For the complete documentation index, see llms.txt. Markdown versions of all pages are available by appending .md to any URL (e.g. /max/get-started.md).

Mojo function

scatter_pull_kernel

def scatter_pull_kernel[dtype: DType, BLOCK_SIZE: Int, ngpus: Int, tp_size: Int, dp_size: Int, simd_width: Int = simd_width_of[dtype, get_gpu_target()]()](output_ptr: UnsafePointer[Scalar[dtype], MutAnyOrigin], input_ptrs: InlineArray[UnsafePointer[Scalar[dtype], ImmutAnyOrigin], dp_size], chunk_num_elems: InlineArray[Int, dp_size], rank_sigs: InlineArray[UnsafePointer[Signal, MutAnyOrigin], 8], my_rank: Int)

Pull-based scatter+broadcast: each GPU reads its chunk from root.

Each GPU determines its replica index (my_rank // tp_size), then copies from input_ptrs[replica] on the root GPU to its own output buffer.

Was this page helpful?

Thank you! We'll create more content like this.

Thank you for helping us improve!

😔 What went wrong?

Some code doesn’t work

It includes inaccurate information

It's missing information I need

It was difficult to understand

Other

Submit