class FeedForward(nn.Module):
def __init__(self, config: LMConfig):
super().__init__()
if config.hidden_dim is None:
hidden_dim = 4 * config.dim
hidden_dim = int(2 * hidden_dim / 3)
config.hidden_dim = config.multiple_of * ((hidden_dim + config.multiple_of - 1) // config.multiple_of)
self.w1 = nn.Linear(config.dim, config.hidden_dim, bias=False)
self.w2 = nn.Linear(config.hidden_dim, config.dim, bias=False)
self.w3 = nn.Linear(config.dim, config.hidden_dim, bias=False)
self.dropout = nn.Dropout(config.dropout)
def forward(self, x):
return self.dropout(self.w2(F.silu(self.w1(x)) * self.w3(x)))
class MoEGate(nn.Module):
def __init__(self, config: LMConfig):
super().__init__()
self.config = config
self.top_k = config.num_experts_per_tok
self.n_routed_experts = config.n_routed_experts
self.scoring_func = config.scoring_func
self.alpha = config.aux_loss_alpha
self.seq_aux = config.seq_aux
self.norm_topk_prob = config.norm_topk_prob
self.gating_dim = config.dim
self.weight = nn.Parameter(torch.empty((self.n_routed_experts, self.gating_dim)))
self.reset_parameters()
def reset_parameters(self) -> None:
import torch.nn.init as init
init.kaiming_uniform_(self.weight, a=math.sqrt(5))
def forward(self, hidden_states):
bsz, seq_len, h = hidden_states.shape
hidden_states = hidden_states.view(-1, h)
logits = F.linear(hidden_states, self.weight, None)
if self.scoring_func == 'softmax':
scores = logits.softmax(dim=-1)
else:
raise NotImplementedError(f'insupportable scoring function for MoE gating: {self.scoring_func}')
topk_weight, topk_idx = torch.topk(scores, k=self.top_k, dim=-1, sorted=False)
if self.top_k > 1 and self.norm_topk_prob:
denominator = topk_weight.sum(dim=-1, keepdim=True) + 1e-20
topk_weight = topk_weight / denominator
if self.training and self.alpha > 0.0:
scores_for_aux = scores
aux_topk = self.top_k
topk_idx_for_aux_loss = topk_idx.view(bsz, -1)
if self.seq_aux:
scores_for_seq_aux = scores_for_aux.view(bsz, seq_len, -1)
ce = torch.zeros(bsz, self.n_routed_experts, device=hidden_states.device)
ce.scatter_add_(1, topk_idx_for_aux_loss,
torch.ones(bsz, seq_len * aux_topk, device=hidden_states.device)).div_(
seq_len * aux_topk / self.n_routed_experts)
aux_loss = (ce * scores_for_seq_aux.mean(dim=1)).sum(dim=1).mean() * self.alpha
else:
mask_ce = F.one_hot(topk_idx_for_aux_loss.view(-1), num_classes=self.n_routed_experts)
ce = mask_ce.float().mean(0)
Pi = scores_for_aux.mean(0)
fi = ce * self.n_routed_experts
aux_loss = (Pi * fi).sum() * self.alpha
else:
aux_loss = 0
return topk_idx, topk_weight, aux_loss
class MOEFeedForward(nn.Module):
def __init__(self, config: LMConfig):
super().__init__()
self.config = config
self.experts = nn.ModuleList([
FeedForward(config)
for _ in range(config.n_routed_experts)
])
self.gate = MoEGate(config)
if config.n_shared_experts is not None:
self.shared_experts = FeedForward(config)
def forward(self, x):
identity = x
orig_shape = x.shape
bsz, seq_len, _ = x.shape
topk_idx, topk_weight, aux_loss = self.gate(x)
x = x.view(-1, x.shape[-1])
flat_topk_idx = topk_idx.view(-1)
if self.training:
x = x.repeat_interleave(self.config.num_experts_per_tok, dim=0)
y = torch.empty_like(x, dtype=torch.float16)
for i, expert in enumerate(self.experts):
y[flat_topk_idx == i] = expert(x[flat_topk_idx == i]).to(y.dtype)
y = (y.view(*topk_weight.shape, -1) * topk_weight.unsqueeze(-1)).sum(dim=1)
y = y.view(*orig_shape)
else:
y = self.moe_infer(x, flat_topk_idx, topk_weight.view(-1, 1)).view(*orig_shape)
if self.config.n_shared_experts is not None:
y = y + self.shared_experts(identity)
self.aux_loss = aux_loss
return y
@torch.no_grad()
def moe_infer(self, x, flat_expert_indices, flat_expert_weights):
expert_cache = torch.zeros_like(x)
idxs = flat_expert_indices.argsort()
tokens_per_expert = flat_expert_indices.bincount().cpu().numpy().cumsum(0)
token_idxs = idxs // self.config.num_experts_per_tok
for i, end_idx in enumerate(tokens_per_expert):
start_idx = 0 if i == 0 else tokens_per_expert[i - 1]
if start_idx == end_idx:
continue
expert = self.experts[i]
exp_token_idx = token_idxs[start_idx:end_idx]
expert_tokens = x[exp_token_idx]
expert_out = expert(expert_tokens).to(expert_cache.dtype)
expert_out.mul_(flat_expert_weights[idxs[start_idx:end_idx]])
expert_cache.scatter_add_(0, exp_token_idx.view(-1, 1).repeat(1, x.shape[-1]), expert_out)
return expert_cache