Issue with the Xformers.

[rrunner77]

Hi,

I am getting this error when running stable diffusion. I seen that the last development (3 weeks ago) should solve this topic, but may be something need to be changed also on the side of A1111 or stable diffusion.

`ckF` is not supported because:
    max(query.shape[-1], value.shape[-1]) > 256
    dtype=torch.float32 (supported: {torch.bfloat16, torch.float16})

I have compiled the Xformers from this source code:
pip3.10 install -v -U git+https://github.com/ROCm/xformers@for_upstream
and it seems to be working fine when I run:

python -m xformers.info
xFormers 0.0.30+4cfab36.d20250207
memory_efficient_attention.ckF:                    available
memory_efficient_attention.ckB:                    available
memory_efficient_attention.ck_decoderF:            available
memory_efficient_attention.ck_splitKF:             available
memory_efficient_attention.cutlassF-pt:            available
memory_efficient_attention.cutlassB-pt:            available
memory_efficient_attention.fa2F@0.0.0:             unavailable
memory_efficient_attention.fa2B@0.0.0:             unavailable
memory_efficient_attention.fa3F@0.0.0:             unavailable
memory_efficient_attention.fa3B@0.0.0:             unavailable
memory_efficient_attention.triton_splitKF:         available
indexing.scaled_index_addF:                        available
indexing.scaled_index_addB:                        available
indexing.index_select:                             available
sequence_parallel_fused.write_values:              available
sequence_parallel_fused.wait_values:               available
sequence_parallel_fused.cuda_memset_32b_async:     available
sp24.sparse24_sparsify_both_ways:                  available
sp24.sparse24_apply:                               available
sp24.sparse24_apply_dense_output:                  available
sp24._sparse24_gemm:                               available
sp24._cslt_sparse_mm_search@0.0.0:                 available
sp24._cslt_sparse_mm@0.0.0:                        available
swiglu.dual_gemm_silu:                             available
swiglu.gemm_fused_operand_sum:                     available
swiglu.fused.p.cpp:                                available
is_triton_available:                               True
pytorch.version:                                   2.7.0.dev20250206+rocm6.3
pytorch.cuda:                                      available
gpu.compute_capability:                            11.0
gpu.name:                                          AMD Radeon RX 7900 XTX
dcgm_profiler:                                     unavailable
build.info:                                        available
build.cuda_version:                                None
build.hip_version:                                 None
build.python_version:                              3.10.16
build.torch_version:                               2.7.0.dev20250206+rocm6.3
build.env.TORCH_CUDA_ARCH_LIST:                    None
build.env.PYTORCH_ROCM_ARCH:                       None
build.env.XFORMERS_BUILD_TYPE:                     None
build.env.XFORMERS_ENABLE_DEBUG_ASSERTIONS:        None
build.env.NVCC_FLAGS:                              None
build.env.XFORMERS_PACKAGE_FROM:                   None
source.privacy:                                    open source

even I see that the cutlassF-pt and cutlassB-pt are now available.

I am pretty new to this topic so may be I am asking stupid questions.

But when running without --precision full --no-half there is a very different error. I assume that the Xformers are ok but torch or A1111 can not work with them for some reason