Closed bmikaili closed 6 months ago
I think it's a bug that I accidentally wrote when cleaning up the code. I would fix it later today :D
Ah okay, happens :D At least you know what it is!
Should be fixed, thanks for notifying it!
Actually have to reopen this 😅 Now I am getting this error. I run it on a GTX 4090 on runpod.io.
Welcome to Ubuntu 22.04.3 LTS (GNU/Linux 5.4.0-152-generic x86_64)
if (final_output) {
set_results_to_output<output_vec_size>(value, base_offsets);
} else {
#pragma unroll
for (int i = 0; i < output_vec_size; i++) {
*(out[i]) = get_accumulated_output(out[i], value[i]);
}
}
} else {
if (accumulate) {
#pragma unroll
for (int i = 0; i < output_vec_size; i++) {
value[i] = reducer::combine((*acc)[i], value[i]);
}
}
if (final_output) {
set_results_to_output<output_vec_size>(value, base_offsets);
} else {
*acc = value;
}
}
}
}
return value;
}
};
extern "C"
__launch_bounds__(512, 4)
__global__ void reduction_prod_kernel(ReduceJitOp r){
r.run();
}
nvrtc: error: invalid value for --gpu-architecture (-arch)
I've just updated the code. If you're still encountering the same error, could you walk me through the steps you took that led to it? This will help me understand the issue better. Thanks!
All good now! Thanks so much for fixing it
I'm running train_densify_prune on a scene of my own. After about 15k iterations i get:
I run it with
python train_densify_prune.py -s dataset
on a dataset converted withconvert.py
.Any ideas what could be going wrong?