fix ci bug

python/jittor/test/test_models.py

@@ -61,6 +61,7 @@ class test_models(unittest.TestCase):
 
     @unittest.skipIf(not jt.has_cuda, "Cuda not found")
     @jt.flag_scope(use_cuda=1)
+    @torch.no_grad()
     def test_models(self):
         def to_cuda(x):
             if jt.has_cuda:
@@ -96,6 +97,9 @@ class test_models(unittest.TestCase):
             diff = relative_error.mean()
             assert diff < threshold, f"[*] {test_model} forward fails..., Relative Error: {diff}"
             print(f"[*] {test_model} forword passes with Relative Error {diff}")
+            jt.clean()
+            jt.gc()
+            torch.cuda.empty_cache()
         print('all models pass test.')
         
 if __name__ == "__main__":

src/mem/allocator/cuda_device_allocator.cc

@@ -19,7 +19,10 @@ void* CudaDeviceAllocator::alloc(size_t size, size_t& allocation) {
     try {
         checkCudaErrors(cudaMalloc(&ptr, size));
         return ptr;
-    } catch (...) {}
+    } catch (...) {
+        // clean the last error
+        cudaGetLastError();
+    }
     LOGw << "Unable to alloc cuda device memory, use unify memory instead. "
         "This may cause low performance.";
     display_memory_info(__FILELINE__);