feat(model): 添加QK归一化和门控注意力支持

khaosz/config/model_config.py

@@ -1,37 +1,43 @@
 import json
 
 from dataclasses import asdict, dataclass
-from typing import Any, Dict, Optional, Self
+from typing import Optional, Self
+
 
 @dataclass
 class ModelConfig:
     # basic config
     vocab_size: Optional[int] = None
-    n_dim: Optional[int] = None
+    dim: Optional[int] = None
-    n_head: Optional[int] = None
+
-    n_layer: Optional[int] = None
+    n_layers: Optional[int] = None
-    m_len: Optional[int] = None
     norm_eps: Optional[float] = None
-    d_ffn: Optional[int] = None
+    dim_ffn: Optional[int] = None
     tie_weight: Optional[bool] = None
     
+    # RoPE
+    max_len: Optional[int] = None
+    rope_theta: Optional[float] = None
+    
     # GQA
-    n_kvhead: Optional[int] = None
+    n_heads: Optional[int] = None
+    n_kv_heads: Optional[int] = None
+    use_qk_norm: Optional[bool] = None
+    use_gated_attention: Optional[bool] = None
     
     
     def load(self, config_path: str) -> Self:
+        config = {}
         with open(config_path, 'r') as f:
-            config: Dict[str, Any] = json.load(f)
+            config.update(json.load(f)) 
+            
         for key, value in config.items():
             if hasattr(self, key):
                 setattr(self, key, value)
     
         return self
                     
-    def save(self, config_path: str) -> None:
+    def save(self, config_path: str):
-        config_dict = asdict(self)
+        config_dict = {k: v for k, v in asdict(self).items() if v is not None}
-        config_dict = {k: v for k, v in config_dict.items() if v is not None}
         with open(config_path, 'w') as f:
             json.dump(config_dict, f, indent=4)
-
-

khaosz/inference/core.py

@@ -100,7 +100,7 @@ class GeneratorCore:
     ) -> List[int]:
         cur_cache_pos = start_pos
         
-        for _ in range(len(ids), self.config.m_len):
+        for _ in range(len(ids), self.config.max_len):
             next_token_id, cache_increase = self.generate_iterator(
                 input_ids, temperature, top_k, top_p, attn_mask, kv_caches, cur_cache_pos)
             
@@ -127,7 +127,7 @@ class EmbeddingEncoderCore:
         with_batch = isinstance(sentence, list)
         ids = self.tokenizer.encode(sentence)
         batch_ids = ids if with_batch else [ids]
-        max_model_len = self.config.m_len
+        max_model_len = self.config.max_len
         
         all_fragments = []
         fragment_origin_idx = []
@@ -195,10 +195,10 @@ class KVCacheManager:
         self.batch_size = batch_size
         self.device = device
         self.dtype = dtype
-        self.num_layers = config.n_layer
+        self.num_layers = config.n_layers
-        self.max_len = config.m_len
+        self.max_len = config.max_len
-        self.num_heads = config.n_kvhead
+        self.num_heads = config.n_kv_heads
-        self.head_dim = config.n_dim //config.n_head
+        self.head_dim = config.dim //config.n_heads
         
         self._kv_cache: Tuple[Tensor, Tensor] = None
         self._seq_mask: Tensor = None

khaosz/inference/generator.py

@@ -167,7 +167,7 @@ class StreamGenerator(GeneratorCore):
         self.model.eval()
         kv_caches = cache_manager.get_kvcache()
         
-        for _ in range(len(ids), self.config.m_len):
+        for _ in range(len(ids), self.config.max_len):
             next_token_id, cache_increase = self.generate_iterator(
                 input_ids, temperature, top_k, top_p, kv_caches=kv_caches, start_pos=cur_cache_pos)
             
@@ -219,7 +219,7 @@ class BatchGenerator(GeneratorCore):
         start_cache_pos = max_ids_len
         cur_cache_pos = 0
         
-        while max_ids_len < self.config.m_len and sum(activate_task_mask) != 0:
+        while max_ids_len < self.config.max_len and sum(activate_task_mask) != 0:
             kv_caches = cache_manager.get_kvcache()
             attn_mask =cache_manager.get_seq_mask()
             

khaosz/model/module.py

@@ -102,23 +102,20 @@ class RotaryEmbedding(nn.Module):
 
 
 class Linear(nn.Module):
-    def __init__(self, in_dim: int, out_dim: int, bias: bool = False, weight_param=None, bias_param=None):
+    def __init__(self, in_dim: int, out_dim: int, bias: bool = False):
         super().__init__()
-        weight_param = torch.empty((out_dim, in_dim)) if weight_param is None else weight_param
+        self.weight = nn.Parameter(torch.empty((out_dim, in_dim)))
-        bias_param = torch.zeros(out_dim) if bias_param is None else bias_param
+        self.bias = nn.Parameter(torch.zeros(out_dim)) if bias else None
-        
-        self.weight = nn.Parameter(weight_param)
-        self.bias = nn.Parameter(bias_param) if bias else None
 
     def forward(self, x: Tensor) -> Tensor:
         return F.linear(x, self.weight, self.bias)
 
 
 class RMSNorm(nn.Module):
-    def __init__(self, n_dim, norm_eps):
+    def __init__(self, dim, norm_eps):
         super().__init__()
-        self.weight = nn.Parameter(torch.ones(n_dim))
+        self.weight = nn.Parameter(torch.ones(dim))
-        self.normalized_shape = (n_dim, )
+        self.normalized_shape = (dim, )
         self.norm_eps = norm_eps
 
     def forward(self, x: Tensor) -> Tensor:
@@ -127,41 +124,70 @@ class RMSNorm(nn.Module):
     
     
 class MLP(nn.Module):
-    def __init__(self, n_dim: int, d_ffn: int):
+    def __init__(self, dim: int, dim_feed_forward: int):
         super().__init__()
-        self.up = Linear(n_dim, d_ffn)
+        self.up = Linear(dim, dim_feed_forward)
-        self.gate = Linear(n_dim, d_ffn)
+        self.gate = Linear(dim, dim_feed_forward)
-        self.down = Linear(d_ffn, n_dim)
+        self.down = Linear(dim_feed_forward, dim)
     
     def forward(self, x: Tensor) -> Tensor:
         gated = self.up(x) * F.silu(self.gate(x))
         out = self.down(gated)
         return out
 
+class Attention(nn.Module):
+    
+    def forward(self, q: Tensor, k: Tensor, v: Tensor, mask: Optional[Tensor] = None, is_causal: bool= False):
+        # (bsz, seq_len, n_heads, head_dim) -> (bsz, n_heads, seq_len, head_dim)
+        q, k, v = q.permute(0, 2, 1, 3), k.permute(0, 2, 1, 3), v.permute(0, 2, 1, 3)
+        # (bsz, n_heads, seq_len, head_dim) - > (bsz, seq_len, n_heads*head_dim)
+        sdqa_out = F.scaled_dot_product_attention(q, k, v, mask, is_causal=is_causal).permute(0, 2, 1, 3).contiguous().flatten(2)
+        
+        return sdqa_out
+
 
 class GQA(nn.Module):
     def __init__(
         self, 
-        n_dim: int, 
+        dim: int, 
-        n_head: int, 
+        n_heads: int, 
-        n_kvhead: int, 
+        n_kv_heads: int, 
+        use_qk_norm: bool,
+        norm_eps: float,
+        use_gated_attention: bool,
         layer_id: int
     ):
         super().__init__()
-        assert n_dim % n_head == 0
+        assert dim % n_heads == 0
-        assert n_head % n_kvhead == 0
+        assert n_heads % n_kv_heads == 0
         
-        self.head_dim = n_dim // n_head
+        self.head_dim = dim // n_heads
         self.layer_id = layer_id
-        self.n_dim = n_dim
+        self.dim = dim
-        self.n_heads = n_head
+        self.n_heads = n_heads
-        self.n_kvheads = n_kvhead
+        self.n_kv_heads = n_kv_heads
-        self.n_rep = n_head // n_kvhead
+        self.n_rep = n_heads // n_kv_heads
+        self.use_qk_norm = use_qk_norm
+        self.use_gated_attention = use_gated_attention
         
-        self.q_proj = Linear(n_dim, n_head * self.head_dim)
+        self.attention = Attention()
-        self.k_proj = Linear(n_dim, n_kvhead * self.head_dim)
+
-        self.v_proj = Linear(n_dim, n_kvhead * self.head_dim)
+        self.q_proj = Linear(dim, n_heads * self.head_dim)
-        self.o_proj = Linear(n_dim, n_dim)
+        self.k_proj = Linear(dim, n_kv_heads * self.head_dim)
+        self.v_proj = Linear(dim, n_kv_heads * self.head_dim)
+        self.o_proj = Linear(dim, dim)
+        
+        if self.use_qk_norm:
+            self.q_norm = RMSNorm(self.head_dim, norm_eps)
+            self.k_norm = RMSNorm(self.head_dim, norm_eps)
+        
+        if self.use_gated_attention:
+            self.gate = Linear(dim, dim)
+
+    def _split_heads(self, x: Tensor, n_heads) -> Tensor:
+        batch_size, seq_len, _ = x.shape
+        x = x.reshape(batch_size, seq_len, n_heads, self.head_dim)
+        return x
     
     def forward(
         self,
@@ -174,10 +200,13 @@ class GQA(nn.Module):
         bsz, seq_len, _ = x.size()
         # x(bsz, seq_len, n_heads * head_dim) -> (bsz, seq_len, n_heads, head_dim)
         q = self._split_heads(self.q_proj(x), self.n_heads)
-        k = self._split_heads(self.k_proj(x), self.n_kvheads)
+        k = self._split_heads(self.k_proj(x), self.n_kv_heads)
-        v = self._split_heads(self.v_proj(x), self.n_kvheads)
+        v = self._split_heads(self.v_proj(x), self.n_kv_heads)
         q, k = apply_rotary_emb(q, rotary_emb), apply_rotary_emb(k, rotary_emb)
         
+        if self.use_qk_norm:
+            q, k = self.q_norm(q), self.k_norm(k)
+        
         if kv_cache is not None:
             k_cache, v_cache = kv_cache
             
@@ -190,27 +219,34 @@ class GQA(nn.Module):
             v = v_cache[:bsz, :start_pos + seq_len, self.layer_id]
         
         k, v = repeat_kv(k, self.n_rep), repeat_kv(v, self.n_rep)
+        sdqa_out = self.attention(q, k, v, mask, is_causal=(mask == None))
         
-        # (bsz, seq_len, n_heads, head_dim) -> (bsz, n_heads, seq_len, head_dim)
+        if self.use_gated_attention:
-        q, k, v = q.permute(0, 2, 1, 3), k.permute(0, 2, 1, 3), v.permute(0, 2, 1, 3)
+            sdqa_out = sdqa_out * F.sigmoid(self.gate(x))
-        sdqa_out = F.scaled_dot_product_attention(q, k, v, mask, is_causal=(mask == None)).permute(0, 2, 1, 3)
+        
-        out = self.o_proj(sdqa_out.contiguous().view(bsz, seq_len, -1))
+        out = self.o_proj(sdqa_out)
 
         return out
-    
+
-    def _split_heads(self, x: Tensor, n_heads) -> Tensor:
-        batch_size, seq_len, _ = x.shape
-        x = x.reshape(batch_size, seq_len, n_heads, self.head_dim)
-        return x
-    
 
 class DecoderBlock(nn.Module):
-    def __init__(self, n_dim, n_head, d_ffn, n_kvhead, norm_eps, layer_id):
+    def __init__(
+        self, 
+        dim: int, 
+        n_heads: int, 
+        dim_ffn: int, 
+        n_kv_heads: int, 
+        norm_eps: int, 
+        use_qk_norm: bool, 
+        use_gated_attention: bool, 
+        layer_id: int
+    ):
         super().__init__()
-        self.attention = GQA(n_dim, n_head, n_kvhead, layer_id)
+        self.attention = GQA(dim, n_heads, n_kv_heads, 
-        self.norm_attn = RMSNorm(n_dim, norm_eps)
+                             use_qk_norm, norm_eps, use_gated_attention, layer_id)
-        self.ffn = MLP(n_dim, d_ffn)
+        self.input_norm = RMSNorm(dim, norm_eps)
-        self.norm_ffn = RMSNorm(n_dim, norm_eps)
+        self.mlp = MLP(dim, dim_ffn)
+        self.post_attention_norm = RMSNorm(dim, norm_eps)
 
     def forward(
         self,
@@ -222,7 +258,7 @@ class DecoderBlock(nn.Module):
     ) -> Tensor:
         # attention
         attn_output = self.attention(
-            self.norm_attn(x), 
+            self.input_norm(x), 
             rotary_emb, 
             attention_mask, 
             kv_cache, 
@@ -231,16 +267,15 @@ class DecoderBlock(nn.Module):
         x = attn_output + x
         
         # feed forward
-        x = self.ffn(self.norm_ffn(x)) + x
+        x = self.mlp(self.post_attention_norm(x)) + x
         
         return x
 
 
 class Embedding(nn.Module):
-    def __init__(self, vocab_size: int, embedding_dim: int, weight_param=None):
+    def __init__(self, vocab_size: int, embedding_dim: int):
         super().__init__()
-        weight_param = torch.empty((vocab_size, embedding_dim)) if weight_param is None else weight_param
+        self.weight = nn.Parameter(torch.empty((vocab_size, embedding_dim)))
-        self.weight = nn.Parameter(weight_param)
     
     def forward(self, x: Tensor) -> Tensor:
         return F.embedding(x, self.weight)

khaosz/model/transformer.py

@@ -59,16 +59,17 @@ class Transformer(nn.Module):
     def __init__(self, config: ModelConfig):
         super().__init__()
         self.config = config
-        self.rotary_embeding = RotaryEmbedding(config.n_dim // config.n_head, config.m_len)
+        self.rotary_embeding = RotaryEmbedding(config.dim // config.n_heads, config.max_len)
-        self.embed_tokens = Embedding(config.vocab_size, config.n_dim)
+        self.embed_tokens = Embedding(config.vocab_size, config.dim)
         
         self.layers = nn.ModuleList([
-            DecoderBlock(config.n_dim, config.n_head, config.d_ffn, config.n_kvhead, config.norm_eps, layer_id)
+            DecoderBlock(config.dim, config.n_heads, config.dim_ffn, config.n_kv_heads, 
-            for layer_id in range(config.n_layer)
+                         config.norm_eps, config.use_qk_norm, config.use_gated_attention, layer_id)
+            for layer_id in range(config.n_layers)
         ])
         
-        self.norm = RMSNorm(config.n_dim, config.norm_eps)
+        self.norm = RMSNorm(config.dim, config.norm_eps)
-        self.lm_head = Linear(config.n_dim, config.vocab_size)
+        self.lm_head = Linear(config.dim, config.vocab_size)
         
         if self.config.tie_weight == True:
             self.lm_head.weight = self.embed_tokens.weight

tests/conftest.py

@@ -83,19 +83,19 @@ def base_test_env(request: pytest.FixtureRequest):
     n_dim_choices = [8, 16, 32]
     n_head_choices = [2, 4]
     
-    n_dim = int(np.random.choice(n_dim_choices))
+    dim = int(np.random.choice(n_dim_choices))
-    n_head = int(np.random.choice(n_head_choices))
+    n_heads = int(np.random.choice(n_head_choices))
-    n_kvhead = n_head // 2
+    n_kv_heads = n_heads // 2
-    d_ffn = n_dim * 2
+    dim_ffn = dim * 2
 
     config = {
         "vocab_size": 1000,
-        "n_dim": n_dim,
+        "dim": dim,
-        "n_head": n_head,
+        "n_heads": n_heads,
-        "n_kvhead": n_kvhead,
+        "n_kv_heads": n_kv_heads,
-        "d_ffn": d_ffn,
+        "dim_ffn": dim_ffn,
-        "m_len": 1024,
+        "max_len": 1024,
-        "n_layer": 4,
+        "n_layers": 4,
         "norm_eps": 1e-5
     }
     

tests/test_module.py

@@ -22,12 +22,12 @@ def test_env(request: pytest.FixtureRequest):
     
     config = {
         "vocab_size": 1000,
-        "n_dim": 128,
+        "dim": 128,
-        "n_head": 4,
+        "n_heads": 4,
-        "n_kvhead": 2,
+        "n_kv_heads": 2,
-        "d_ffn": 256,
+        "dim_ffn": 256,
-        "m_len": 64,
+        "max_len": 64,
-        "n_layer": 2,
+        "n_layers": 2,
         "norm_eps": 1e-5
     }
     with open(config_path, 'w') as f:
@@ -64,9 +64,9 @@ def test_model_parameter(test_env):
 def test_transformer(test_env):
     model = test_env["model"]
     input_ids = torch.randint(0, test_env["transformer_config"].vocab_size, 
-                              (4, test_env["transformer_config"].m_len))
+                              (4, test_env["transformer_config"].max_len))
     output_logits = model(input_ids)["logits"]
-    target_shape = (4, test_env["transformer_config"].m_len, test_env["transformer_config"].vocab_size)
+    target_shape = (4, test_env["transformer_config"].max_len, test_env["transformer_config"].vocab_size)
     assert output_logits.shape == target_shape
     
 # generator
@@ -80,7 +80,7 @@ def test_embedding_encoder_core(test_env):
     
     single_emb = encoder.encode("测试文本")
     assert isinstance(single_emb, torch.Tensor)
-    assert single_emb.shape[-1] == test_env["transformer_config"].n_dim
+    assert single_emb.shape[-1] == test_env["transformer_config"].dim
     
 
     batch_emb = encoder.encode(["测试1", "测试2"])

tests/test_tie_weight.py

@@ -16,12 +16,12 @@ def transformer_test_env():
     
     config = {
         "vocab_size": 1000,
-        "n_dim": 128,
+        "dim": 128,
-        "n_head": 4,
+        "n_heads": 4,
-        "n_kvhead": 2,
+        "n_kv_heads": 2,
-        "d_ffn": 256,
+        "dim_ffn": 256,
-        "m_len": 64,
+        "max_len": 64,
-        "n_layer": 2,
+        "n_layers": 2,
         "norm_eps": 1e-5
     }
     

tools/benchmark.py

@@ -28,7 +28,7 @@ class GenerationBenchmark:
     def _initialize_kv_cache(self, batch_size: int) -> list:
         """初始化KV缓存"""
         config = self.config
-        shape = (batch_size, config.m_len, config.n_layer, config.n_kvhead, config.n_dim // config.n_head)
+        shape = (batch_size, config.max_len, config.n_layers, config.n_kv_heads, config.dim // config.n_heads)
         k_cache = torch.zeros(shape, device=self.device, dtype=self.dtype)
         v_cache = torch.zeros(shape, device=self.device, dtype=self.dtype)
         return (k_cache, v_cache)
@@ -175,12 +175,12 @@ def print_benchmark_result(result: BenchmarkResult):
 if __name__ == "__main__":
     config = ModelConfig(
         vocab_size=10000,
-        n_dim=1536,
+        dim=1536,
-        n_head=24,
+        n_heads=24,
-        n_kvhead=4,
+        n_kv_heads=4,
-        d_ffn=6912,
+        dim_ffn=6912,
-        m_len=2048,
+        max_len=2048,
-        n_layer=24,
+        n_layers=24,
         norm_eps=1e-5,
     )
     

tools/train.py

@@ -111,7 +111,7 @@ def train(
     parameter.load(param_path)
 
     if window_size is None:
-        window_size = parameter.config.m_len
+        window_size = parameter.config.max_len
 
     model = parameter.model