huggingface/transformers · error · ImportError

DeepGEMM kernel is missing required symbols: {', '.join(miss

Error message

DeepGEMM kernel is missing required symbols: {', '.join(missing)}. `kernels` is either not installed or uses an incompatible version. Please install a compatible version ({KERNELS_MIN_VERSION} <= version < {KERNELS_MAX_VERSION}), e.g. `pip install kernels=={KERNELS_MIN_VERSION}`

What it means

Error "DeepGEMM kernel is missing required symbols: {', '.join(missing)}. `kernels` is either not installed or uses an incompatible version. Please install a compatible version ({KERNELS_MIN_VERSION} <= version < {KERNELS_MAX_VERSION}), e.g. `pip install kernels=={KERNELS_MIN_VERSION}`" thrown in huggingface/transformers.

Source

Thrown at src/transformers/integrations/deepgemm.py:284

    missing = [
        name
        for name, attr in [
            ("fp8_fp4_gemm_nt", fp8_fp4_matmul),
            ("m_grouped_fp8_fp4_gemm_nt_contiguous", grouped_fp8_fp4_matmul_nt),
            ("m_grouped_fp8_fp4_gemm_nn_contiguous", grouped_fp8_fp4_matmul_nn),
            ("m_grouped_bf16_gemm_nt_contiguous", grouped_bf16_matmul_nt),
            ("m_grouped_bf16_gemm_nn_contiguous", grouped_bf16_matmul_nn),
            ("utils.per_token_cast_to_fp8", per_token_cast_to_fp8),
            ("transform_sf_into_required_layout", transform_sf_into_required_layout),
            ("transform_weights_for_mega_moe", transform_weights_for_mega_moe),
            ("get_symm_buffer_for_mega_moe", get_symm_buffer_for_mega_moe),
            ("get_mk_alignment_for_contiguous_layout", get_mk_alignment),
            ("fp8_fp4_mega_moe", fp8_fp4_mega_moe),
        ]
        if attr is None
    ]
    if missing:
        raise ImportError(
            f"DeepGEMM kernel is missing required symbols: {', '.join(missing)}. {_MISSING_KERNELS_MESSAGE}"
        )

    _DEEPGEMM = DeepGEMM(
        fp8_fp4_matmul=fp8_fp4_matmul,
        grouped_fp8_fp4_matmul_nt=grouped_fp8_fp4_matmul_nt,
        grouped_fp8_fp4_matmul_nn=grouped_fp8_fp4_matmul_nn,
        grouped_bf16_matmul_nt=grouped_bf16_matmul_nt,
        grouped_bf16_matmul_nn=grouped_bf16_matmul_nn,
        per_token_cast_to_fp8=per_token_cast_to_fp8,
        transform_sf_into_required_layout=transform_sf_into_required_layout,
        transform_weights_for_mega_moe=transform_weights_for_mega_moe,
        get_symm_buffer_for_mega_moe=get_symm_buffer_for_mega_moe,
        fp8_fp4_mega_moe=fp8_fp4_mega_moe,
        m_alignment=get_mk_alignment(),
    )

View on GitHub (pinned to a597f97485)

Solutions

1. Install a compatible kernels version within the stated range, e.g. `pip install kernels==<min_version>`.

When it happens

Trigger: Raised when the loaded DeepGEMM kernel lacks required symbols because the kernels package version is incompatible.

Common situations: kernels package missing or outside the supported version range while loading a DeepGEMM-quantized model.

AI-assisted analysis of huggingface/transformers@a597f97485 (2026-08-14). Data as JSON: /api/errors/381a2736e369e50d. Report an issue: GitHub.