clip.hpp

@@ -553,9 +553,10 @@ protected:
     void init_params(struct ggml_context* ctx, const String2GGMLType& tensor_types = {}, const std::string prefix = "") {
         enum ggml_type token_wtype = GGML_TYPE_F32;
         if (!force_clip_f32) {
-            token_wtype = get_type(prefix + "token_embedding.weight", tensor_types, GGML_TYPE_F32);
-            if (!support_get_rows(token_wtype)) {
-                token_wtype = GGML_TYPE_F32;
+            auto tensor_type                = tensor_types.find(prefix + "token_embedding.weight");
+            std::set<ggml_type> allow_types = {GGML_TYPE_F16, GGML_TYPE_Q8_0, GGML_TYPE_Q5_1, GGML_TYPE_Q5_0, GGML_TYPE_Q4_1, GGML_TYPE_Q4_0};
+            if (tensor_type != tensor_types.end() && allow_types.find(tensor_type->second) != allow_types.end()) {
+                token_wtype = tensor_type->second;
             }
         }
         enum ggml_type position_wtype       = GGML_TYPE_F32;

ggml

@@ -1 +1 @@
-Subproject commit 7bffd79a4bec72e9a3bfbedb582a218b84401c13
+Subproject commit 553c44706c3cc6e4077f4ab214923fc4c20a013c

ggml_extend.hpp

@@ -1980,24 +1980,13 @@ public:
     }
 };
 
-__STATIC_INLINE__ bool support_get_rows(ggml_type wtype) {
-    std::set<ggml_type> allow_types = {GGML_TYPE_F16, GGML_TYPE_Q8_0, GGML_TYPE_Q5_1, GGML_TYPE_Q5_0, GGML_TYPE_Q4_1, GGML_TYPE_Q4_0};
-    if (allow_types.find(wtype) != allow_types.end()) {
-        return true;
-    }
-    return false;
-}
-
 class Embedding : public UnaryBlock {
 protected:
     int64_t embedding_dim;
     int64_t num_embeddings;
     void init_params(struct ggml_context* ctx, const String2GGMLType& tensor_types, const std::string prefix = "") {
         enum ggml_type wtype = get_type(prefix + "weight", tensor_types, GGML_TYPE_F32);
-        if (!support_get_rows(wtype)) {
-            wtype = GGML_TYPE_F32;
-        }
-        params["weight"] = ggml_new_tensor_2d(ctx, wtype, embedding_dim, num_embeddings);
+        params["weight"]     = ggml_new_tensor_2d(ctx, wtype, embedding_dim, num_embeddings);
     }
 
 public: