[skip ci] Auto-format code with clang-format and black

Author: github-actions[bot]

csrc/lib/ops/matmul/op.cu

@@ -20,8 +20,10 @@ __global__ void matMulKernel(ScalarT* M, ScalarT* N, ScalarT* P, int32_t Width)
 
     fp32_t Pvalue = 0.0F;
     for (int32_t ph = 0; ph < Width / TILE_SIZE; ++ph) {
-        Mds[threadIdx.y][threadIdx.x] = M[Row * Width + (ph * TILE_SIZE + threadIdx.x)];
-        Nds[threadIdx.y][threadIdx.x] = N[(ph * TILE_SIZE + threadIdx.y) * Width + Col];
+        Mds[threadIdx.y][threadIdx.x] =
+            M[Row * Width + (ph * TILE_SIZE + threadIdx.x)];
+        Nds[threadIdx.y][threadIdx.x] =
+            N[(ph * TILE_SIZE + threadIdx.y) * Width + Col];
         __syncthreads();
 
         for (int32_t k = 0; k < TILE_SIZE; ++k) {

src/pmpp/models/attention.py

@@ -62,9 +62,7 @@ def forward(
         # v -> (num_heads, kv_len, head_size)
         v = v.view(kv_len, self.num_heads, self.head_size).transpose(0, 1)
         # scores -> (num_heads, q_len, kv_len)
-        scores = torch.matmul(q, k.transpose(-1, -2)) / math.sqrt(
-            self.head_size
-        )
+        scores = torch.matmul(q, k.transpose(-1, -2)) / math.sqrt(self.head_size)
         # scores -> (num_heads, q_len, kv_len)
         scores = scores + mask if mask is not None else scores
         # scores -> (num_heads, q_len, kv_len)
@@ -247,9 +245,7 @@ def forward(
             # out -> (seq_len, embed_dim)
             # k -> (kv_len, embed_dim)
             # v -> (kv_len, embed_dim)
-            out, k, v = self.cached_mha(
-                embedded_prompt, self.k_cache, self.v_cache
-            )
+            out, k, v = self.cached_mha(embedded_prompt, self.k_cache, self.v_cache)
 
         # Update k cache and v cache
         # [NOTE]