import math, torch, torch.nn.functional as F from torch import nn from transformers.activations import ACT2FN from transformers import PreTrainedModel, GenerationMixin, PretrainedConfig from transformers.modeling_outputs import MoeCausalLMOutputWithPast # 🌏🌎🌍🌏🌎🌍🌏🌎🌍🌏🌎🌍🌏🌎🌍🌏🌎🌍🌏🌎🌍🌏🌎🌍🌏🌎🌍🌏🌎🌍🌏🌎🌍🌏🌎🌍🌏🌎🌍🌏🌎🌍🌏🌎🌍🌏🌎🌍🌏🌎🌍🌏 # MiniMind Config # 🌏🌎🌍🌏🌎🌍🌏🌎🌍🌏🌎🌍🌏🌎🌍🌏🌎🌍🌏🌎🌍🌏🌎🌍🌏🌎🌍🌏🌎🌍🌏🌎🌍🌏🌎🌍🌏🌎🌍🌏🌎🌍🌏🌎🌍🌏🌎🌍🌏🌎🌍🌏 class MiniMindConfig(PretrainedConfig): model_type = "minimind" def __init__(self, hidden_size=768, num_hidden_layers=8, use_moe=False, **kwargs): super().__init__(**kwargs) self.hidden_size = hidden_size self.num_hidden_layers = num_hidden_layers self.use_moe = use_moe self.dropout = kwargs.get("dropout", 0.0) self.vocab_size = kwargs.get("vocab_size", 6400) self.bos_token_id = kwargs.get("bos_token_id", 1) self.eos_token_id = kwargs.get("eos_token_id", 2) self.flash_attn = kwargs.get("flash_attn", True) self.num_attention_heads = kwargs.get("num_attention_heads", 8) self.attention_type = kwargs.get("attention_type", "gqa") num_key_value_heads = kwargs.get("num_key_value_heads", 4) if self.attention_type == "mha": num_key_value_heads = self.num_attention_heads elif self.attention_type == "mqa": num_key_value_heads = 1 self.num_key_value_heads = num_key_value_heads if self.num_attention_heads % self.num_key_value_heads != 0: raise ValueError("num_attention_heads must be divisible by num_key_value_heads") self.head_dim = kwargs.get("head_dim", self.hidden_size // self.num_attention_heads) self.hidden_act = kwargs.get("hidden_act", 'silu') self.intermediate_size = kwargs.get("intermediate_size", math.ceil(hidden_size * math.pi / 64) * 64) self.max_position_embeddings = kwargs.get("max_position_embeddings", 32768) self.rms_norm_eps = kwargs.get("rms_norm_eps", 1e-6) self.rope_theta = kwargs.get("rope_theta", 1e6) self.tie_word_embeddings = kwargs.get("tie_word_embeddings", True) self.inference_rope_scaling = kwargs.get("inference_rope_scaling", False) self.rope_scaling = { "beta_fast": 32, "beta_slow": 1, "factor": 16, "original_max_position_embeddings": 2048, "attention_factor": 1.0, "type": "yarn" } if self.inference_rope_scaling else None ### MoE specific configs (ignored if use_moe = False) self.num_experts = kwargs.get("num_experts", 4) self.num_experts_per_tok = kwargs.get("num_experts_per_tok", 1) self.moe_intermediate_size = kwargs.get("moe_intermediate_size", self.intermediate_size) self.norm_topk_prob = kwargs.get("norm_topk_prob", True) self.router_aux_loss_coef = kwargs.get("router_aux_loss_coef", 5e-4) # 🌏🌎🌍🌏🌎🌍🌏🌎🌍🌏🌎🌍🌏🌎🌍🌏🌎🌍🌏🌎🌍🌏🌎🌍🌏🌎🌍🌏🌎🌍🌏🌎🌍🌏🌎🌍🌏🌎🌍🌏🌎🌍🌏🌎🌍🌏🌎🌍🌏🌎🌍🌏 # MiniMind Model # 🌏🌎🌍🌏🌎🌍🌏🌎🌍🌏🌎🌍🌏🌎🌍🌏🌎🌍🌏🌎🌍🌏🌎🌍🌏🌎🌍🌏🌎🌍🌏🌎🌍🌏🌎🌍🌏🌎🌍🌏🌎🌍🌏🌎🌍🌏🌎🌍🌏🌎🌍🌏 class RMSNorm(torch.nn.Module): def __init__(self, dim: int, eps: float = 1e-5): super().__init__() self.eps = eps self.weight = nn.Parameter(torch.ones(dim)) def norm(self, x): #rsqrt是sqrt的倒数，keepdim是保持维度不变，这里是把求平均的维度位置为1 return x * torch.rsqrt(x.pow(2).mean(-1, keepdim=True) + self.eps) def forward(self, x): #self.weight是一个可学习的参数 return (self.weight * self.norm(x.float())).type_as(x) def precompute_freqs_cis(dim: int, end: int = int(32 * 1024), rope_base: float = 1e6, rope_scaling: dict = None): freqs, attn_factor = 1.0 / (rope_base ** (torch.arange(0, dim, 2)[: (dim // 2)].float() / dim)), 1.0 """ 把原本只能支持较短上下文的RoPE，平滑地扩展到更长上下文 普通rope的频率是f，位置m的旋转角度是mf，如果上下文长度很长，m很大，角度会转的太快，模型泛化不好 YaRN的做法是对一部分频率进行缩放，让他们转的慢一点 高频部分频率保持原样，保留短上下文能力， 中频部分频率平滑过渡 低频部分缩小，用于支持长上下文 """ if rope_scaling is not None: # YaRN: f'(i) = f(i)((1-γ) + γ/s), where γ∈[0,1] is linear ramp，y=1是表示低频部分缩小，y=0表示高频部分不变 """ orig_max是原模型训练时支持的最大位置长度 factor是上下文扩展倍数 beta_fast和beat_slow用来决定哪些维度的频率保持不变 attn_factir是乘在cos/sin上的attention的缩放因子 """ orig_max, factor, beta_fast, beta_slow, attn_factor = ( rope_scaling.get("original_max_position_embeddings", 2048), rope_scaling.get("factor", 16), rope_scaling.get("beta_fast", 32.0), rope_scaling.get("beta_slow", 1.0), rope_scaling.get("attention_factor", 1.0) ) #只有当你实际预计算的长度end超过原始最大长度时，才需要做rope scaling if end / orig_max > 1.0: #inv_dim推理公式，用beta，也就是这里的b控制周期和原始上下文长度之间的关系 inv_dim = lambda b: (dim * math.log(orig_max / (b * 2 * math.pi))) / (2 * math.log(rope_base)) #low，high是两个分界点 #0到low的区间，ramp=0，频率不变，low到high的区间，ramp从0平滑加到1，high到dim//2的区间，ramp=1，频率除以factor low, high = max(math.floor(inv_dim(beta_fast)), 0), min(math.ceil(inv_dim(beta_slow)), dim // 2 - 1) #torch.clamp是裁剪或限制张量中元素的数值范围，torch.clamp(input,min,max),不满足范围内的用最大最小值代替 ramp = torch.clamp((torch.arange(dim // 2, device=freqs.device).float() - low) / max(high - low, 0.001), 0, 1) freqs = freqs * (1 - ramp + ramp / factor) t = torch.arange(end, device=freqs.device) #[end,dim//2] freqs = torch.outer(t, freqs).float() #[end,dim] freqs_cos = torch.cat([torch.cos(freqs), torch.cos(freqs)], dim=-1) * attn_factor freqs_sin = torch.cat([torch.sin(freqs), torch.sin(freqs)], dim=-1) * attn_factor return freqs_cos, freqs_sin def apply_rotary_pos_emb(q, k, cos, sin, unsqueeze_dim=1): # rope作用在qk上，传统的正弦位置编码一般加在token embedding上，不用作用在v上，因为注意力分数计算只用qk #常见的qk shape是[baych,num_heads,seq_len,head_dim]要对头和seq_len做一个维度交换以适应硬件要求 #...表示省略前面所有的维度，前面的维度都按原样保留 def rotate_half(x): return torch.cat((-x[..., x.shape[-1] // 2:], x[..., : x.shape[-1] // 2]), dim=-1) q_embed = ((q * cos.unsqueeze(unsqueeze_dim)) + (rotate_half(q) * sin.unsqueeze(unsqueeze_dim))).to(q.dtype) k_embed = ((k * cos.unsqueeze(unsqueeze_dim)) + (rotate_half(k) * sin.unsqueeze(unsqueeze_dim))).to(k.dtype) return q_embed, k_embed def repeat_kv(x: torch.Tensor, n_rep: int) -> torch.Tensor: """ 用于GQA和MQA attention 把k/v heads复制成和query_heads一样多，方便后面做attention计算 """ bs, slen, num_key_value_heads, head_dim = x.shape if n_rep == 1: return x #None是在这里插入一个新维度，expand通常不真的复制维度，只是创建一个广播视图，所以更省内存 return (x[:, :, :, None, :].expand(bs, slen, num_key_value_heads, n_rep, head_dim).reshape(bs, slen, num_key_value_heads * n_rep, head_dim)) class Attention(nn.Module): def __init__(self, config: MiniMindConfig): super().__init__() #qk用几个头 self.num_key_value_heads = config.num_attention_heads if config.num_key_value_heads is None else config.num_key_value_heads self.n_local_heads = config.num_attention_heads 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) self.o_proj = nn.Linear(config.num_attention_heads * self.head_dim, config.hidden_size, bias=False) self.q_norm = RMSNorm(self.head_dim, eps=config.rms_norm_eps) self.k_norm = RMSNorm(self.head_dim, eps=config.rms_norm_eps) self.attn_dropout = nn.Dropout(config.dropout) self.resid_dropout = nn.Dropout(config.dropout) self.dropout = config.dropout #self.flash是布尔值 self.flash = hasattr(torch.nn.functional, 'scaled_dot_product_attention') and config.flash_attn def forward(self, x, position_embeddings, past_key_value=None, use_cache=False, attention_mask=None): bsz, seq_len, _ = x.shape xq, xk, xv = self.q_proj(x), self.k_proj(x), self.v_proj(x) xq = xq.view(bsz, seq_len, self.n_local_heads, self.head_dim) xk = xk.view(bsz, seq_len, self.n_local_kv_heads, self.head_dim) xv = xv.view(bsz, seq_len, self.n_local_kv_heads, self.head_dim) xq, xk = self.q_norm(xq), self.k_norm(xk) cos, sin = position_embeddings xq, xk = apply_rotary_pos_emb(xq, xk, cos, sin) if past_key_value is not None: xk = torch.cat([past_key_value[0], xk], dim=1) xv = torch.cat([past_key_value[1], xv], dim=1) past_kv = (xk, xv) if use_cache else None #这段是在判断，当前情况能不能走Pytorch中内置的高效attention实现，scaled_dot_product_attention 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 (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: #@是矩阵乘法，*是逐元素相乘，两个张量形状必须相同 scores = (xq @ xk.transpose(-2, -1)) / math.sqrt(self.head_dim) #torch.full是生成一个形状指定，且所有元素都填充为同一个固定值的张量 #triu是pytorch中用于提取矩阵上三角的函数，diagonal参数=0时，保留主对角线及以上的元素，>0时保留主对角线上方的第N条对角线及以上的元素，<0时保留主对角线下方及以下的元素,其余用0替代 #triu只对张量的后两个维度进行操作 if self.is_causal: scores[:, :, :, -seq_len:] += torch.full((seq_len, seq_len), float("-inf"), device=scores.device).triu(1) #casual mask和attention_mask管的是两件事 #casual_mask是阻止当前位置看到未来token，attention_mask是告诉模型哪些是有效的，哪些token不应该看见 #attention_mask是在padding时忽略padding部分的注意力分数 if attention_mask is not None: scores += (1.0 - attention_mask.unsqueeze(1).unsqueeze(2)) * -1e9 #对注意力分数做一个正则化 output = self.attn_dropout(F.softmax(scores.float(), dim=-1).type_as(xq)) @ xv output = output.transpose(1, 2).reshape(bsz, seq_len, -1) #还要对输出结果做一个正则化 output = self.resid_dropout(self.o_proj(output)) return output, past_kv class FeedForward(nn.Module): def __init__(self, config: MiniMindConfig, intermediate_size: int = None): super().__init__() intermediate_size = intermediate_size or config.intermediate_size self.gate_proj = nn.Linear(config.hidden_size, intermediate_size, bias=False) self.down_proj = nn.Linear(intermediate_size, config.hidden_size, bias=False) self.up_proj = nn.Linear(config.hidden_size, intermediate_size, bias=False) self.act_fn = ACT2FN[config.hidden_act] def forward(self, x): return self.down_proj(self.act_fn(self.gate_proj(x)) * self.up_proj(x)) class MOEFeedForward(nn.Module): def __init__(self, config: MiniMindConfig): super().__init__() self.config = config self.gate = nn.Linear(config.hidden_size, config.num_experts, bias=False) self.experts = nn.ModuleList([FeedForward(config, intermediate_size=config.moe_intermediate_size) for _ in range(config.num_experts)]) self.act_fn = ACT2FN[config.hidden_act] def forward(self, x): batch_size, seq_len, hidden_dim = x.shape #[bs*seq_len,hidden_dim] x_flat = x.view(-1, hidden_dim) #[bs*seq_len,num_experts] scores = F.softmax(self.gate(x_flat), dim=-1) #[bs*seq_len,num_experts_per_tok] topk_weight, topk_idx = torch.topk(scores, k=self.config.num_experts_per_tok, dim=-1, sorted=False) if self.config.norm_topk_prob: topk_weight = topk_weight / (topk_weight.sum(dim=-1, keepdim=True) + 1e-20) #[bs*seq_len,hidden_dim] y = torch.zeros_like(x_flat) for i, expert in enumerate(self.experts): #[bs*seq_len,num_experts_per_tok] mask = (topk_idx == i) #any()用来判断是否至少有一个元素为真 if mask.any(): #nonzero用来返回非零元素的索引，flatten将结果展成一维向量 #[num_selected_tokens] token_idx = mask.any(dim=-1).nonzero().flatten() #[num_selected_tokens,1]，这里mask是一个和top_weight同维度的布尔矩阵，会把top_weight中和mask矩阵位置相同且为True的部分提取出来 weight = topk_weight[mask].view(-1, 1) #x_flat[token_idx]->[num_selected_tokens,hidden_dim] #y[token_idx] += expert_output，index_add_(dim,index,source) y.index_add_(0, token_idx, (expert(x_flat[token_idx]) * weight).to(y.dtype)) elif self.training: #某个专家没参与计算，也把它挂到计算图上 y[0, 0] += 0 * sum(p.sum() for p in expert.parameters()) if self.training and self.config.router_aux_loss_coef > 0: #F.one_hot(topk_idx, self.config.num_experts)->[bs*seq_len,num_experts_per_tok,num_experts]，会把原来的每一个数字变成一组one-hot向量 #做mean(0)后，会在第一个维度做平均，[num_experts_per_tok,num_experts] load = F.one_hot(topk_idx, self.config.num_experts).float().mean(0) self.aux_loss = (load * scores.mean(0)).sum() * self.config.num_experts * self.config.router_aux_loss_coef else: self.aux_loss = scores.new_zeros(1).squeeze() return y.view(batch_size, seq_len, hidden_dim) class MiniMindBlock(nn.Module): def __init__(self, layer_id: int, config: MiniMindConfig): super().__init__() self.self_attn = Attention(config) self.input_layernorm = RMSNorm(config.hidden_size, eps=config.rms_norm_eps) self.post_attention_layernorm = RMSNorm(config.hidden_size, eps=config.rms_norm_eps) self.mlp = FeedForward(config) if not config.use_moe else MOEFeedForward(config) def forward(self, hidden_states, position_embeddings, past_key_value=None, use_cache=False, attention_mask=None): residual = hidden_states hidden_states, present_key_value = self.self_attn( self.input_layernorm(hidden_states), position_embeddings, past_key_value, use_cache, attention_mask ) hidden_states += residual hidden_states = hidden_states + self.mlp(self.post_attention_layernorm(hidden_states)) return hidden_states, present_key_value class MiniMindModel(nn.Module): def __init__(self, config: MiniMindConfig): super().__init__() self.config = config self.vocab_size, self.num_hidden_layers = config.vocab_size, config.num_hidden_layers self.embed_tokens = nn.Embedding(config.vocab_size, config.hidden_size) self.dropout = nn.Dropout(config.dropout) self.layers = nn.ModuleList([MiniMindBlock(l, config) for l in range(self.num_hidden_layers)]) self.norm = RMSNorm(config.hidden_size, eps=config.rms_norm_eps) #[max_position_embeddings,head_dim] freqs_cos, freqs_sin = precompute_freqs_cis(dim=config.head_dim, end=config.max_position_embeddings, rope_base=config.rope_theta, rope_scaling=config.rope_scaling) self.register_buffer("freqs_cos", freqs_cos, persistent=False) self.register_buffer("freqs_sin", freqs_sin, persistent=False) def forward(self, input_ids, attention_mask=None, past_key_values=None, use_cache=False, **kwargs): batch_size, seq_length = input_ids.shape if hasattr(past_key_values, 'layers'): past_key_values = None past_key_values = past_key_values or [None] * len(self.layers) start_pos = past_key_values[0][0].shape[1] if past_key_values[0] is not None else 0 hidden_states = self.dropout(self.embed_tokens(input_ids)) # Recompute RoPE buffers lost during meta-device init (transformers>=5.x) if self.freqs_cos[0, 0] == 0: freqs_cos, freqs_sin = precompute_freqs_cis(dim=self.config.head_dim, end=self.config.max_position_embeddings, rope_base=self.config.rope_theta, rope_scaling=self.config.rope_scaling) self.freqs_cos, self.freqs_sin = freqs_cos.to(hidden_states.device), freqs_sin.to(hidden_states.device) position_embeddings = (self.freqs_cos[start_pos:start_pos + seq_length], self.freqs_sin[start_pos:start_pos + seq_length]) presents = [] for layer, past_key_value in zip(self.layers, past_key_values): hidden_states, present = layer( hidden_states, position_embeddings, past_key_value=past_key_value, use_cache=use_cache, attention_mask=attention_mask ) presents.append(present) hidden_states = self.norm(hidden_states) #sum方法中第二个参数为初始值，new_zeros创建与当前张量同类型，同设备的新零张量。里面的参数是形状，设置一个初值0 aux_loss = sum([l.mlp.aux_loss for l in self.layers if isinstance(l.mlp, MOEFeedForward)], hidden_states.new_zeros(1).squeeze()) return hidden_states, presents, aux_loss class MiniMindForCausalLM(PreTrainedModel, GenerationMixin): config_class = MiniMindConfig _tied_weights_keys = {"lm_head.weight": "model.embed_tokens.weight"} def __init__(self, config: MiniMindConfig = None): self.config = config or MiniMindConfig() super().__init__(self.config) self.model = MiniMindModel(self.config) self.lm_head = nn.Linear(self.config.hidden_size, self.config.vocab_size, bias=False) #nn.Embedding的权重矩阵存储是(input_dim,output_dim),nn.Linear的权重矩阵存储是转秩过后的 if self.config.tie_word_embeddings: self.model.embed_tokens.weight = self.lm_head.weight #在模型所有子模块创建完成后，执行Huggingface规定的后处理初始化逻辑 self.post_init() def forward(self, input_ids, attention_mask=None, past_key_values=None, use_cache=False, logits_to_keep=0, labels=None, **kwargs): hidden_states, past_key_values, aux_loss = self.model(input_ids, attention_mask, past_key_values, use_cache, **kwargs) #创建一个切片对象 x[3:5]和x[slice(3,5)]等价，这里等价于[-logits_to_keep:] slice_indices = slice(-logits_to_keep, None) if isinstance(logits_to_keep, int) else logits_to_keep logits = self.lm_head(hidden_states[:, slice_indices, :]) loss = None if labels is not None: #训练时输入和输出的长度相同，由 12345预测 23456 所以要计算2345的loss x, y = logits[..., :-1, :].contiguous(), labels[..., 1:].contiguous() loss = F.cross_entropy(x.view(-1, x.size(-1)), y.view(-1), ignore_index=-100) #把loss，aux_loss，logits等封装成一个标准输出对象，方便训练，推理,以后可以用字典的方式访问 return MoeCausalLMOutputWithPast(loss=loss, aux_loss=aux_loss, logits=logits, past_key_values=past_key_values, hidden_states=hidden_states) # https://github.com/jingyaogong/minimind/discussions/611 #表示这个函数在推理模式下运行，不计算梯度 @torch.inference_mode() def generate(self, inputs=None, attention_mask=None, max_new_tokens=8192, temperature=0.85, top_p=0.85, top_k=50, eos_token_id=2, streamer=None, use_cache=True, num_return_sequences=1, do_sample=True, repetition_penalty=1.0, **kwargs): #这里的repeat是向量复制，这里是在第0维复制num_return_sequences次，第1维复制1次，也就是不变，repeat（a,b,c)是第一维复制a次，第二维复制b次，第三维复制c次，以此类推 input_ids = kwargs.pop("input_ids", inputs).repeat(num_return_sequences, 1) attention_mask = attention_mask.repeat(num_return_sequences, 1) if attention_mask is not None else None past_key_values = kwargs.pop("past_key_values", None) finished = torch.zeros(input_ids.shape[0], dtype=torch.bool, device=input_ids.device) if streamer: streamer.put(input_ids.cpu()) for _ in range(max_new_tokens): past_len = past_key_values[0][0].shape[1] if past_key_values else 0 outputs = self.forward(input_ids[:, past_len:], attention_mask, past_key_values, use_cache=use_cache, **kwargs) attention_mask = torch.cat([attention_mask, attention_mask.new_ones(attention_mask.shape[0], 1)], -1) if attention_mask is not None else None logits = outputs.logits[:, -1, :] / temperature #重复惩罚，如果某个token已经在当前生成序列里出现过，就降低它下一次再次被选中的概率 if repetition_penalty != 1.0: #torch.unique是去重，给张量 for i in range(input_ids.shape[0]): seen = torch.unique(input_ids[i]); score = logits[i, seen]; logits[i, seen] = torch.where(score > 0, score / repetition_penalty, score * repetition_penalty) if top_k > 0: logits[logits < torch.topk(logits, top_k)[0][..., -1, None]] = -float('inf') #按概率从高到低，只保留累计概率达到top_P之前的token if top_p < 1.0: sorted_logits, sorted_indices = torch.sort(logits, descending=True) mask = torch.cumsum(torch.softmax(sorted_logits, dim=-1), dim=-1) > top_p mask[..., 1:], mask[..., 0] = mask[..., :-1].clone(), 0 logits[mask.scatter(1, sorted_indices, mask)] = -float('inf') next_token = torch.multinomial(torch.softmax(logits, dim=-1),num_samples=1) if do_sample else torch.argmax(logits, dim=-1, keepdim=True) #torch.where是pytorch的条件选择函数，类似三元运算符torch.where(condition,x,y) if eos_token_id is not None: next_token = torch.where(finished.unsqueeze(-1), next_token.new_full((next_token.shape[0], 1), eos_token_id), next_token) input_ids = torch.cat([input_ids, next_token], dim=-1) past_key_values = outputs.past_key_values if use_cache else None if streamer: streamer.put(next_token.cpu()) if eos_token_id is not None: finished |= next_token.squeeze(-1).eq(eos_token_id) if finished.all(): break if streamer: streamer.end() if kwargs.get("return_kv"): return {'generated_ids': input_ids, 'past_kv': past_key_values} return input_ids