spikingjelly.activation_based.triton_kernel.compress module#

spikingjelly.activation_based.triton_kernel.compress.bit_spike_compress(s_seq)[源代码]#

Compress a float32 spike tensor into a compact uint8 representation using bit-packing. API Language: 中文 | English

  • 中文

对脉冲张量进行位压缩

参数:

s_seq (torch.Tensor) -- Spike sequence tensor of float32

返回:

Compressed uint8 tensor (8x smaller)

返回类型:

torch.Tensor

Each element is rounded to 0 or 1 (by threshold 0.5) and packed as a single bit. Works on both CPU and GPU (via Triton kernel on CUDA).

  • English

Bit-compress a spike tensor

参数:

s_seq (torch.Tensor) -- Spike sequence tensor of float32

返回:

Compressed uint8 tensor (8x smaller)

返回类型:

torch.Tensor

spikingjelly.activation_based.triton_kernel.compress.bit_spike_decompress(s_seq_compressed, shape)[源代码]#

Decompress a uint8 bit-packed tensor back to a float32 spike tensor. API Language: 中文 | English

  • 中文

解压位压缩的脉冲张量

参数:

  • s_seq_compressed (torch.Tensor) -- Compressed uint8 tensor from bit_spike_compress()

  • shape (tuple) -- Original shape of the uncompressed tensor

返回:

Decompressed float32 spike tensor (values are 0.0 or 1.0)

返回类型:

torch.Tensor

  • English

Decompress a bit-compressed spike tensor

参数:

  • s_seq_compressed (torch.Tensor) -- Compressed uint8 tensor from bit_spike_compress()

  • shape (tuple) -- Original shape of the uncompressed tensor

返回:

Decompressed float32 spike tensor (values are 0.0 or 1.0)

返回类型:

torch.Tensor