2023-05-04 02:43:17 +02:00
|
|
|
import logging
|
2023-04-10 04:08:40 +02:00
|
|
|
import math
|
|
|
|
import sys
|
2023-05-04 02:43:17 +02:00
|
|
|
from typing import Optional, Tuple
|
|
|
|
|
2023-04-10 04:08:40 +02:00
|
|
|
import torch
|
|
|
|
import torch.nn as nn
|
|
|
|
import transformers.models.llama.modeling_llama
|
|
|
|
|
|
|
|
import modules.shared as shared
|
|
|
|
|
|
|
|
if shared.args.xformers:
|
|
|
|
try:
|
|
|
|
import xformers.ops
|
|
|
|
except Exception:
|
2023-05-04 02:43:17 +02:00
|
|
|
logging.error("xformers not found! Please install it before trying to use it.", file=sys.stderr)
|
2023-04-10 04:08:40 +02:00
|
|
|
|
|
|
|
|
|
|
|
def hijack_llama_attention():
|
|
|
|
if shared.args.xformers:
|
|
|
|
transformers.models.llama.modeling_llama.LlamaAttention.forward = xformers_forward
|
2023-05-04 02:43:17 +02:00
|
|
|
logging.info("Replaced attention with xformers_attention")
|
2023-04-10 04:08:40 +02:00
|
|
|
elif shared.args.sdp_attention:
|
|
|
|
transformers.models.llama.modeling_llama.LlamaAttention.forward = sdp_attention_forward
|
2023-05-04 02:43:17 +02:00
|
|
|
logging.info("Replaced attention with sdp_attention")
|
2023-04-10 04:08:40 +02:00
|
|
|
|
|
|
|
|
|
|
|
def xformers_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,
|
|
|
|
) -> 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 = transformers.models.llama.modeling_llama.apply_rotary_pos_emb(query_states, key_states, cos, sin, position_ids)
|
|
|
|
# [bsz, nh, t, hd]
|
|
|
|
|
|
|
|
if past_key_value is not None:
|
|
|
|
# reuse k, v, self_attention
|
|
|
|
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
|
|
|
|
|
2023-05-04 02:43:17 +02:00
|
|
|
# We only apply xformers optimizations if we don't need to output the whole attention matrix
|
2023-04-10 04:08:40 +02:00
|
|
|
if not output_attentions:
|
|
|
|
query_states = query_states.transpose(1, 2)
|
|
|
|
key_states = key_states.transpose(1, 2)
|
|
|
|
value_states = value_states.transpose(1, 2)
|
2023-05-04 02:43:17 +02:00
|
|
|
|
|
|
|
# This is a nasty hack. We know attention_mask in transformers is either LowerTriangular or all Zeros.
|
|
|
|
# We therefore check if one element in the upper triangular portion is zero. If it is, then the mask is all zeros.
|
2023-04-10 04:08:40 +02:00
|
|
|
if attention_mask is None or attention_mask[0, 0, 0, 1] == 0:
|
|
|
|
# input and output should be of form (bsz, q_len, num_heads, head_dim)
|
|
|
|
attn_output = xformers.ops.memory_efficient_attention(query_states, key_states, value_states, attn_bias=None)
|
|
|
|
else:
|
|
|
|
# input and output should be of form (bsz, q_len, num_heads, head_dim)
|
|
|
|
attn_output = xformers.ops.memory_efficient_attention(query_states, key_states, value_states, attn_bias=xformers.ops.LowerTriangularMask())
|
|
|
|
attn_weights = None
|
|
|
|
else:
|
|
|
|
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))
|
|
|
|
|
|
|
|
# upcast attention to fp32
|
|
|
|
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)
|
|
|
|
return attn_output, attn_weights, past_key_value
|
|
|
|
|
|
|
|
|
|
|
|
def sdp_attention_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,
|
|
|
|
) -> 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 = transformers.models.llama.modeling_llama.apply_rotary_pos_emb(query_states, key_states, cos, sin, position_ids)
|
|
|
|
# [bsz, nh, t, hd]
|
|
|
|
|
|
|
|
if past_key_value is not None:
|
|
|
|
# reuse k, v, self_attention
|
|
|
|
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
|
|
|
|
|
2023-05-04 02:43:17 +02:00
|
|
|
# We only apply sdp attention if we don't need to output the whole attention matrix
|
2023-04-10 04:08:40 +02:00
|
|
|
if not output_attentions:
|
|
|
|
attn_output = torch.nn.functional.scaled_dot_product_attention(query_states, key_states, value_states, attn_mask=attention_mask, is_causal=False)
|
|
|
|
attn_weights = None
|
|
|
|
else:
|
|
|
|
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))
|
|
|
|
|
|
|
|
# upcast attention to fp32
|
|
|
|
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)
|
|
|
|
|
|
|
|
return attn_output, attn_weights, past_key_value
|