amendment

Author: YangShuai52

ggml/src/ggml-cann/aclnn_ops.cpp

@@ -2174,14 +2174,16 @@ static void aclnn_cache_init(ggml_backend_cann_context& ctx, ggml_tensor* dst,
 
     bool is_q = (std::strncmp(dst->name, "Qcur-", 5) == 0);
     bool is_k = (std::strncmp(dst->name, "Kcur-", 5) == 0);
+
+    // used for accuracy testing
     bool is_attention = is_q || is_k;
 
     if(ctx.init_ptr == nullptr || !is_attention) {
         // theta_scale arange, [0,1,...,ne00/2 - 1]
         if(ctx.init_ptr != nullptr){
             ACL_CHECK(aclrtFree(ctx.init_ptr));
         }
-        ACL_CHECK(aclrtMalloc(&ctx.init_ptr,theta_scale_length * sizeof(float_t), ACL_MEM_MALLOC_HUGE_FIRST));
+        ACL_CHECK(aclrtMalloc(&ctx.init_ptr, theta_scale_length * sizeof(float_t), ACL_MEM_MALLOC_HUGE_FIRST));
 
         aclTensor* acl_theta_scale_tensor =
             ggml_cann_create_tensor(ctx.init_ptr, ACL_FLOAT, sizeof(float_t),

ggml/src/ggml-cann/common.h

@@ -371,7 +371,7 @@ struct ggml_backend_cann_context {
     void* init_ptr = nullptr;
     void* sin_ptr = nullptr;
     void* cos_ptr = nullptr;
-    int64_t max_position_length = 200000;
+    int64_t max_prompt_length = 65536;
 #ifdef USE_ACL_GRAPH
     /// Cached CANN ACL graph used for executing the current ggml computation graph.
     std::unique_ptr<ggml_cann_graph> cann_graph;