I try to build a model using relay.build . but when i add “-libs=cublas” to target , it raise RuntimeError like this.it seems that the batch_matmul_cublas has some errors.what’s the reason of this error.And how can i fix it. Thanks very much!!
the code is:
target = “cuda -keys=cuda,gpu -libs=cublas -arch=sm_86 -max_num_threads=1024 -thread_warp_size=32” with tvm.transform.PassContext(opt_level=1): lib = relay.build(mod, target=target, params=params)
RuntimeError: Memory verification failed with the following errors: PrimFunc([placeholder, batch_matmul_cublas]) attrs={“hash”: “2d669cac08cf6bfc”, “target”: cuda -keys=cuda,gpu -arch=sm_86 -libs=cublas -max_num_threads=1024 -thread_warp_size=32, “tir.noalias”: (bool)1, “global_symbol”: “tvmgen_default_fused_nn_batch_matmul_1”, “from_legacy_te_schedule”: (bool)1} { allocate compute[float16 * 2560], storage_scope = global for (i0, 0, 10) { for (i1, 0, 256) { let cse_var_1 = ((i0*256) + i1) compute[cse_var_1] = placeholder[cse_var_1] } } // attr [0] extern_scope = 0 tir.tvm_call_packed(“tvm.contrib.cublas.batch_matmul”, tir.tvm_stack_make_array(placeholder, tir.tvm_stack_make_shape(10, 256, 1), 0, 3, 0h, 0), tir.tvm_stack_make_array(compute, tir.tvm_stack_make_shape(10, 256, 1), 0, 3, 0h, 0), tir.tvm_stack_make_array(batch_matmul_cublas, tir.tvm_stack_make_shape(10, 256, 256), 0, 3, 0h, 0), 0, 1) }
