Setting upper bound for accelerate to avoid torch dependency conflicts

Setting upper bound for accelerate to avoid torch dependency conflicts #957