Spaces:
Sleeping
Sleeping
import torch | |
import torch.nn as nn | |
import gpt_config as config | |
from head import Head | |
class MultiHeadAttention(nn.Module): | |
""" multiple heads of self-attention in parallel """ | |
def __init__(self, num_heads, head_size): | |
super().__init__() | |
self.heads = nn.ModuleList([Head(head_size) for _ in range(num_heads)]) | |
self.proj = nn.Linear(head_size * num_heads, config.n_embd) | |
self.dropout = nn.Dropout(config.dropout) | |
def forward(self, x): | |
out = torch.cat([h(x) for h in self.heads], dim=-1) | |
out = self.dropout(self.proj(out)) | |
return out |