🧠 两小时从 0 训出你的 64M 大模型。Train your own 64M LLM from scratch in 2 hours.
[update] prompt prefill
jingyaogong1279a6168171
diff --git a/model/model_minimind.py b/model/model_minimind.pyindex 830c78d..d826f82 100755--- a/model/model_minimind.py+++ b/model/model_minimind.py@@ -179,7 +179,7 @@ class Attention(nn.Module):xv = xv.view(bsz, seq_len, self.n_local_kv_heads, self.head_dim)cos, sin = position_embeddings- xq, xk = apply_rotary_pos_emb(xq, xk, cos[:seq_len], sin[:seq_len])+ xq, xk = apply_rotary_pos_emb(xq, xk, cos, sin)# kv_cache实现if past_key_value is not None:@@ -193,14 +193,11 @@ class Attention(nn.Module):repeat_kv(xv, self.n_rep).transpose(1, 2))- if self.flash and seq_len > 1 and (attention_mask is None or torch.all(attention_mask == 1)):+ 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)else:scores = (xq @ xk.transpose(-2, -1)) / math.sqrt(self.head_dim)- scores = scores + torch.triu(- torch.full((seq_len, seq_len), float("-inf"), device=scores.device),- diagonal=1- ).unsqueeze(0).unsqueeze(0) # scores+mask+ scores[:, :, :, -seq_len:] += torch.triu(torch.full((seq_len, seq_len), float("-inf"), device=scores.device), diagonal=1)if attention_mask is not None:extended_attention_mask = attention_mask.unsqueeze(1).unsqueeze(2)