General test for verifying spike_matmul matches dense matmul output when using a given SpikeCount LIF quantizer. Args: x (torch.

Test function for SpikeCountBitwiseNode using spike_fake_quant(). Args: x (torch.Tensor, optional): Input tensor with integer values

Updates the cache with the new `recurrent_state`/`attn_state`/`conv_state` for the layer `layer_idx`. Args: recurrent_st

Test function for SpikeCountBinaryLIFNode. Args: x (torch.Tensor, optional): Input tensor with non-negative integer values.

Test function for SpikeCountTernaryLIFNode. Args: x (torch.Tensor, optional): Input tensor with integer values.

Returns the sequence length of the cached states. A layer index can be optionally passed.

Calculate the average firing rate of the spike sequence. For bidirectional coding: use absolute spike values. For unidirecti

Updates the cache with the new `recurrent_state`/`attn_state`/`conv_state` for the layer `layer_idx`. Args: recurrent_st

This method forwards all its arguments to Qwen2TokenizerFast's [`~PreTrainedTokenizer.batch_decode`]. Please refer to the docstring o

This is the equivalent of torch.repeat_interleave(x, dim=1, repeats=n_rep). The hidden states go from (batch, num_key_value_heads, seqlen, he

Rotates half the hidden dims of the input.

Simulate spike-based quantization and dequantization (fake quantization), with optional zero-point shifting. Args: x (torch.Tens

This is the equivalent of torch.repeat_interleave(x, dim=1, repeats=n_rep). The hidden states go from (batch, num_key_value_heads, seqlen, he

This is the equivalent of torch.repeat_interleave(x, dim=1, repeats=n_rep). The hidden states go from (batch, num_key_value_heads, seqlen, he

This is the equivalent of torch.repeat_interleave(x, dim=1, repeats=n_rep). The hidden states go from (batch, num_key_value_heads, seqlen, he

Quantize an integer tensor into spike sequences using a spike encoder. Supports optional zero-point shifting for bidirectional or unipolar mo

Creates a causal 4D mask of shape `(batch_size, 1, query_length, key_value_length)` from a 2D mask of shape `(batch_size, key_value_l

Applies Rotary Position Embedding with Multimodal Sections to the query and key tensors Explanation: Multimodal 3D rotary position embedd

Converts a cache in the legacy cache format into an equivalent `Cache`.

Converts a cache in the legacy cache format into an equivalent `Cache`.

Returns the sequence length of the cached states. A layer index can be optionally passed.

x: [b, l, d]

This is the equivalent of torch.repeat_interleave(x, dim=1, repeats=n_rep). The hidden states go from (batch, num_key_value_heads, seqlen, he

This is the equivalent of torch.repeat_interleave(x, dim=1, repeats=n_rep). The hidden states go from (batch, num_key_value_heads, seqlen, he

This is the equivalent of torch.repeat_interleave(x, dim=1, repeats=n_rep). The hidden states go from (batch, num_key_value_heads, seqlen, he

Rotates half the hidden dims of the input.

Rotates half the hidden dims of the input.

Rotates half the hidden dims of the input.

Convert integer weight tensor into its bitwise representation. - Ensures `w` is integer-valued. - If `w` contains negative values, shift

Add a sequence group to the metadata. Specifically update/append 1. context length. 2. block table. 3. slot mapping.

Create a boolean attention mask for sliding-window + causal attention. True = masked, False = visible.

dynamic RoPE layers should recompute `inv_freq` in the following situations: 1 - growing beyond the cached sequence length (allow sca

Determine whether the given attention type is suitable for causal attention mechanisms. Args: attn_type (AttentionType): The ty

Get the number of images and videos for each sample to calculate the separation length of the sample tensor. These parameters are not

Returns sequence metadata for key and query based on the specified attention type and whether input is a prompt. This function computes

Applies Rotary Position Embedding to the query and key tensors. Args: q (`torch.Tensor`): The query tensor. k (`torch.Tensor`): T

Applies Rotary Position Embedding to the query and key tensors. Args: q (`torch.Tensor`): The query tensor. k (`torch.Tensor`): T

Applies Rotary Position Embedding to the query and key tensors. Args: q (`torch.Tensor`): The query tensor. k (`torch.Tensor`): T

Assume that y has shape (B, D) and bias has shape (D)

Assume that y has shape (B, D) and bias has shape (D)

Full dequantization: decode spike sequence to pulse count, then map to float using dequantizer. Args: spike (torch.Tensor): Spike se

Get Python package dependencies from requirements.txt.

Calculate the 3D rope index based on image and video's temporal, height and width in LLM. Explanation: Each embedding se