{
"cells": [
{
"cell_type": "markdown",
"metadata": {},
"source": [
"# 🧠⇌ 💾 Import / export between toolchains with NIR\n",
"\n",
"**BETA INFORMATION. SUBJECT TO API CHANGE.**\n",
"\n",
"Neuromorphic Intermediate Representation (NIR for short) is a new open-source package for interoperability between the new crop of SNN simulators. Rockpool supports import and export of (`torch`-backed) models using NIR.\n",
"\n",
"![](\"https://github.com/neuromorphs/NIR/raw/main/docs/logo_light.png\")
"
]
},
{
"cell_type": "raw",
"metadata": {
"raw_mimetype": "text/restructuredtext",
"vscode": {
"languageId": "raw"
}
},
"source": [
"These are supported using the :py:func:`.nn.modules.to_nir` and :py:func:`.nn.modules.from_nir` functions from the :py:mod:`rockpool.nn.modules.torch` package.\n",
"\n",
"Using NIR requires the ``nir``, ``nirtorch`` and ``torch`` dependencies.\n",
"These can be installed alongside Rockpool with::\n",
"\n",
"\n",
" pip install 'rockpool[nir]'\n"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"This notebook shows you how to interact with NIR.\n",
"Let's start by generating a simple network.\n",
"\n",
"## Export from Rockpool to NIR"
]
},
{
"cell_type": "code",
"execution_count": 1,
"metadata": {},
"outputs": [
{
"data": {
"text/html": [
"TorchSequential with shape (1, 4) {\n",
" LinearTorch '0_LinearTorch' with shape (1, 2)\n",
" LIFTorch '1_LIFTorch' with shape (2, 2)\n",
" LinearTorch '2_LinearTorch' with shape (2, 4)\n",
" LIFTorch '3_LIFTorch' with shape (4, 4)\n",
"}\n",
"\n"
],
"text/plain": [
"TorchSequential with shape \u001b[1m(\u001b[0m\u001b[1;36m1\u001b[0m, \u001b[1;36m4\u001b[0m\u001b[1m)\u001b[0m \u001b[1m{\u001b[0m\n",
" LinearTorch \u001b[32m'0_LinearTorch'\u001b[0m with shape \u001b[1m(\u001b[0m\u001b[1;36m1\u001b[0m, \u001b[1;36m2\u001b[0m\u001b[1m)\u001b[0m\n",
" LIFTorch \u001b[32m'1_LIFTorch'\u001b[0m with shape \u001b[1m(\u001b[0m\u001b[1;36m2\u001b[0m, \u001b[1;36m2\u001b[0m\u001b[1m)\u001b[0m\n",
" LinearTorch \u001b[32m'2_LinearTorch'\u001b[0m with shape \u001b[1m(\u001b[0m\u001b[1;36m2\u001b[0m, \u001b[1;36m4\u001b[0m\u001b[1m)\u001b[0m\n",
" LIFTorch \u001b[32m'3_LIFTorch'\u001b[0m with shape \u001b[1m(\u001b[0m\u001b[1;36m4\u001b[0m, \u001b[1;36m4\u001b[0m\u001b[1m)\u001b[0m\n",
"\u001b[1m}\u001b[0m\n"
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"# - Import modules to build a simple network\n",
"from rockpool.nn.modules import LIFTorch, LinearTorch\n",
"from rockpool.nn.combinators import Sequential\n",
"\n",
"try:\n",
" from rich import print\n",
"except:\n",
" pass\n",
"\n",
"net = Sequential(\n",
" LinearTorch((1, 2)),\n",
" LIFTorch(2),\n",
" LinearTorch((2, 4)),\n",
" LIFTorch(4),\n",
")\n",
"print(net)"
]
},
{
"cell_type": "raw",
"metadata": {
"raw_mimetype": "text/restructuredtext",
"vscode": {
"languageId": "raw"
}
},
"source": [
"To handle import and export, Rockpool provides the helper functions :py:func:`.from_nir` and :py:func:`.to_nir`.\n",
"Let's convert our Rockpool network to a NIR representation, and write it out to disk."
]
},
{
"cell_type": "code",
"execution_count": 2,
"metadata": {},
"outputs": [
{
"data": {
"text/html": [
"NIRGraph(\n",
" nodes={\n",
" 'input': Input(input_type={'input': array([1, 1])}),\n",
" '0_LinearTorch': Linear(weight=tensor([[-1.5277],\n",
" [-2.2684]])),\n",
" '1_LIFTorch': CubaLIF(\n",
" tau_syn=array([0.02, 0.02], dtype=float32),\n",
" tau_mem=array([0.02, 0.02], dtype=float32),\n",
" r=array([19.024588, 19.024588], dtype=float32),\n",
" v_leak=array([0., 0.], dtype=float32),\n",
" v_threshold=array([1., 1.], dtype=float32),\n",
" w_in=array([1., 1.], dtype=float32),\n",
" input_type={'input': array([2])},\n",
" output_type={'output': array([2])},\n",
" metadata={}\n",
" ),\n",
" '2_LinearTorch': Linear(\n",
" weight=tensor([[-0.4406, 0.1872],\n",
" [ 0.2237, -1.0186],\n",
" [ 0.2433, 0.9899],\n",
" [-1.5922, 0.0632]])\n",
" ),\n",
" '3_LIFTorch': CubaLIF(\n",
" tau_syn=array([0.02, 0.02, 0.02, 0.02], dtype=float32),\n",
" tau_mem=array([0.02, 0.02, 0.02, 0.02], dtype=float32),\n",
" r=array([19.024588, 19.024588, 19.024588, 19.024588], dtype=float32),\n",
" v_leak=array([0., 0., 0., 0.], dtype=float32),\n",
" v_threshold=array([1., 1., 1., 1.], dtype=float32),\n",
" w_in=array([1., 1., 1., 1.], dtype=float32),\n",
" input_type={'input': array([4])},\n",
" output_type={'output': array([4])},\n",
" metadata={}\n",
" ),\n",
" 'output': Output(output_type={'output': array([1, 1, 4])})\n",
" },\n",
" edges=[\n",
" ('input', '0_LinearTorch'),\n",
" ('0_LinearTorch', '1_LIFTorch'),\n",
" ('3_LIFTorch', 'output'),\n",
" ('2_LinearTorch', '3_LIFTorch'),\n",
" ('1_LIFTorch', '2_LinearTorch')\n",
" ],\n",
" input_type={'input': {'input': array([1, 1])}},\n",
" output_type={'output': {'output': array([1, 1, 4])}},\n",
" metadata={}\n",
")\n",
"\n"
],
"text/plain": [
"\u001b[1;35mNIRGraph\u001b[0m\u001b[1m(\u001b[0m\n",
" \u001b[33mnodes\u001b[0m=\u001b[1m{\u001b[0m\n",
" \u001b[32m'input'\u001b[0m: \u001b[1;35mInput\u001b[0m\u001b[1m(\u001b[0m\u001b[33minput_type\u001b[0m=\u001b[1m{\u001b[0m\u001b[32m'input'\u001b[0m: \u001b[1;35marray\u001b[0m\u001b[1m(\u001b[0m\u001b[1m[\u001b[0m\u001b[1;36m1\u001b[0m, \u001b[1;36m1\u001b[0m\u001b[1m]\u001b[0m\u001b[1m)\u001b[0m\u001b[1m}\u001b[0m\u001b[1m)\u001b[0m,\n",
" \u001b[32m'0_LinearTorch'\u001b[0m: \u001b[1;35mLinear\u001b[0m\u001b[1m(\u001b[0m\u001b[33mweight\u001b[0m=\u001b[1;35mtensor\u001b[0m\u001b[1m(\u001b[0m\u001b[1m[\u001b[0m\u001b[1m[\u001b[0m\u001b[1;36m-1.5277\u001b[0m\u001b[1m]\u001b[0m,\n",
" \u001b[1m[\u001b[0m\u001b[1;36m-2.2684\u001b[0m\u001b[1m]\u001b[0m\u001b[1m]\u001b[0m\u001b[1m)\u001b[0m\u001b[1m)\u001b[0m,\n",
" \u001b[32m'1_LIFTorch'\u001b[0m: \u001b[1;35mCubaLIF\u001b[0m\u001b[1m(\u001b[0m\n",
" \u001b[33mtau_syn\u001b[0m=\u001b[1;35marray\u001b[0m\u001b[1m(\u001b[0m\u001b[1m[\u001b[0m\u001b[1;36m0.02\u001b[0m, \u001b[1;36m0.02\u001b[0m\u001b[1m]\u001b[0m, \u001b[33mdtype\u001b[0m=\u001b[35mfloat32\u001b[0m\u001b[1m)\u001b[0m,\n",
" \u001b[33mtau_mem\u001b[0m=\u001b[1;35marray\u001b[0m\u001b[1m(\u001b[0m\u001b[1m[\u001b[0m\u001b[1;36m0.02\u001b[0m, \u001b[1;36m0.02\u001b[0m\u001b[1m]\u001b[0m, \u001b[33mdtype\u001b[0m=\u001b[35mfloat32\u001b[0m\u001b[1m)\u001b[0m,\n",
" \u001b[33mr\u001b[0m=\u001b[1;35marray\u001b[0m\u001b[1m(\u001b[0m\u001b[1m[\u001b[0m\u001b[1;36m19.024588\u001b[0m, \u001b[1;36m19.024588\u001b[0m\u001b[1m]\u001b[0m, \u001b[33mdtype\u001b[0m=\u001b[35mfloat32\u001b[0m\u001b[1m)\u001b[0m,\n",
" \u001b[33mv_leak\u001b[0m=\u001b[1;35marray\u001b[0m\u001b[1m(\u001b[0m\u001b[1m[\u001b[0m\u001b[1;36m0\u001b[0m., \u001b[1;36m0\u001b[0m.\u001b[1m]\u001b[0m, \u001b[33mdtype\u001b[0m=\u001b[35mfloat32\u001b[0m\u001b[1m)\u001b[0m,\n",
" \u001b[33mv_threshold\u001b[0m=\u001b[1;35marray\u001b[0m\u001b[1m(\u001b[0m\u001b[1m[\u001b[0m\u001b[1;36m1\u001b[0m., \u001b[1;36m1\u001b[0m.\u001b[1m]\u001b[0m, \u001b[33mdtype\u001b[0m=\u001b[35mfloat32\u001b[0m\u001b[1m)\u001b[0m,\n",
" \u001b[33mw_in\u001b[0m=\u001b[1;35marray\u001b[0m\u001b[1m(\u001b[0m\u001b[1m[\u001b[0m\u001b[1;36m1\u001b[0m., \u001b[1;36m1\u001b[0m.\u001b[1m]\u001b[0m, \u001b[33mdtype\u001b[0m=\u001b[35mfloat32\u001b[0m\u001b[1m)\u001b[0m,\n",
" \u001b[33minput_type\u001b[0m=\u001b[1m{\u001b[0m\u001b[32m'input'\u001b[0m: \u001b[1;35marray\u001b[0m\u001b[1m(\u001b[0m\u001b[1m[\u001b[0m\u001b[1;36m2\u001b[0m\u001b[1m]\u001b[0m\u001b[1m)\u001b[0m\u001b[1m}\u001b[0m,\n",
" \u001b[33moutput_type\u001b[0m=\u001b[1m{\u001b[0m\u001b[32m'output'\u001b[0m: \u001b[1;35marray\u001b[0m\u001b[1m(\u001b[0m\u001b[1m[\u001b[0m\u001b[1;36m2\u001b[0m\u001b[1m]\u001b[0m\u001b[1m)\u001b[0m\u001b[1m}\u001b[0m,\n",
" \u001b[33mmetadata\u001b[0m=\u001b[1m{\u001b[0m\u001b[1m}\u001b[0m\n",
" \u001b[1m)\u001b[0m,\n",
" \u001b[32m'2_LinearTorch'\u001b[0m: \u001b[1;35mLinear\u001b[0m\u001b[1m(\u001b[0m\n",
" \u001b[33mweight\u001b[0m=\u001b[1;35mtensor\u001b[0m\u001b[1m(\u001b[0m\u001b[1m[\u001b[0m\u001b[1m[\u001b[0m\u001b[1;36m-0.4406\u001b[0m, \u001b[1;36m0.1872\u001b[0m\u001b[1m]\u001b[0m,\n",
" \u001b[1m[\u001b[0m \u001b[1;36m0.2237\u001b[0m, \u001b[1;36m-1.0186\u001b[0m\u001b[1m]\u001b[0m,\n",
" \u001b[1m[\u001b[0m \u001b[1;36m0.2433\u001b[0m, \u001b[1;36m0.9899\u001b[0m\u001b[1m]\u001b[0m,\n",
" \u001b[1m[\u001b[0m\u001b[1;36m-1.5922\u001b[0m, \u001b[1;36m0.0632\u001b[0m\u001b[1m]\u001b[0m\u001b[1m]\u001b[0m\u001b[1m)\u001b[0m\n",
" \u001b[1m)\u001b[0m,\n",
" \u001b[32m'3_LIFTorch'\u001b[0m: \u001b[1;35mCubaLIF\u001b[0m\u001b[1m(\u001b[0m\n",
" \u001b[33mtau_syn\u001b[0m=\u001b[1;35marray\u001b[0m\u001b[1m(\u001b[0m\u001b[1m[\u001b[0m\u001b[1;36m0.02\u001b[0m, \u001b[1;36m0.02\u001b[0m, \u001b[1;36m0.02\u001b[0m, \u001b[1;36m0.02\u001b[0m\u001b[1m]\u001b[0m, \u001b[33mdtype\u001b[0m=\u001b[35mfloat32\u001b[0m\u001b[1m)\u001b[0m,\n",
" \u001b[33mtau_mem\u001b[0m=\u001b[1;35marray\u001b[0m\u001b[1m(\u001b[0m\u001b[1m[\u001b[0m\u001b[1;36m0.02\u001b[0m, \u001b[1;36m0.02\u001b[0m, \u001b[1;36m0.02\u001b[0m, \u001b[1;36m0.02\u001b[0m\u001b[1m]\u001b[0m, \u001b[33mdtype\u001b[0m=\u001b[35mfloat32\u001b[0m\u001b[1m)\u001b[0m,\n",
" \u001b[33mr\u001b[0m=\u001b[1;35marray\u001b[0m\u001b[1m(\u001b[0m\u001b[1m[\u001b[0m\u001b[1;36m19.024588\u001b[0m, \u001b[1;36m19.024588\u001b[0m, \u001b[1;36m19.024588\u001b[0m, \u001b[1;36m19.024588\u001b[0m\u001b[1m]\u001b[0m, \u001b[33mdtype\u001b[0m=\u001b[35mfloat32\u001b[0m\u001b[1m)\u001b[0m,\n",
" \u001b[33mv_leak\u001b[0m=\u001b[1;35marray\u001b[0m\u001b[1m(\u001b[0m\u001b[1m[\u001b[0m\u001b[1;36m0\u001b[0m., \u001b[1;36m0\u001b[0m., \u001b[1;36m0\u001b[0m., \u001b[1;36m0\u001b[0m.\u001b[1m]\u001b[0m, \u001b[33mdtype\u001b[0m=\u001b[35mfloat32\u001b[0m\u001b[1m)\u001b[0m,\n",
" \u001b[33mv_threshold\u001b[0m=\u001b[1;35marray\u001b[0m\u001b[1m(\u001b[0m\u001b[1m[\u001b[0m\u001b[1;36m1\u001b[0m., \u001b[1;36m1\u001b[0m., \u001b[1;36m1\u001b[0m., \u001b[1;36m1\u001b[0m.\u001b[1m]\u001b[0m, \u001b[33mdtype\u001b[0m=\u001b[35mfloat32\u001b[0m\u001b[1m)\u001b[0m,\n",
" \u001b[33mw_in\u001b[0m=\u001b[1;35marray\u001b[0m\u001b[1m(\u001b[0m\u001b[1m[\u001b[0m\u001b[1;36m1\u001b[0m., \u001b[1;36m1\u001b[0m., \u001b[1;36m1\u001b[0m., \u001b[1;36m1\u001b[0m.\u001b[1m]\u001b[0m, \u001b[33mdtype\u001b[0m=\u001b[35mfloat32\u001b[0m\u001b[1m)\u001b[0m,\n",
" \u001b[33minput_type\u001b[0m=\u001b[1m{\u001b[0m\u001b[32m'input'\u001b[0m: \u001b[1;35marray\u001b[0m\u001b[1m(\u001b[0m\u001b[1m[\u001b[0m\u001b[1;36m4\u001b[0m\u001b[1m]\u001b[0m\u001b[1m)\u001b[0m\u001b[1m}\u001b[0m,\n",
" \u001b[33moutput_type\u001b[0m=\u001b[1m{\u001b[0m\u001b[32m'output'\u001b[0m: \u001b[1;35marray\u001b[0m\u001b[1m(\u001b[0m\u001b[1m[\u001b[0m\u001b[1;36m4\u001b[0m\u001b[1m]\u001b[0m\u001b[1m)\u001b[0m\u001b[1m}\u001b[0m,\n",
" \u001b[33mmetadata\u001b[0m=\u001b[1m{\u001b[0m\u001b[1m}\u001b[0m\n",
" \u001b[1m)\u001b[0m,\n",
" \u001b[32m'output'\u001b[0m: \u001b[1;35mOutput\u001b[0m\u001b[1m(\u001b[0m\u001b[33moutput_type\u001b[0m=\u001b[1m{\u001b[0m\u001b[32m'output'\u001b[0m: \u001b[1;35marray\u001b[0m\u001b[1m(\u001b[0m\u001b[1m[\u001b[0m\u001b[1;36m1\u001b[0m, \u001b[1;36m1\u001b[0m, \u001b[1;36m4\u001b[0m\u001b[1m]\u001b[0m\u001b[1m)\u001b[0m\u001b[1m}\u001b[0m\u001b[1m)\u001b[0m\n",
" \u001b[1m}\u001b[0m,\n",
" \u001b[33medges\u001b[0m=\u001b[1m[\u001b[0m\n",
" \u001b[1m(\u001b[0m\u001b[32m'input'\u001b[0m, \u001b[32m'0_LinearTorch'\u001b[0m\u001b[1m)\u001b[0m,\n",
" \u001b[1m(\u001b[0m\u001b[32m'0_LinearTorch'\u001b[0m, \u001b[32m'1_LIFTorch'\u001b[0m\u001b[1m)\u001b[0m,\n",
" \u001b[1m(\u001b[0m\u001b[32m'3_LIFTorch'\u001b[0m, \u001b[32m'output'\u001b[0m\u001b[1m)\u001b[0m,\n",
" \u001b[1m(\u001b[0m\u001b[32m'2_LinearTorch'\u001b[0m, \u001b[32m'3_LIFTorch'\u001b[0m\u001b[1m)\u001b[0m,\n",
" \u001b[1m(\u001b[0m\u001b[32m'1_LIFTorch'\u001b[0m, \u001b[32m'2_LinearTorch'\u001b[0m\u001b[1m)\u001b[0m\n",
" \u001b[1m]\u001b[0m,\n",
" \u001b[33minput_type\u001b[0m=\u001b[1m{\u001b[0m\u001b[32m'input'\u001b[0m: \u001b[1m{\u001b[0m\u001b[32m'input'\u001b[0m: \u001b[1;35marray\u001b[0m\u001b[1m(\u001b[0m\u001b[1m[\u001b[0m\u001b[1;36m1\u001b[0m, \u001b[1;36m1\u001b[0m\u001b[1m]\u001b[0m\u001b[1m)\u001b[0m\u001b[1m}\u001b[0m\u001b[1m}\u001b[0m,\n",
" \u001b[33moutput_type\u001b[0m=\u001b[1m{\u001b[0m\u001b[32m'output'\u001b[0m: \u001b[1m{\u001b[0m\u001b[32m'output'\u001b[0m: \u001b[1;35marray\u001b[0m\u001b[1m(\u001b[0m\u001b[1m[\u001b[0m\u001b[1;36m1\u001b[0m, \u001b[1;36m1\u001b[0m, \u001b[1;36m4\u001b[0m\u001b[1m]\u001b[0m\u001b[1m)\u001b[0m\u001b[1m}\u001b[0m\u001b[1m}\u001b[0m,\n",
" \u001b[33mmetadata\u001b[0m=\u001b[1m{\u001b[0m\u001b[1m}\u001b[0m\n",
"\u001b[1m)\u001b[0m\n"
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"# - Import utilities to import and export NIR graphs\n",
"from rockpool.nn.modules import from_nir, to_nir\n",
"import nir\n",
"import torch\n",
"\n",
"# - Convert the Rockpool network to the NIR representation\n",
"nir_graph = to_nir(net)\n",
"print(nir_graph)\n",
"\n",
"# - Write the exported graph to disk\n",
"nir.write('rockpool.nir', nir_graph)"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Import from NIR to Rockpool and deploy"
]
},
{
"cell_type": "raw",
"metadata": {
"raw_mimetype": "text/restructuredtext",
"vscode": {
"languageId": "raw"
}
},
"source": [
"Importing from NIR to Rockpool requires the :py:func:`.from_nir` helper function, along with ``nir.read``."
]
},
{
"cell_type": "code",
"execution_count": 3,
"metadata": {},
"outputs": [
{
"data": {
"text/html": [
"GraphExecutor(\n",
" (0_LinearTorch): LinearTorch()\n",
" (1_LIFTorch): LIFTorch()\n",
" (2_LinearTorch): LinearTorch()\n",
" (3_LIFTorch): LIFTorch()\n",
" (input): Identity()\n",
" (output): Identity()\n",
")\n",
"\n"
],
"text/plain": [
"\u001b[1;35mGraphExecutor\u001b[0m\u001b[1m(\u001b[0m\n",
" \u001b[1m(\u001b[0m0_LinearTorch\u001b[1m)\u001b[0m: \u001b[1;35mLinearTorch\u001b[0m\u001b[1m(\u001b[0m\u001b[1m)\u001b[0m\n",
" \u001b[1m(\u001b[0m1_LIFTorch\u001b[1m)\u001b[0m: \u001b[1;35mLIFTorch\u001b[0m\u001b[1m(\u001b[0m\u001b[1m)\u001b[0m\n",
" \u001b[1m(\u001b[0m2_LinearTorch\u001b[1m)\u001b[0m: \u001b[1;35mLinearTorch\u001b[0m\u001b[1m(\u001b[0m\u001b[1m)\u001b[0m\n",
" \u001b[1m(\u001b[0m3_LIFTorch\u001b[1m)\u001b[0m: \u001b[1;35mLIFTorch\u001b[0m\u001b[1m(\u001b[0m\u001b[1m)\u001b[0m\n",
" \u001b[1m(\u001b[0minput\u001b[1m)\u001b[0m: \u001b[1;35mIdentity\u001b[0m\u001b[1m(\u001b[0m\u001b[1m)\u001b[0m\n",
" \u001b[1m(\u001b[0moutput\u001b[1m)\u001b[0m: \u001b[1;35mIdentity\u001b[0m\u001b[1m(\u001b[0m\u001b[1m)\u001b[0m\n",
"\u001b[1m)\u001b[0m\n"
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"# - Read a NIR file from disk\n",
"nir_graph = nir.read('rockpool.nir')\n",
"\n",
"# - Use `from_nir` to convert to Rockpool\n",
"torch_net = from_nir(nir_graph)\n",
"\n",
"# - Alternatively, load and convert directly\n",
"torch_net = from_nir('rockpool.nir')\n",
"print(torch_net)"
]
},
{
"cell_type": "raw",
"metadata": {
"raw_mimetype": "text/restructuredtext",
"vscode": {
"languageId": "raw"
}
},
"source": [
"Note that ``torch_net`` is a ``torch.nn.Module``, and not a Rockpool-native module.\n",
"This is a limitation caused by relying on the ``nirtorch`` package for NIR import.\n",
"Nevertheless, this object contains Rockpool modules :py:class:`.LinearTorch` and :py:class:`.LIFTorch`, and can be simulated.\n",
"Since it is a ``nirtorch.GraphExecutor`` module, it returns a tuple ``(output, state)``."
]
},
{
"cell_type": "code",
"execution_count": 4,
"metadata": {},
"outputs": [],
"source": [
"# - Simulate the module\n",
"output, state = torch_net(torch.rand(1, 100, 1))"
]
},
{
"cell_type": "raw",
"metadata": {
"raw_mimetype": "text/restructuredtext",
"vscode": {
"languageId": "raw"
}
},
"source": [
"It also supports the Rockpool :py:meth:`~.Module.as_graph` method, required for deployment."
]
},
{
"cell_type": "code",
"execution_count": 5,
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"XyloSim with shape (16, 1000, 8)"
]
},
"execution_count": 5,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"from rockpool.devices.xylo.syns61201 import mapper, config_from_specification, XyloSim\n",
"from rockpool.transform.quantize_methods import channel_quantize\n",
"\n",
"# - Map, quantize and get a Xylo HW configuration\n",
"spec = mapper(torch_net.as_graph())\n",
"config, is_valid, msg = config_from_specification(**channel_quantize(**spec))\n",
"\n",
"# - Produce a Xylo simulation matching the NIR graph\n",
"XyloSim.from_config(config)"
]
},
{
"cell_type": "markdown",
"metadata": {
"vscode": {
"languageId": "raw"
}
},
"source": [
"## See also\n",
"For more information about NIR, as well as how NIR integrates with other libraries, see https://neuroir.org."
]
}
],
"metadata": {
"kernelspec": {
"display_name": "py38",
"language": "python",
"name": "python3"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 3
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.10.14"
}
},
"nbformat": 4,
"nbformat_minor": 2
}