hopefully fixed the ooms for good

README.md

@@ -32,6 +32,7 @@ What does it mean? You get llama.cpp with a fancy UI, persistent stories, editin
   - CLBlast - tested with https://github.com/CNugteren/CLBlast . If you wish to compile it you will need to reference the OpenCL files. It will only generate the ".lib" file if you compile using MSVC.
   - OpenBLAS - tested with https://github.com/xianyi/OpenBLAS . 
   - Move the respectives .lib files to the /lib folder of your project, overwriting the older files.
+  - Also, replace the existing versions of the corresponding .dll files located in the project directory root (e.g. libopenblas.dll).
   - Make the KoboldCPP project using the instructions above. 
 
 ## OSX and Linux

otherarch/gpt2_v2.cpp

@@ -371,7 +371,7 @@ bool gpt2_eval(
     const int n_vocab = hparams.n_vocab;
 
     //todo: there is a bug that causes the buffer to oom and I cannot figure it out, hack to increase size for now  
-    static size_t buf_size = 512u*1024*1024;    
+    static size_t buf_size = 256u*1024*1024;    
     static void * buf = malloc(buf_size);
 
     if (mem_per_token > 0 && mem_per_token*N*1.6 > buf_size) {

otherarch/gptj_v2.cpp

@@ -382,11 +382,11 @@ bool gptj_eval(
     const int d_key = n_embd/n_head;
 
     //todo: there is a bug that causes the buffer to oom and I cannot figure it out, hack to increase size for now  
-    static size_t buf_size = 512u*1024*1024;    
+    static size_t buf_size = 256u*1024*1024;    
     static void * buf = malloc(buf_size);
 
-    if (mem_per_token > 0 && mem_per_token*N*1.4 > buf_size) {
-        const size_t buf_size_new = 1.6*(mem_per_token*N); // add 10% to account for ggml object overhead
+    if (mem_per_token > 0 && mem_per_token*N*1.6 > buf_size) {
+        const size_t buf_size_new = 1.8*(mem_per_token*N); // add 10% to account for ggml object overhead
         //printf("\n%s: reallocating buffer from %zu to %zu bytes\n", __func__, buf_size, buf_size_new);
 
         // reallocate

otherarch/neox.cpp

@@ -364,11 +364,11 @@ bool stablelm_eval(
     const int n_vocab = hparams.n_vocab;
     const int n_rot   = hparams.n_rot;
 
-    static size_t buf_size = 512u*1024*1024;    
+    static size_t buf_size = 256u*1024*1024;    
     static void * buf = malloc(buf_size);
 
-    if (mem_per_token > 0 && mem_per_token*N*1.4 > buf_size) {
-        const size_t buf_size_new = 1.6*(mem_per_token*N); // add 10% to account for ggml object overhead
+    if (mem_per_token > 0 && mem_per_token*N*1.6 > buf_size) {
+        const size_t buf_size_new = 1.8*(mem_per_token*N); // add 10% to account for ggml object overhead
         //printf("\n%s: reallocating buffer from %zu to %zu bytes\n", __func__, buf_size, buf_size_new);
 
         // reallocate