Lower text and remove punctuation, extra whitespace.

Lower text and remove punctuation, articles and extra whitespace.

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.

Naive constant-space memory following the recurrence formula M_t = f(M_{t-1}, q).

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

Return set of symbol pairs in a word. Word is represented as tuple of symbols (symbols being variable-length strings).

Lower text and remove punctuation, articles and extra whitespace.

Lower text and remove punctuation, extra whitespace.

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

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

Returns list of utf-8 byte and a mapping to unicode strings. We specifically avoids mapping to whitespace/control characters the bpe code bar

Return model weight files saved by HF `save_pretrained`. Supports: - *.safetensors (sharded or single) - pytorch_model*.bin (shar

Padding x tensor with `pad_size` on the seq_len dim (dim=1) Assumes that we only have tensors of either size 4 or 3

Post-process the response. Args: response (str): Model response. model_name (str): Model name. Returns: str: Po

Build FlexAttention mask with custom sliding window and attention sink.

Build FlexAttention mask with custom sliding window and attention sink.