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModel.from_pretrained(model_name).to(device).half().eval()

# This line triggers the AttributeError
model = ipex.optimize(model, dtype=torch.float16, inplace=True)

# Input text
texts = ["What is TSNE?", "How do vector embeddings work?"]
inputs = tokenizer(texts, return_tensors="pt", padding=True, truncation=True)
inputs = {k: v.to(device).half() for k, v in inputs.items()}

# Inference
with torch.no_grad():
    out = model(**inputs)
    pooled = mean_pooling(out.last_hidden_state, inputs["attention_mask"])
    embeddings = F.normalize(pooled, p=2, dim=1)

print("Shape:", embeddings.shape)
print("First 5 dims:", embeddings[0][:5].tolist())

Versions

collect_env.py fails with

[W523 22:33:19.194673362 OperatorEntry.cpp:154] Warning: Warning only once for all operators,  other operators may also be overridden.
  Overriding a previously registered kernel for the same operator and the same dispatch key
  operator: aten::_addmm_activation(Tensor self, Tensor mat1, Tensor mat2, *, Scalar beta=1, Scalar alpha=1, bool use_gelu=False) -> Tensor
    registered at /pytorch/build/aten/src/ATen/RegisterSchema.cpp:6
  dispatch key: AutocastCPU
  previous kernel: registered at /pytorch/aten/src/ATen/autocast_mode.cpp:327
       new kernel: registered at /opt/workspace/ipex-cpu-dev/csrc/cpu/autocast/autocast_mode.cpp:112 (function operator())
Collecting environment information...
Traceback (most recent call last):
  File "/opt/backboost/temp/collect_env.py", line 616, in <module>
    main()
  File "/opt/backboost/temp/collect_env.py", line 610, in main
    output = get_pretty_env_info()
             ^^^^^^^^^^^^^^^^^^^^^
  File "/opt/backboost/temp/collect_env.py", line 605, in get_pretty_env_info
    return pretty_str(get_env_info())
                      ^^^^^^^^^^^^^^
  File "/opt/backboost/temp/collect_env.py", line 449, in get_env_info
    pyenv, pip_list_output = get_python_packages(run_lambda)
                             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/opt/backboost/temp/collect_env.py", line 442, in get_python_packages
    pkgs_filtered = filter_python_packages(out)
                    ^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/opt/backboost/temp/collect_env.py", line 407, in filter_python_packages
    for line in data.splitlines()
                ^^^^^^^^^^^^^^^
AttributeError: 'NoneType' object has no attribute 'splitlines'

[jingxu10]

Could you share a full log of the no attribute error?

[kdbeall]

@jingxu10 Solved it with monkey patching

[Unbreak4ble]

@jingxu10 Solved it with monkey patching

can you give more detail?

[kdbeall]

if hasattr(torch, "xpu") and not hasattr(torch.xpu, "has_2d_block_array"):
    torch.xpu.has_2d_block_array = lambda: False