fix

Author: orca-zhang

examples/server/atomic_hash_map.hpp

@@ -295,8 +295,8 @@ namespace atomic {
   friend ConstIterator;
 
   hash_map(size_t maxSize,
-                    float maxLoadFactor = 0.8f,
-                    const Allocator& alloc = Allocator())
+           float maxLoadFactor = 0.8f,
+           const Allocator& alloc = Allocator())
     : allocator_(alloc)
   {
     size_t capacity = size_t(maxSize / (maxLoadFactor > 1.0f ? 1.0f : maxLoadFactor) + 128);

src/llama-model.cpp

@@ -2949,7 +2949,8 @@ bool llama_model::load_tensors(llama_model_loader & ml) {
                             ggml_context * ctx = ctx_for_buft(buft);
 
                             auto trans_wkv_b = ggml_transpose(ctx, layer.wkv_b);
-                            auto wkv_b_copied = ggml_dup(ctx, trans_wkv_b);
+                            auto wkv_b_copied = ggml_new_tensor_2d(ctx, trans_wkv_b->type, trans_wkv_b->ne[0], trans_wkv_b->ne[1]);
+                            ggml_cpy(ctx, trans_wkv_b, wkv_b_copied);
                             layer.wk_b = ggml_view_2d(ctx, wkv_b_copied, wkv_b_copied->ne[0], n_embd_head_qk_nope, n_head, 0);
                             layer.wv_b = ggml_view_2d(ctx, wkv_b_copied, wkv_b_copied->ne[0], n_embd_head_v, n_head, n_embd_head_qk_nope * n_head);
                         }