fix ubatch, autoselect vulkan dgpu if possible

Author: LostRuins

gpttype_adapter.cpp

@@ -1053,6 +1053,7 @@ ModelLoadResult gpttype_load_model(const load_model_inputs inputs, FileFormat in
         #endif
 
         llama_ctx_params.n_batch = kcpp_params->n_batch;
+        llama_ctx_params.n_ubatch = kcpp_params->n_ubatch;
         llama_ctx_params.n_threads = kcpp_params->n_threads;
         llama_ctx_params.n_threads_batch = kcpp_params->n_threads_batch;
 

koboldcpp.py

@@ -1767,6 +1767,7 @@ def hide_tooltip(event):
     CLDevicesNames = ["","","",""]
     CUDevicesNames = ["","","","",""]
     VKDevicesNames = ["","","",""]
+    VKIsDGPU = [0,0,0,0]
     MaxMemory = [0]
 
     tabcontent = {}
@@ -2005,11 +2006,18 @@ def auto_gpu_heuristics():
         try: # Get Vulkan names
             output = subprocess.run(['vulkaninfo','--summary'], capture_output=True, text=True, check=True, encoding='utf-8').stdout
             devicelist = [line.split("=")[1].strip() for line in output.splitlines() if "deviceName" in line]
+            devicetypes = [line.split("=")[1].strip() for line in output.splitlines() if "deviceType" in line]
             idx = 0
             for dname in devicelist:
                 if idx<len(VKDevicesNames):
                     VKDevicesNames[idx] = dname
                     idx += 1
+            if len(devicetypes) == len(devicelist):
+                idx = 0
+                for dvtype in devicetypes:
+                    if idx<len(VKIsDGPU):
+                        VKIsDGPU[idx] = (1 if dvtype=="PHYSICAL_DEVICE_TYPE_DISCRETE_GPU" else 0)
+                        idx += 1
         except Exception as e:
             pass
 
@@ -2029,6 +2037,12 @@ def auto_gpu_heuristics():
                 runopts_var.set("Use CuBLAS")
             elif "Use hipBLAS (ROCm)" in runopts:
                 runopts_var.set("Use hipBLAS (ROCm)")
+        elif exitcounter < 100 and (1 in VKIsDGPU) and runmode_untouched and "Use Vulkan" in runopts:
+            for i in range(0,len(VKIsDGPU)):
+                if VKIsDGPU[i]==1:
+                    runopts_var.set("Use Vulkan")
+                    gpu_choice_var.set(str(i+1))
+                    break
 
         changed_gpu_choice_var()
         return