Views
No views yet

输入 x
│
├─> w1(x) ──> silu ─┐
│ × ──> w2 ──> dropout ──> 输出
└─> w3(x) ─────────┘模型架构超参数:
dim: int = 512,
n_layers: int = 8,
n_heads: int = 8,
n_kv_heads: int = 2,
vocab_size: int = 6600,
multiple_of: int = 64,
norm_eps: float = 1e-5,
max_seq_len: int = 8192,
rope_theta: int = 1e6,
dropout: float = 0.0,
flash_attn: bool = True,LLM 总参数量: 26194432
LoRA 参数量: 262144
LoRA 参数占比: 1.00%# 在 sp_ids 对应的位置增加额外的惩罚
...
loss_mask[sp_ids] = 10 # 惩罚系数,原本是1