Skip to content

[Question]: Can I add more supported head sizes for attention_v1? #759

@hoangvictor

Description

@hoangvictor

Problem Description

In this file aiter/csrc/kernels/attention_v1.cu, can I add a case of 256 for the head_size?

#define CALL_CUSTOM_LAUNCHER_BLK_HEAD(T, KVT, KV_DTYPE)                       \
    switch(head_size)                                                         \
    {                                                                         \
    case 64: CALL_CUSTOM_LAUNCHER_BLK(T, KVT, KV_DTYPE, 64); break;           \
    case 128: CALL_CUSTOM_LAUNCHER_BLK(T, KVT, KV_DTYPE, 128); break;         \
    default: TORCH_CHECK(false, "Unsupported head size: ", head_size); break; \
    }

Operating System

22.04.5 LTS (Jammy Jellyfish)

CPU

Intel(R) Xeon(R) Platinum 8468V

GPU

AMD Instinct MI300X

ROCm Version

ROCM 6.3.1

ROCm Component

No response

Steps to Reproduce

No response

(Optional for Linux users) Output of /opt/rocm/bin/rocminfo --support

No response

Additional Information

No response

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions