diff --git a/docs/core_concept/brainpy_transform_concept-old.ipynb b/docs/core_concept/brainpy_transform_concept-old.ipynb
deleted file mode 100644
index c8b3a771b..000000000
--- a/docs/core_concept/brainpy_transform_concept-old.ipynb
+++ /dev/null
@@ -1,654 +0,0 @@
-{
- "cells": [
-  {
-   "cell_type": "markdown",
-   "metadata": {
-    "collapsed": true,
-    "jupyter": {
-     "outputs_hidden": true
-    }
-   },
-   "source": [
-    "# Concept 1: Object-oriented Transformation"
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "metadata": {},
-   "source": [
-    "@[Chaoming Wang](https://github.com/chaoming0625)"
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "metadata": {},
-   "source": [
-    "Most computation in BrainPy relies on [JAX](https://jax.readthedocs.io/en/latest/).\n",
-    "JAX has provided wonderful transformations, including differentiation, vecterization, parallelization and just-in-time compilation, for Python programs. If you are not familiar with it, please see its [documentation](https://jax.readthedocs.io/en/latest/).\n",
-    "\n",
-    "However, JAX only supports functional programming, i.e., transformations for Python functions. This is not what we want. Brain Dynamics Modeling need object-oriented programming.\n",
-    "\n",
-    "To meet this requirement, BrainPy defines the interface for object-oriented (OO) transformations. These OO transformations can be easily performed for BrainPy objects.\n",
-    "\n",
-    "In this section, let's talk about the BrainPy concept of object-oriented transformations."
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": 4,
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "import brainpy as bp\n",
-    "import brainpy.math as bm\n",
-    "\n",
-    "# bm.set_platform('cpu')"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": 5,
-   "metadata": {
-    "collapsed": false,
-    "jupyter": {
-     "outputs_hidden": false
-    }
-   },
-   "outputs": [
-    {
-     "data": {
-      "text/plain": [
-       "'2.3.0'"
-      ]
-     },
-     "execution_count": 5,
-     "metadata": {},
-     "output_type": "execute_result"
-    }
-   ],
-   "source": [
-    "bp.__version__"
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "metadata": {},
-   "source": [
-    "## Illustrating example: Training a network"
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "metadata": {},
-   "source": [
-    "To illustrate this concept, we need a demonstration example. Here, we choose the popular neural network training as the illustrating case.\n",
-    "\n",
-    "In this training case, we want to teach the neural network to correctly classify a random array as two labels (`True` or `False`). That is, we have the training data:"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": 6,
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "num_in = 100\n",
-    "num_sample = 256\n",
-    "X = bm.random.rand(num_sample, num_in)\n",
-    "Y = (bm.random.rand(num_sample) < 0.5).astype(float)"
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "metadata": {},
-   "source": [
-    "We use a two-layer feedforward network:"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": 7,
-   "metadata": {},
-   "outputs": [
-    {
-     "name": "stdout",
-     "output_type": "stream",
-     "text": [
-      "Sequential(\n",
-      "  [0] Linear0\n",
-      "  [1] relu\n",
-      "  [2] Linear1\n",
-      ")\n"
-     ]
-    }
-   ],
-   "source": [
-    "class Linear(bp.BrainPyObject):\n",
-    "    def __init__(self, n_in, n_out):\n",
-    "        super().__init__()\n",
-    "        self.num_in = n_in\n",
-    "        self.num_out = n_out\n",
-    "        init = bp.init.XavierNormal()\n",
-    "        self.W = bm.Variable(init((n_in, n_out)))\n",
-    "        self.b = bm.Variable(bm.zeros((1, n_out)))\n",
-    "\n",
-    "    def __call__(self, x):\n",
-    "        return x @ self.W + self.b\n",
-    "\n",
-    "\n",
-    "net = bp.Sequential(Linear(num_in, 20),\n",
-    "                    bm.relu,\n",
-    "                    Linear(20, 2))\n",
-    "print(net)"
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "metadata": {},
-   "source": [
-    "Here, we use a supervised learning training paradigm. "
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": 8,
-   "metadata": {},
-   "outputs": [
-    {
-     "name": "stdout",
-     "output_type": "stream",
-     "text": [
-      "Train 400 epoch, loss = 0.6710\n",
-      "Train 800 epoch, loss = 0.5992\n",
-      "Train 1200 epoch, loss = 0.5332\n",
-      "Train 1600 epoch, loss = 0.4720\n",
-      "Train 2000 epoch, loss = 0.4189\n",
-      "Train 2400 epoch, loss = 0.3736\n",
-      "Train 2800 epoch, loss = 0.3335\n",
-      "Train 3200 epoch, loss = 0.2972\n",
-      "Train 3600 epoch, loss = 0.2644\n",
-      "Train 4000 epoch, loss = 0.2346\n"
-     ]
-    }
-   ],
-   "source": [
-    "rng = bm.random.RandomState(123)\n",
-    "\n",
-    "\n",
-    "# Loss function\n",
-    "@bm.to_object(child_objs=net, dyn_vars=rng)\n",
-    "def loss():\n",
-    "    # shuffle the data\n",
-    "    key = rng.split_key()\n",
-    "    x_data = rng.permutation(X, key=key)\n",
-    "    y_data = rng.permutation(Y, key=key)\n",
-    "    # prediction\n",
-    "    predictions = net(dict(), x_data)\n",
-    "    # loss\n",
-    "    l = bp.losses.cross_entropy_loss(predictions, y_data)\n",
-    "    return l\n",
-    "\n",
-    "\n",
-    "# Gradient function\n",
-    "grad = bm.grad(loss, grad_vars=net.vars(), return_value=True)\n",
-    "\n",
-    "# Optimizer\n",
-    "optimizer = bp.optim.SGD(lr=1e-2, train_vars=net.vars())\n",
-    "\n",
-    "\n",
-    "# Training step\n",
-    "@bm.to_object(child_objs=(grad, optimizer))\n",
-    "def train(i):\n",
-    "    grads, l = grad()\n",
-    "    optimizer.update(grads)\n",
-    "    return l\n",
-    "\n",
-    "\n",
-    "num_step = 400\n",
-    "for i in range(0, 4000, num_step):\n",
-    "    # train 400 steps once\n",
-    "    ls = bm.for_loop(train, operands=bm.arange(i, i + num_step))\n",
-    "    print(f'Train {i + num_step} epoch, loss = {bm.mean(ls):.4f}')"
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "metadata": {},
-   "source": [
-    "In the above example, we have seen classical elements in a neural network training, such as \n",
-    "\n",
-    "- `net`: neural network\n",
-    "- `loss`: loss function\n",
-    "- `grad`: gradient function\n",
-    "- `optimizer`: parameter optimizer\n",
-    "- `train`: training step"
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "metadata": {},
-   "source": [
-    "In BrainPy, all these elements can be defined as class objects and can be used for performing OO transformations. "
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "metadata": {},
-   "source": [
-    "In essence, the concept of BrainPy object-oriented transformation has three components:\n",
-    "\n",
-    "- `BrainPyObject`: the base class for object-oriented programming\n",
-    "- `Variable`: the varibles in the class object, whose values are ready to be changed/updated during transformation\n",
-    "- `ObjectTransform`: the transformations for computation involving `BrainPyObject` and `Variable`"
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "metadata": {},
-   "source": [
-    "## ``BrainPyObject`` and its ``Variable``"
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "metadata": {},
-   "source": [
-    "``BrainPyObject`` is the base class for object-oriented programming in BrainPy. \n",
-    "It can be viewed as a container which contains all needed [Variable](../tutorial_math/arrays_and_variables.ipynb) for our computation."
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "metadata": {},
-   "source": [
-    "![](./imgs/net_with_two_linear.png)"
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "metadata": {},
-   "source": [
-    "In the above example, ``Linear`` object has two ``Variable``: *W* and *b*. The ``net`` we defined is further composed of two ``Linear`` objects. We can expect that four variables can be retrieved from it."
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": 9,
-   "metadata": {},
-   "outputs": [
-    {
-     "data": {
-      "text/plain": [
-       "dict_keys(['Linear0.W', 'Linear0.b', 'Linear1.W', 'Linear1.b'])"
-      ]
-     },
-     "execution_count": 9,
-     "metadata": {},
-     "output_type": "execute_result"
-    }
-   ],
-   "source": [
-    "net.vars().keys()"
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "metadata": {},
-   "source": [
-    "An important question is, **how to define `Variable` in a `BrainPyObject` so that we can retrieve all of them?**"
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "metadata": {},
-   "source": [
-    "Actually, all Variable instance which can be accessed by `self.` attribue can be retrived from a `BrainPyObject` recursively. \n",
-    "No matter how deep the composition of ``BrainPyObject``, once `BrainPyObject` instance and their `Variable` instances can be accessed by `self.` operation, all of them will be retrieved. "
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": 10,
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "class SuperLinear(bp.BrainPyObject):\n",
-    "    def __init__(self, ):\n",
-    "        super().__init__()\n",
-    "        self.l1 = Linear(10, 20)\n",
-    "        self.v1 = bm.Variable(3)\n",
-    "        \n",
-    "sl = SuperLinear()"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": 11,
-   "metadata": {},
-   "outputs": [
-    {
-     "data": {
-      "text/plain": [
-       "dict_keys(['SuperLinear0.v1', 'Linear2.W', 'Linear2.b'])"
-      ]
-     },
-     "execution_count": 11,
-     "metadata": {},
-     "output_type": "execute_result"
-    }
-   ],
-   "source": [
-    "# retrieve Variable\n",
-    "sl.vars().keys()"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": 12,
-   "metadata": {},
-   "outputs": [
-    {
-     "data": {
-      "text/plain": [
-       "dict_keys(['SuperLinear0', 'Linear2'])"
-      ]
-     },
-     "execution_count": 12,
-     "metadata": {},
-     "output_type": "execute_result"
-    }
-   ],
-   "source": [
-    "# retrieve BrainPyObject\n",
-    "sl.nodes().keys()"
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "metadata": {},
-   "source": [
-    "However, we cannot access the ``BrainPyObject`` or ``Variable`` which is in a Python container (like tuple, list, or dict). For this case, we can register our objects and variables through ``.register_implicit_vars()`` and ``.register_implicit_nodes()``:"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": 13,
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "class SuperSuperLinear(bp.BrainPyObject):\n",
-    "    def __init__(self, register=False):\n",
-    "        super().__init__()\n",
-    "        self.ss = [SuperLinear(), SuperLinear()]\n",
-    "        self.vv = {'v_a': bm.Variable(3)}\n",
-    "        if register:\n",
-    "            self.register_implicit_nodes(self.ss)\n",
-    "            self.register_implicit_vars(self.vv)"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": 14,
-   "metadata": {},
-   "outputs": [
-    {
-     "name": "stdout",
-     "output_type": "stream",
-     "text": [
-      "dict_keys([])\n",
-      "dict_keys(['SuperSuperLinear0'])\n"
-     ]
-    }
-   ],
-   "source": [
-    "# without register\n",
-    "ssl = SuperSuperLinear(register=False)\n",
-    "print(ssl.vars().keys())\n",
-    "print(ssl.nodes().keys())"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": 15,
-   "metadata": {},
-   "outputs": [
-    {
-     "name": "stdout",
-     "output_type": "stream",
-     "text": [
-      "dict_keys(['SuperSuperLinear1.v_a', 'SuperLinear3.v1', 'SuperLinear4.v1', 'Linear5.W', 'Linear5.b', 'Linear6.W', 'Linear6.b'])\n",
-      "dict_keys(['SuperSuperLinear1', 'SuperLinear3', 'SuperLinear4', 'Linear5', 'Linear6'])\n"
-     ]
-    }
-   ],
-   "source": [
-    "# with register\n",
-    "ssl = SuperSuperLinear(register=True)\n",
-    "print(ssl.vars().keys())\n",
-    "print(ssl.nodes().keys())"
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "metadata": {},
-   "source": [
-    "## Transform a function to `BrainPyObject`"
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "metadata": {},
-   "source": [
-    "![](./imgs/loss_with_net_and_rng.png)"
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "metadata": {},
-   "source": [
-    "Let's go back to our network training.\n",
-    "After the definition of `net`, we further define a ``loss`` function whose computation involves the ``net`` object for neural network prediction and a ``rng`` Variable for data shuffling. \n",
-    "\n",
-    "This Python function is then transformed into a ``BrainPyObject`` instance by ``brainpy.math.to_object`` interface. "
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": 16,
-   "metadata": {},
-   "outputs": [
-    {
-     "data": {
-      "text/plain": [
-       "FunAsObject(nodes=[Sequential0],\n",
-       "            num_of_vars=1)"
-      ]
-     },
-     "execution_count": 16,
-     "metadata": {},
-     "output_type": "execute_result"
-    }
-   ],
-   "source": [
-    "loss"
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "metadata": {},
-   "source": [
-    "All `Variable` used in this instance can also be retrieved through:"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": 17,
-   "metadata": {},
-   "outputs": [
-    {
-     "data": {
-      "text/plain": [
-       "dict_keys(['loss0._var0', 'Linear0.W', 'Linear0.b', 'Linear1.W', 'Linear1.b'])"
-      ]
-     },
-     "execution_count": 17,
-     "metadata": {},
-     "output_type": "execute_result"
-    }
-   ],
-   "source": [
-    "loss.vars().keys()"
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "metadata": {},
-   "source": [
-    "Note that, when using `to_object()`, we need to explicitly declare all `BrainPyObject` and `Variable` used in this Python function. \n",
-    "Due to the recursive retrieval property of `BrainPyObject`, we only need to specify the latest composition object.\n",
-    "\n",
-    "In the above `loss` object, we do not need to specify two ``Linear`` object. Instead, we only need to give the top level object ``net`` into ``to_object()`` transform. \n",
-    "\n",
-    "Similarly, when we transform ``train`` function into a ``BrainPyObject``, we just need to point out the ``grad`` and ``opt`` we have used, rather than the previous *loss*, *net* or *rng*. "
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "metadata": {},
-   "source": [
-    "![](./imgs/train_with_grad_and_opt.png)"
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "metadata": {},
-   "source": [
-    "## BrainPy object-oriented transformations"
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "metadata": {},
-   "source": [
-    "BrainPy object-oriented transformations are designed to work on ``BrainPyObject``. \n",
-    "These transformations include autograd ``brainpy.math.grad()`` and JIT ``brainpy.math.jit()``."
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "metadata": {},
-   "source": [
-    "In our case, we used two OO transformations provided in BrainPy. \n",
-    "\n",
-    "First, ``grad`` object is defined with the ``loss`` function. Within it, we need to specify what variables we need to compute their gradients through `grad_vars`. "
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "metadata": {},
-   "source": [
-    "Note that, the OO transformation of any ``BrainPyObject`` results in another ``BrainPyObject`` object. Therefore, it can be recersively used as a component to form the larger scope of object-oriented programming and object-oriented transformation. "
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": 18,
-   "metadata": {},
-   "outputs": [
-    {
-     "data": {
-      "text/plain": [
-       "GradientTransform(target=loss0, \n",
-       "                  num_of_grad_vars=4, \n",
-       "                  num_of_dyn_vars=1)"
-      ]
-     },
-     "execution_count": 18,
-     "metadata": {},
-     "output_type": "execute_result"
-    }
-   ],
-   "source": [
-    "grad"
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "metadata": {},
-   "source": [
-    "![](./imgs/grad_with_loss.png)"
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "metadata": {},
-   "source": [
-    "Next, we train 400 steps once by using a ``for_loop`` transformation. Different from ``grad`` which return a `BrainPyObject` instance, `for_loop` direactly returns the loop results. "
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "metadata": {},
-   "source": [
-    "![](./imgs/for-loop-train.png)"
-   ]
-  }
- ],
- "metadata": {
-  "kernelspec": {
-   "display_name": "BrainPy",
-   "language": "python",
-   "name": "brainpy"
-  },
-  "language_info": {
-   "codemirror_mode": {
-    "name": "ipython",
-    "version": 3
-   },
-   "file_extension": ".py",
-   "mimetype": "text/x-python",
-   "name": "python",
-   "nbconvert_exporter": "python",
-   "pygments_lexer": "ipython3",
-   "version": "3.6.6"
-  },
-  "latex_envs": {
-   "LaTeX_envs_menu_present": true,
-   "autoclose": false,
-   "autocomplete": true,
-   "bibliofile": "biblio.bib",
-   "cite_by": "apalike",
-   "current_citInitial": 1,
-   "eqLabelWithNumbers": true,
-   "eqNumInitial": 1,
-   "hotkeys": {
-    "equation": "Ctrl-E",
-    "itemize": "Ctrl-I"
-   },
-   "labels_anchors": false,
-   "latex_user_defs": false,
-   "report_style_numbering": false,
-   "user_envs_cfg": false
-  },
-  "toc": {
-   "base_numbering": 1,
-   "nav_menu": {},
-   "number_sections": false,
-   "sideBar": true,
-   "skip_h1_title": false,
-   "title_cell": "Table of Contents",
-   "title_sidebar": "Contents",
-   "toc_cell": false,
-   "toc_position": {
-    "height": "calc(100% - 180px)",
-    "left": "10px",
-    "top": "150px",
-    "width": "245.75px"
-   },
-   "toc_section_display": true,
-   "toc_window_display": true
-  }
- },
- "nbformat": 4,
- "nbformat_minor": 4
-}
diff --git a/docs/core_concept/imgs/for-loop-train.png b/docs/core_concept/imgs/for-loop-train.png
deleted file mode 100644
index 5c380e5a7..000000000
Binary files a/docs/core_concept/imgs/for-loop-train.png and /dev/null differ
diff --git a/docs/core_concept/imgs/grad_with_loss.png b/docs/core_concept/imgs/grad_with_loss.png
deleted file mode 100644
index 64e7d6ab9..000000000
Binary files a/docs/core_concept/imgs/grad_with_loss.png and /dev/null differ
diff --git a/docs/core_concept/imgs/loss_with_net_and_rng.png b/docs/core_concept/imgs/loss_with_net_and_rng.png
deleted file mode 100644
index 94e4b2af5..000000000
Binary files a/docs/core_concept/imgs/loss_with_net_and_rng.png and /dev/null differ
diff --git a/docs/core_concept/imgs/train_with_grad_and_opt.png b/docs/core_concept/imgs/train_with_grad_and_opt.png
deleted file mode 100644
index e5ff0ca30..000000000
Binary files a/docs/core_concept/imgs/train_with_grad_and_opt.png and /dev/null differ
diff --git a/docs/index.rst b/docs/index.rst
index 1cf3db2f3..583a30e08 100644
--- a/docs/index.rst
+++ b/docs/index.rst
@@ -195,8 +195,8 @@ Learn more
 
    core_concepts.rst
    tutorials.rst
-   advanced_tutorials.rst
    toolboxes.rst
+   advanced_tutorials.rst
    FAQ.rst
    api.rst
 
diff --git a/docs/tutorial_FAQs/how_to_debug.ipynb b/docs/tutorial_FAQs/how_to_debug.ipynb
index 8b51d0bc4..a6f124288 100644
--- a/docs/tutorial_FAQs/how_to_debug.ipynb
+++ b/docs/tutorial_FAQs/how_to_debug.ipynb
@@ -154,12 +154,11 @@
   {
    "cell_type": "markdown",
    "source": [
-    "``jax.disable_jit()`` works for all brainpy transformations, for example:\n",
+    "``jax.disable_jit()`` works for most brainpy transformations, including:\n",
     "\n",
     "- ``brainpy.math.jit()``\n",
     "- ``brainpy.math.grad()``\n",
     "- ``brainpy.math.vector_grad()``\n",
-    "- ``brainpy.math.for_loop()``\n",
     "- ``brainpy.math.while_loop()``\n",
     "- ``brainpy.math.cond()``\n",
     "- ``brainpy.math.ifelse()``"
@@ -167,6 +166,28 @@
    "metadata": {
     "collapsed": false
    }
+  },
+  {
+   "cell_type": "markdown",
+   "source": [
+    "## ``brainpy.DSRunner(..., jit=False)``\n",
+    "\n",
+    "If users are using ``brainpy.DSRunner``, you can initialize ``brainpy.DSRunner(..., jit=False)`` to disable JIT compilation when simulating a brain dynamics model.\n"
+   ],
+   "metadata": {
+    "collapsed": false
+   }
+  },
+  {
+   "cell_type": "markdown",
+   "source": [
+    "## ``brainpy.for_loop(..., jit=False)``\n",
+    "\n",
+    "Similarly, if users are using ``brainpy.for_loop``, you can put a ``jit=False`` argument into the ``for_loop`` transformation, then the JIT compilation will be removed."
+   ],
+   "metadata": {
+    "collapsed": false
+   }
   }
  ],
  "metadata": {