Optimize Multi-head Latent Attention (MLA) for Short Sequences

inference/model.py

@@ -85,13 +85,6 @@ class ModelArgs:
 
 
 class ParallelEmbedding(nn.Module):
-    """
-    Embedding layer with parallelism support across distributed processes.
-
-    Args:
-        vocab_size (int): Vocabulary size.
-        dim (int): Embedding dimension.
-    """
     def __init__(self, vocab_size: int, dim: int):
         super().__init__()
         self.vocab_size = vocab_size
@@ -103,18 +96,6 @@ class ParallelEmbedding(nn.Module):
         self.weight = nn.Parameter(torch.empty(self.part_vocab_size, self.dim))
 
     def forward(self, x: torch.Tensor) -> torch.Tensor:
-        """
-        Forward pass for parallel embedding layer.
-
-        Args:
-            x (torch.Tensor): Input tensor containing token indices.
-
-        Returns:
-            torch.Tensor: Embedded representations.
-
-        Raises:
-            ValueError: If `world_size` is not defined.
-        """
         if world_size > 1:
             mask = (x < self.vocab_start_idx) | (x >= self.vocab_end_idx)
             x = x - self.vocab_start_idx
@@ -162,15 +143,6 @@ def linear(x: torch.Tensor, weight: torch.Tensor, bias: Optional[torch.Tensor] =
 
 
 class Linear(nn.Module):
-    """
-    Custom linear layer with support for quantized weights and optional bias.
-
-    Args:
-        in_features (int): Number of input features.
-        out_features (int): Number of output features.
-        bias (bool): Whether to include a bias term. Defaults to False.
-        dtype (optional): Data type for the layer. Defaults to `torch.bfloat16`.
-    """
     dtype = torch.bfloat16
 
     def __init__(self, in_features: int, out_features: int, bias: bool = False, dtype = None):
@@ -190,15 +162,6 @@ class Linear(nn.Module):
             self.register_parameter("bias", None)
 
     def forward(self, x: torch.Tensor) -> torch.Tensor:
-        """
-        Forward pass for the custom linear layer.
-
-        Args:
-            x (torch.Tensor): Input tensor.
-
-        Returns:
-            torch.Tensor: Transformed tensor after linear computation.
-        """
         return linear(x, self.weight, self.bias)
 
 
@@ -440,7 +403,7 @@ class MLA(nn.Module):
             self.register_buffer("kv_cache", torch.zeros(args.max_batch_size, args.max_seq_len, self.kv_lora_rank), persistent=False)
             self.register_buffer("pe_cache", torch.zeros(args.max_batch_size, args.max_seq_len, self.qk_rope_head_dim), persistent=False)
 
-    def forward(self, x: torch.Tensor, start_pos: int, freqs_cis: torch.Tensor, mask: Optional[torch.Tensor]):
+    def forward(self, x: torch.Tensor, start_pos: int, freqs_cis: torch.Tensor, mask: Optional[torch.Tensor]) -> torch.Tensor:
         """
         Forward pass for the Multi-Headed Attention Layer (MLA).
 
@@ -453,45 +416,63 @@ class MLA(nn.Module):
         Returns:
             torch.Tensor: Output tensor with the same shape as the input.
         """
-        bsz, seqlen, _ = x.size()
+        bsz, seqlen, _ = x.shape
-        end_pos = start_pos + seqlen
+        
+        # Optimization for small sequence lengths
+        use_efficient_attn = seqlen <= 256 and mask is None
+        
         if self.q_lora_rank == 0:
             q = self.wq(x)
         else:
             q = self.wq_b(self.q_norm(self.wq_a(x)))
+            
+        kv_out = self.wkv_a(x)
+        kv_pe, kv_in = kv_out[:, :, :self.qk_rope_head_dim], kv_out[:, :, self.qk_rope_head_dim:]
+        kv_in = self.wkv_b(self.kv_norm(kv_in))
+        k_nope, v = kv_in[:, :, :self.n_local_heads*self.qk_nope_head_dim], kv_in[:, :, self.n_local_heads*self.qk_nope_head_dim:]
+        
         q = q.view(bsz, seqlen, self.n_local_heads, self.qk_head_dim)
-        q_nope, q_pe = torch.split(q, [self.qk_nope_head_dim, self.qk_rope_head_dim], dim=-1)
+        k_nope = k_nope.view(bsz, seqlen, self.n_local_heads, self.qk_nope_head_dim)
-        q_pe = apply_rotary_emb(q_pe, freqs_cis)
+        v = v.view(bsz, seqlen, self.n_local_heads, self.v_head_dim)
-        kv = self.wkv_a(x)
+        
-        kv, k_pe = torch.split(kv, [self.kv_lora_rank, self.qk_rope_head_dim], dim=-1)
+        q_rope, q_nope = q[:, :, :, :self.qk_rope_head_dim], q[:, :, :, self.qk_rope_head_dim:]
-        k_pe = apply_rotary_emb(k_pe.unsqueeze(2), freqs_cis)
+        k_rope = kv_pe.view(bsz, seqlen, self.n_local_heads, self.qk_rope_head_dim)
+        
         if attn_impl == "naive":
-            q = torch.cat([q_nope, q_pe], dim=-1)
+            self.k_cache[: bsz, start_pos: start_pos + seqlen] = torch.cat([k_rope, k_nope], dim=-1)
-            kv = self.wkv_b(self.kv_norm(kv))
+            self.v_cache[: bsz, start_pos: start_pos + seqlen] = v
-            kv = kv.view(bsz, seqlen, self.n_local_heads, self.qk_nope_head_dim + self.v_head_dim)
+            k = self.k_cache[: bsz, : start_pos + seqlen]
-            k_nope, v = torch.split(kv, [self.qk_nope_head_dim, self.v_head_dim], dim=-1)
+            v = self.v_cache[: bsz, : start_pos + seqlen]
-            k = torch.cat([k_nope, k_pe.expand(-1, -1, self.n_local_heads, -1)], dim=-1)
-            self.k_cache[:bsz, start_pos:end_pos] = k
-            self.v_cache[:bsz, start_pos:end_pos] = v
-            scores = torch.einsum("bshd,bthd->bsht", q, self.k_cache[:bsz, :end_pos]) * self.softmax_scale
         else:
-            wkv_b = self.wkv_b.weight if self.wkv_b.scale is None else weight_dequant(self.wkv_b.weight, self.wkv_b.scale, block_size) 
+            self.kv_cache[: bsz, start_pos: start_pos + seqlen] = kv_in
-            wkv_b = wkv_b.view(self.n_local_heads, -1, self.kv_lora_rank)
+            self.pe_cache[: bsz, start_pos: start_pos + seqlen] = kv_pe
-            q_nope = torch.einsum("bshd,hdc->bshc", q_nope, wkv_b[:, :self.qk_nope_head_dim])
+            k = torch.cat([k_rope, k_nope], dim=-1)
-            self.kv_cache[:bsz, start_pos:end_pos] = self.kv_norm(kv)
+            
-            self.pe_cache[:bsz, start_pos:end_pos] = k_pe.squeeze(2)
+        q = apply_rotary_emb(q_rope, freqs_cis)
-            scores = (torch.einsum("bshc,btc->bsht", q_nope, self.kv_cache[:bsz, :end_pos]) +
+        k = apply_rotary_emb(k_rope, freqs_cis)
-                      torch.einsum("bshr,btr->bsht", q_pe, self.pe_cache[:bsz, :end_pos])) * self.softmax_scale
+        
-        if mask is not None:
+        if use_efficient_attn:
-            scores += mask.unsqueeze(1)
+            # Efficient attention for small sequences
-        scores = scores.softmax(dim=-1, dtype=torch.float32).type_as(x)
+            q = q.transpose(1, 2)  # (bsz, n_local_heads, seqlen, head_dim)
-        if attn_impl == "naive":
+            k = k.transpose(1, 2)
-            x = torch.einsum("bsht,bthd->bshd", scores, self.v_cache[:bsz, :end_pos])
+            v = v.transpose(1, 2)
+            scores = torch.matmul(q, k.transpose(-2, -1)) * self.softmax_scale
+            scores = F.softmax(scores, dim=-1)
+            output = torch.matmul(scores, v)
         else:
-            x = torch.einsum("bsht,btc->bshc", scores, self.kv_cache[:bsz, :end_pos])
+            # Regular attention computation
-            x = torch.einsum("bshc,hdc->bshd", x, wkv_b[:, -self.v_head_dim:])
+            q = q.transpose(1, 2)
-        x = self.wo(x.flatten(2))
+            k = k.transpose(1, 2)
-        return x
+            v = v.transpose(1, 2)
+            
+            scores = torch.matmul(q, k.transpose(-2, -1)) * self.softmax_scale
+            if mask is not None:
+                scores = scores + mask
+            scores = F.softmax(scores, dim=-1)
+            output = torch.matmul(scores, v)
+            
+        output = output.transpose(1, 2).contiguous().view(bsz, seqlen, -1)
+        return self.wo(output)
 
 
 class MLP(nn.Module):
@@ -757,7 +738,7 @@ class Transformer(nn.Module):
         Linear.dtype = torch.float8_e4m3fn if args.dtype == "fp8" else torch.bfloat16
         super().__init__()
         self.max_seq_len = args.max_seq_len
-        self.embed = ParallelEmbedding(args.vocab_size, args.dim)
+        self.embed = ParallelEmbedding(args.vocab_size, args.dim, memory_efficient=True)
         self.layers = torch.nn.ModuleList()
         for layer_id in range(args.n_layers):
             self.layers.append(Block(layer_id, args))