Skip to content

🧠 两小时从 0 训出你的 64M 大模型。Train your own 64M LLM from scratch in 2 hours.

[update] open causal

jingyaogong2ab6455d9dbf
diff --git a/model/model_minimind.py b/model/model_minimind.py
index 70ee32b..d3518ad 100755
--- a/model/model_minimind.py
+++ b/model/model_minimind.py
@@ -95,6 +95,7 @@ class Attention(nn.Module):
self.n_local_kv_heads = self.num_key_value_heads
self.n_rep = self.n_local_heads // self.n_local_kv_heads
self.head_dim = config.head_dim
+ self.is_causal = True
self.q_proj = nn.Linear(config.hidden_size, config.num_attention_heads * self.head_dim, bias=False)
self.k_proj = nn.Linear(config.hidden_size, self.num_key_value_heads * self.head_dim, bias=False)
self.v_proj = nn.Linear(config.hidden_size, self.num_key_value_heads * self.head_dim, bias=False)
@@ -120,11 +121,11 @@ class Attention(nn.Module):
xv = torch.cat([past_key_value[1], xv], dim=1)
past_kv = (xk, xv) if use_cache else None
xq, xk, xv = (xq.transpose(1, 2), repeat_kv(xk, self.n_rep).transpose(1, 2), repeat_kv(xv, self.n_rep).transpose(1, 2))
- if self.flash and (seq_len > 1) and (past_key_value is None) and (attention_mask is None or torch.all(attention_mask == 1)):
- output = F.scaled_dot_product_attention(xq, xk, xv, dropout_p=self.dropout if self.training else 0.0, is_causal=True)
+ if self.flash and (seq_len > 1) and (not self.is_causal or past_key_value is None) and (attention_mask is None or torch.all(attention_mask == 1)):
+ output = F.scaled_dot_product_attention(xq, xk, xv, dropout_p=self.dropout if self.training else 0.0, is_causal=self.is_causal)
else:
scores = (xq @ xk.transpose(-2, -1)) / math.sqrt(self.head_dim)
- scores[:, :, :, -seq_len:] += torch.full((seq_len, seq_len), float("-inf"), device=scores.device).triu(1)
+ if self.is_causal: scores[:, :, :, -seq_len:] += torch.full((seq_len, seq_len), float("-inf"), device=scores.device).triu(1)
if attention_mask is not None: scores += (1.0 - attention_mask.unsqueeze(1).unsqueeze(2)) * -1e9
output = self.attn_dropout(F.softmax(scores.float(), dim=-1).type_as(xq)) @ xv
output = output.transpose(1, 2).reshape(bsz, seq_len, -1)