|
""" |
|
Monkey patch the llama implementation in the huggingface/transformers library. |
|
Avoid bugs in mps backend by not using in-place operations. |
|
""" |
|
import math |
|
from typing import List, Optional, Tuple |
|
|
|
import torch |
|
from torch import nn |
|
import transformers |
|
|
|
|
|
def rotate_half(x): |
|
"""Rotates half the hidden dims of the input.""" |
|
x1 = x[..., : x.shape[-1] // 2].clone() |
|
x2 = x[..., x.shape[-1] // 2 :].clone() |
|
return torch.cat((-x2, x1), dim=-1) |
|
|
|
|
|
def apply_rotary_pos_emb(q, k, cos, sin, position_ids): |
|
gather_indices = position_ids[:, None, :, None] |
|
gather_indices = gather_indices.repeat(1, cos.shape[1], 1, cos.shape[3]) |
|
cos = torch.gather(cos.repeat(gather_indices.shape[0], 1, 1, 1), 2, gather_indices) |
|
sin = torch.gather(sin.repeat(gather_indices.shape[0], 1, 1, 1), 2, gather_indices) |
|
q_embed = (q * cos) + (rotate_half(q) * sin) |
|
k_embed = (k * cos) + (rotate_half(k) * sin) |
|
return q_embed, k_embed |
|
|
|
|
|
def forward( |
|
self, |
|
hidden_states: torch.Tensor, |
|
attention_mask: Optional[torch.Tensor] = None, |
|
position_ids: Optional[torch.LongTensor] = None, |
|
past_key_value: Optional[Tuple[torch.Tensor]] = None, |
|
output_attentions: bool = False, |
|
use_cache: bool = False, |
|
padding_mask: Optional[torch.LongTensor] = None, |
|
) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]: |
|
bsz, q_len, _ = hidden_states.size() |
|
|
|
query_states = ( |
|
self.q_proj(hidden_states) |
|
.view(bsz, q_len, self.num_heads, self.head_dim) |
|
.transpose(1, 2) |
|
) |
|
key_states = ( |
|
self.k_proj(hidden_states) |
|
.view(bsz, q_len, self.num_heads, self.head_dim) |
|
.transpose(1, 2) |
|
) |
|
value_states = ( |
|
self.v_proj(hidden_states) |
|
.view(bsz, q_len, self.num_heads, self.head_dim) |
|
.transpose(1, 2) |
|
) |
|
|
|
kv_seq_len = key_states.shape[-2] |
|
if past_key_value is not None: |
|
kv_seq_len += past_key_value[0].shape[-2] |
|
cos, sin = self.rotary_emb(value_states, seq_len=kv_seq_len) |
|
query_states, key_states = apply_rotary_pos_emb( |
|
query_states, key_states, cos, sin, position_ids |
|
) |
|
|
|
|
|
if past_key_value is not None: |
|
|
|
key_states = torch.cat([past_key_value[0], key_states], dim=2) |
|
value_states = torch.cat([past_key_value[1], value_states], dim=2) |
|
|
|
past_key_value = (key_states, value_states) if use_cache else None |
|
|
|
attn_weights = torch.matmul(query_states, key_states.transpose(2, 3)) / math.sqrt( |
|
self.head_dim |
|
) |
|
|
|
if attn_weights.size() != (bsz, self.num_heads, q_len, kv_seq_len): |
|
raise ValueError( |
|
f"Attention weights should be of size {(bsz * self.num_heads, q_len, kv_seq_len)}, but is" |
|
f" {attn_weights.size()}" |
|
) |
|
|
|
if attention_mask is not None: |
|
if attention_mask.size() != (bsz, 1, q_len, kv_seq_len): |
|
raise ValueError( |
|
f"Attention mask should be of size {(bsz, 1, q_len, kv_seq_len)}, but is {attention_mask.size()}" |
|
) |
|
attn_weights = attn_weights + attention_mask |
|
attn_weights = torch.max( |
|
attn_weights, torch.tensor(torch.finfo(attn_weights.dtype).min) |
|
) |
|
|
|
|
|
attn_weights = nn.functional.softmax(attn_weights, dim=-1, dtype=torch.float32).to( |
|
query_states.dtype |
|
) |
|
attn_output = torch.matmul(attn_weights, value_states) |
|
|
|
if attn_output.size() != (bsz, self.num_heads, q_len, self.head_dim): |
|
raise ValueError( |
|
f"`attn_output` should be of size {(bsz, self.num_heads, q_len, self.head_dim)}, but is" |
|
f" {attn_output.size()}" |
|
) |
|
|
|
attn_output = attn_output.transpose(1, 2) |
|
attn_output = attn_output.reshape(bsz, q_len, self.hidden_size) |
|
|
|
attn_output = self.o_proj(attn_output) |
|
|
|
if not output_attentions: |
|
attn_weights = None |
|
|
|
return attn_output, attn_weights, past_key_value |
|
|
|
|
|
def replace_llama_attn_with_non_inplace_operations(): |
|
"""Avoid bugs in mps backend by not using in-place operations.""" |
|
transformers.models.llama.modeling_llama.LlamaAttention.forward = forward |
|
|
