iicd/xautodl
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Update xmisc.scheduler/sampler

Browse Source
This commit is contained in:
D-X-Y
2021-06-11 11:46:18 +08:00
parent 9bf0fa5f04
commit 48163c792c
17 changed files with 807 additions and 201 deletions
Download Patch File Download Diff File Expand all files Collapse all files

102
notebooks/spaces-xmisc/random-search-transformer.ipynb Normal file
View File

@@ -0,0 +1,102 @@
{
"cells": [
{
"cell_type": "code",
"execution_count": 1,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"library path: /Users/xuanyidong/Desktop/XAutoDL/lib\n"
]
}
],
"source": [
"#####################################################\n",
"# Copyright (c) Xuanyi Dong [GitHub D-X-Y], 2021.03 #\n",
"#####################################################\n",
"import abc, os, sys\n",
"from pathlib import Path\n",
"\n",
"__file__ = os.path.dirname(os.path.realpath(\"__file__\"))\n",
"\n",
"lib_dir = (Path(__file__).parent / \"..\" / \"lib\").resolve()\n",
"print(\"library path: {:}\".format(lib_dir))\n",
"assert lib_dir.exists(), \"{:} does not exist\".format(lib_dir)\n",
"if str(lib_dir) not in sys.path:\n",
" sys.path.insert(0, str(lib_dir))"
]
},
{
"cell_type": "code",
"execution_count": 2,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"1.7.0\n",
"True\n",
"OrderedDict()\n",
"OrderedDict()\n",
"set()\n",
"OrderedDict()\n",
"OrderedDict()\n",
"OrderedDict()\n",
"OrderedDict()\n",
"OrderedDict()\n",
"OrderedDict()\n"
]
},
{
"name": "stderr",
"output_type": "stream",
"text": [
"/Users/xuanyidong/anaconda3/lib/python3.8/site-packages/torch/nn/modules/container.py:551: UserWarning: Setting attributes on ParameterDict is not supported.\n",
" warnings.warn(\"Setting attributes on ParameterDict is not supported.\")\n"
]
}
],
"source": [
"# Test the Linear layer\n",
"import spaces\n",
"import torch\n",
"from xlayers import super_core\n",
"\n",
"print(torch.__version__)\n",
"mlp = super_core.SuperMLPv2(10, 12, 32)"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": []
}
],
"metadata": {
"kernelspec": {
"display_name": "Python 3",
"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": "3.8.3"
}
},
"nbformat": 4,
"nbformat_minor": 4
}

119
notebooks/spaces-xmisc/scheduler.ipynb Normal file
View File

File diff suppressed because one or more lines are too long

110
notebooks/spaces-xmisc/synthetic-data.ipynb Normal file
View File

File diff suppressed because one or more lines are too long

129
notebooks/spaces-xmisc/synthetic-env.ipynb Normal file
View File

File diff suppressed because one or more lines are too long

152
notebooks/spaces-xmisc/synthetic-visualize-env.ipynb Normal file
View File

@@ -0,0 +1,152 @@
{
"cells": [
{
"cell_type": "code",
"execution_count": 1,
"id": "filled-multiple",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"The root path: /Users/xuanyidong/Desktop/AutoDL-Projects\n",
"The library path: /Users/xuanyidong/Desktop/AutoDL-Projects/lib\n"
]
}
],
"source": [
"import os, sys\n",
"import torch\n",
"from pathlib import Path\n",
"import numpy as np\n",
"import matplotlib\n",
"from matplotlib import cm\n",
"matplotlib.use(\"agg\")\n",
"import matplotlib.pyplot as plt\n",
"import matplotlib.ticker as ticker\n",
"\n",
"\n",
"__file__ = os.path.dirname(os.path.realpath(\"__file__\"))\n",
"root_dir = (Path(__file__).parent / \"..\").resolve()\n",
"lib_dir = (root_dir / \"lib\").resolve()\n",
"print(\"The root path: {:}\".format(root_dir))\n",
"print(\"The library path: {:}\".format(lib_dir))\n",
"assert lib_dir.exists(), \"{:} does not exist\".format(lib_dir)\n",
"if str(lib_dir) not in sys.path:\n",
" sys.path.insert(0, str(lib_dir))\n",
"\n",
"from datasets import ConstantGenerator, SinGenerator, SyntheticDEnv\n",
"from datasets import DynamicQuadraticFunc\n",
"from datasets.synthetic_example import create_example_v1"
]
},
{
"cell_type": "code",
"execution_count": 2,
"id": "detected-second",
"metadata": {},
"outputs": [],
"source": [
"def draw_fig(save_dir, timestamp, xaxis, yaxis):\n",
" save_path = save_dir / '{:04d}'.format(timestamp)\n",
" # print('Plot the figure at timestamp-{:} into {:}'.format(timestamp, save_path))\n",
" dpi, width, height = 40, 1500, 1500\n",
" figsize = width / float(dpi), height / float(dpi)\n",
" LabelSize, LegendFontsize, font_gap = 80, 80, 5\n",
"\n",
" fig = plt.figure(figsize=figsize)\n",
" \n",
" cur_ax = fig.add_subplot(1, 1, 1)\n",
" cur_ax.scatter(xaxis, yaxis, color=\"k\", s=10, alpha=0.9, label=\"Timestamp={:02d}\".format(timestamp))\n",
" cur_ax.set_xlabel(\"X\", fontsize=LabelSize)\n",
" cur_ax.set_ylabel(\"f(X)\", rotation=0, fontsize=LabelSize)\n",
" cur_ax.set_xlim(-6, 6)\n",
" cur_ax.set_ylim(-40, 40)\n",
" for tick in cur_ax.xaxis.get_major_ticks():\n",
" tick.label.set_fontsize(LabelSize - font_gap)\n",
" tick.label.set_rotation(10)\n",
" for tick in cur_ax.yaxis.get_major_ticks():\n",
" tick.label.set_fontsize(LabelSize - font_gap)\n",
" \n",
" plt.legend(loc=1, fontsize=LegendFontsize)\n",
" fig.savefig(str(save_path) + '.pdf', dpi=dpi, bbox_inches=\"tight\", format=\"pdf\")\n",
" fig.savefig(str(save_path) + '.png', dpi=dpi, bbox_inches=\"tight\", format=\"png\")\n",
" plt.close(\"all\")\n",
"\n",
"\n",
"def visualize_env(save_dir):\n",
" save_dir.mkdir(parents=True, exist_ok=True)\n",
" dynamic_env, function = create_example_v1(100, num_per_task=500)\n",
" \n",
" additional_xaxis = np.arange(-6, 6, 0.1)\n",
" for timestamp, dataset in dynamic_env:\n",
" num = dataset.shape[0]\n",
" # timeaxis = (torch.zeros(num) + timestamp).numpy()\n",
" xaxis = dataset[:,0].numpy()\n",
" xaxis = np.concatenate((additional_xaxis, xaxis))\n",
" # compute the ground truth\n",
" function.set_timestamp(timestamp)\n",
" yaxis = function(xaxis)\n",
" draw_fig(save_dir, timestamp, xaxis, yaxis)\n",
"\n",
"home_dir = Path.home()\n",
"desktop_dir = home_dir / 'Desktop'\n",
"vis_save_dir = desktop_dir / 'vis-synthetic'\n",
"visualize_env(vis_save_dir)"
]
},
{
"cell_type": "code",
"execution_count": 3,
"id": "rapid-uruguay",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"ffmpeg -y -i /Users/xuanyidong/Desktop/vis-synthetic/%04d.png -pix_fmt yuv420p -vf fps=2 -vf scale=1000:1000 -vb 5000k /Users/xuanyidong/Desktop/vis-synthetic/vis.mp4\n"
]
},
{
"data": {
"text/plain": [
"0"
]
},
"execution_count": 3,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"# Plot the data\n",
"cmd = 'ffmpeg -y -i {:}/%04d.png -pix_fmt yuv420p -vf fps=2 -vf scale=1000:1000 -vb 5000k {:}/vis.mp4'.format(vis_save_dir, vis_save_dir)\n",
"print(cmd)\n",
"os.system(cmd)"
]
}
],
"metadata": {
"kernelspec": {
"display_name": "Python 3",
"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": "3.8.8"
}
},
"nbformat": 4,
"nbformat_minor": 5
}

277
notebooks/spaces-xmisc/test-transformer-encoder.ipynb Normal file
View File

@@ -0,0 +1,277 @@
{
"cells": [
{
"cell_type": "code",
"execution_count": 1,
"id": "3f754c96",
"metadata": {},
"outputs": [],
"source": [
"import torch\n",
"from xautodl import spaces\n",
"from xautodl.xlayers import super_core\n",
"\n",
"def _create_stel(input_dim, output_dim, order):\n",
" return super_core.SuperSequential(\n",
" super_core.SuperLinear(input_dim, output_dim),\n",
" super_core.SuperTransformerEncoderLayer(\n",
" output_dim,\n",
" num_heads=spaces.Categorical(2, 4, 6),\n",
" mlp_hidden_multiplier=spaces.Categorical(1, 2, 4),\n",
" order=order,\n",
" ),\n",
" )"
]
},
{
"cell_type": "code",
"execution_count": 2,
"id": "81d42f4b",
"metadata": {},
"outputs": [],
"source": [
"batch, seq_dim, input_dim = 1, 4, 6\n",
"order = super_core.LayerOrder.PreNorm"
]
},
{
"cell_type": "code",
"execution_count": 3,
"id": "8056b37c",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"SuperSequential(\n",
" (0): SuperSequential(\n",
" (0): SuperLinear(in_features=6, out_features=Categorical(candidates=[12, 24, 36], default_index=None), bias=True)\n",
" (1): SuperTransformerEncoderLayer(\n",
" (norm1): SuperLayerNorm1D(shape=Categorical(candidates=[12, 24, 36], default_index=None), eps=1e-06, elementwise_affine=True)\n",
" (mha): SuperSelfAttention(\n",
" input_dim=Categorical(candidates=[12, 24, 36], default_index=None), proj_dim=Categorical(candidates=[12, 24, 36], default_index=None), num_heads=Categorical(candidates=[2, 4, 6], default_index=None), mask=False, infinity=1000000000.0\n",
" (q_fc): SuperLinear(in_features=Categorical(candidates=[12, 24, 36], default_index=None), out_features=Categorical(candidates=[12, 24, 36], default_index=None), bias=False)\n",
" (k_fc): SuperLinear(in_features=Categorical(candidates=[12, 24, 36], default_index=None), out_features=Categorical(candidates=[12, 24, 36], default_index=None), bias=False)\n",
" (v_fc): SuperLinear(in_features=Categorical(candidates=[12, 24, 36], default_index=None), out_features=Categorical(candidates=[12, 24, 36], default_index=None), bias=False)\n",
" (attn_drop): SuperDrop(p=0.0, dims=[-1, -1, -1, -1], recover=True)\n",
" )\n",
" (drop): Dropout(p=0.0, inplace=False)\n",
" (norm2): SuperLayerNorm1D(shape=Categorical(candidates=[12, 24, 36], default_index=None), eps=1e-06, elementwise_affine=True)\n",
" (mlp): SuperMLPv2(\n",
" in_features=Categorical(candidates=[12, 24, 36], default_index=None), hidden_multiplier=Categorical(candidates=[1, 2, 4], default_index=None), out_features=Categorical(candidates=[12, 24, 36], default_index=None), drop=None, fc1 -> act -> drop -> fc2 -> drop,\n",
" (_params): ParameterDict(\n",
" (fc1_super_weight): Parameter containing: [torch.FloatTensor of size 144x36]\n",
" (fc1_super_bias): Parameter containing: [torch.FloatTensor of size 144]\n",
" (fc2_super_weight): Parameter containing: [torch.FloatTensor of size 36x144]\n",
" (fc2_super_bias): Parameter containing: [torch.FloatTensor of size 36]\n",
" )\n",
" (act): GELU()\n",
" (drop): Dropout(p=0.0, inplace=False)\n",
" )\n",
" )\n",
" )\n",
" (1): SuperSequential(\n",
" (0): SuperLinear(in_features=Categorical(candidates=[12, 24, 36], default_index=None), out_features=Categorical(candidates=[24, 36, 48], default_index=None), bias=True)\n",
" (1): SuperTransformerEncoderLayer(\n",
" (norm1): SuperLayerNorm1D(shape=Categorical(candidates=[24, 36, 48], default_index=None), eps=1e-06, elementwise_affine=True)\n",
" (mha): SuperSelfAttention(\n",
" input_dim=Categorical(candidates=[24, 36, 48], default_index=None), proj_dim=Categorical(candidates=[24, 36, 48], default_index=None), num_heads=Categorical(candidates=[2, 4, 6], default_index=None), mask=False, infinity=1000000000.0\n",
" (q_fc): SuperLinear(in_features=Categorical(candidates=[24, 36, 48], default_index=None), out_features=Categorical(candidates=[24, 36, 48], default_index=None), bias=False)\n",
" (k_fc): SuperLinear(in_features=Categorical(candidates=[24, 36, 48], default_index=None), out_features=Categorical(candidates=[24, 36, 48], default_index=None), bias=False)\n",
" (v_fc): SuperLinear(in_features=Categorical(candidates=[24, 36, 48], default_index=None), out_features=Categorical(candidates=[24, 36, 48], default_index=None), bias=False)\n",
" (attn_drop): SuperDrop(p=0.0, dims=[-1, -1, -1, -1], recover=True)\n",
" )\n",
" (drop): Dropout(p=0.0, inplace=False)\n",
" (norm2): SuperLayerNorm1D(shape=Categorical(candidates=[24, 36, 48], default_index=None), eps=1e-06, elementwise_affine=True)\n",
" (mlp): SuperMLPv2(\n",
" in_features=Categorical(candidates=[24, 36, 48], default_index=None), hidden_multiplier=Categorical(candidates=[1, 2, 4], default_index=None), out_features=Categorical(candidates=[24, 36, 48], default_index=None), drop=None, fc1 -> act -> drop -> fc2 -> drop,\n",
" (_params): ParameterDict(\n",
" (fc1_super_weight): Parameter containing: [torch.FloatTensor of size 192x48]\n",
" (fc1_super_bias): Parameter containing: [torch.FloatTensor of size 192]\n",
" (fc2_super_weight): Parameter containing: [torch.FloatTensor of size 48x192]\n",
" (fc2_super_bias): Parameter containing: [torch.FloatTensor of size 48]\n",
" )\n",
" (act): GELU()\n",
" (drop): Dropout(p=0.0, inplace=False)\n",
" )\n",
" )\n",
" )\n",
" (2): SuperSequential(\n",
" (0): SuperLinear(in_features=Categorical(candidates=[24, 36, 48], default_index=None), out_features=Categorical(candidates=[36, 72, 100], default_index=None), bias=True)\n",
" (1): SuperTransformerEncoderLayer(\n",
" (norm1): SuperLayerNorm1D(shape=Categorical(candidates=[36, 72, 100], default_index=None), eps=1e-06, elementwise_affine=True)\n",
" (mha): SuperSelfAttention(\n",
" input_dim=Categorical(candidates=[36, 72, 100], default_index=None), proj_dim=Categorical(candidates=[36, 72, 100], default_index=None), num_heads=Categorical(candidates=[2, 4, 6], default_index=None), mask=False, infinity=1000000000.0\n",
" (q_fc): SuperLinear(in_features=Categorical(candidates=[36, 72, 100], default_index=None), out_features=Categorical(candidates=[36, 72, 100], default_index=None), bias=False)\n",
" (k_fc): SuperLinear(in_features=Categorical(candidates=[36, 72, 100], default_index=None), out_features=Categorical(candidates=[36, 72, 100], default_index=None), bias=False)\n",
" (v_fc): SuperLinear(in_features=Categorical(candidates=[36, 72, 100], default_index=None), out_features=Categorical(candidates=[36, 72, 100], default_index=None), bias=False)\n",
" (attn_drop): SuperDrop(p=0.0, dims=[-1, -1, -1, -1], recover=True)\n",
" )\n",
" (drop): Dropout(p=0.0, inplace=False)\n",
" (norm2): SuperLayerNorm1D(shape=Categorical(candidates=[36, 72, 100], default_index=None), eps=1e-06, elementwise_affine=True)\n",
" (mlp): SuperMLPv2(\n",
" in_features=Categorical(candidates=[36, 72, 100], default_index=None), hidden_multiplier=Categorical(candidates=[1, 2, 4], default_index=None), out_features=Categorical(candidates=[36, 72, 100], default_index=None), drop=None, fc1 -> act -> drop -> fc2 -> drop,\n",
" (_params): ParameterDict(\n",
" (fc1_super_weight): Parameter containing: [torch.FloatTensor of size 400x100]\n",
" (fc1_super_bias): Parameter containing: [torch.FloatTensor of size 400]\n",
" (fc2_super_weight): Parameter containing: [torch.FloatTensor of size 100x400]\n",
" (fc2_super_bias): Parameter containing: [torch.FloatTensor of size 100]\n",
" )\n",
" (act): GELU()\n",
" (drop): Dropout(p=0.0, inplace=False)\n",
" )\n",
" )\n",
" )\n",
")\n"
]
}
],
"source": [
"out1_dim = spaces.Categorical(12, 24, 36)\n",
"out2_dim = spaces.Categorical(24, 36, 48)\n",
"out3_dim = spaces.Categorical(36, 72, 100)\n",
"layer1 = _create_stel(input_dim, out1_dim, order)\n",
"layer2 = _create_stel(out1_dim, out2_dim, order)\n",
"layer3 = _create_stel(out2_dim, out3_dim, order)\n",
"model = super_core.SuperSequential(layer1, layer2, layer3)\n",
"print(model)"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "4fd53a7c",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"> \u001b[0;32m/Users/xuanyidong/anaconda3/lib/python3.8/site-packages/xautodl-0.9.9-py3.8.egg/xautodl/xlayers/super_transformer.py\u001b[0m(116)\u001b[0;36mforward_raw\u001b[0;34m()\u001b[0m\n",
"\u001b[0;32m 114 \u001b[0;31m \u001b[0;32mimport\u001b[0m \u001b[0mpdb\u001b[0m\u001b[0;34m;\u001b[0m \u001b[0mpdb\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mset_trace\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
"\u001b[0m\u001b[0;32m 115 \u001b[0;31m \u001b[0;31m# feed-forward layer -- MLP\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
"\u001b[0m\u001b[0;32m--> 116 \u001b[0;31m \u001b[0my\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mnorm2\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mx\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
"\u001b[0m\u001b[0;32m 117 \u001b[0;31m \u001b[0mouts\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mx\u001b[0m \u001b[0;34m+\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mmlp\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0my\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
"\u001b[0m\u001b[0;32m 118 \u001b[0;31m \u001b[0;32melif\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_order\u001b[0m \u001b[0;32mis\u001b[0m \u001b[0mLayerOrder\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mPostNorm\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
"\u001b[0m\n",
"ipdb> print(self)\n",
"SuperTransformerEncoderLayer(\n",
" (norm1): SuperLayerNorm1D(shape=Categorical(candidates=[36, 72, 100], default_index=None), eps=1e-06, elementwise_affine=True)\n",
" (mha): SuperSelfAttention(\n",
" input_dim=Categorical(candidates=[36, 72, 100], default_index=None), proj_dim=Categorical(candidates=[36, 72, 100], default_index=None), num_heads=Categorical(candidates=[2, 4, 6], default_index=None), mask=False, infinity=1000000000.0\n",
" (q_fc): SuperLinear(in_features=Categorical(candidates=[36, 72, 100], default_index=None), out_features=Categorical(candidates=[36, 72, 100], default_index=None), bias=False)\n",
" (k_fc): SuperLinear(in_features=Categorical(candidates=[36, 72, 100], default_index=None), out_features=Categorical(candidates=[36, 72, 100], default_index=None), bias=False)\n",
" (v_fc): SuperLinear(in_features=Categorical(candidates=[36, 72, 100], default_index=None), out_features=Categorical(candidates=[36, 72, 100], default_index=None), bias=False)\n",
" (attn_drop): SuperDrop(p=0.0, dims=[-1, -1, -1, -1], recover=True)\n",
" )\n",
" (drop): Dropout(p=0.0, inplace=False)\n",
" (norm2): SuperLayerNorm1D(shape=Categorical(candidates=[36, 72, 100], default_index=None), eps=1e-06, elementwise_affine=True)\n",
" (mlp): SuperMLPv2(\n",
" in_features=Categorical(candidates=[36, 72, 100], default_index=None), hidden_multiplier=Categorical(candidates=[1, 2, 4], default_index=None), out_features=Categorical(candidates=[36, 72, 100], default_index=None), drop=None, fc1 -> act -> drop -> fc2 -> drop,\n",
" (_params): ParameterDict(\n",
" (fc1_super_weight): Parameter containing: [torch.FloatTensor of size 400x100]\n",
" (fc1_super_bias): Parameter containing: [torch.FloatTensor of size 400]\n",
" (fc2_super_weight): Parameter containing: [torch.FloatTensor of size 100x400]\n",
" (fc2_super_bias): Parameter containing: [torch.FloatTensor of size 100]\n",
" )\n",
" (act): GELU()\n",
" (drop): Dropout(p=0.0, inplace=False)\n",
" )\n",
")\n",
"ipdb> print(inputs.shape)\n",
"torch.Size([1, 4, 100])\n",
"ipdb> print(x.shape)\n",
"torch.Size([1, 4, 96])\n",
"ipdb> print(self.mha)\n",
"SuperSelfAttention(\n",
" input_dim=Categorical(candidates=[36, 72, 100], default_index=None), proj_dim=Categorical(candidates=[36, 72, 100], default_index=None), num_heads=Categorical(candidates=[2, 4, 6], default_index=None), mask=False, infinity=1000000000.0\n",
" (q_fc): SuperLinear(in_features=Categorical(candidates=[36, 72, 100], default_index=None), out_features=Categorical(candidates=[36, 72, 100], default_index=None), bias=False)\n",
" (k_fc): SuperLinear(in_features=Categorical(candidates=[36, 72, 100], default_index=None), out_features=Categorical(candidates=[36, 72, 100], default_index=None), bias=False)\n",
" (v_fc): SuperLinear(in_features=Categorical(candidates=[36, 72, 100], default_index=None), out_features=Categorical(candidates=[36, 72, 100], default_index=None), bias=False)\n",
" (attn_drop): SuperDrop(p=0.0, dims=[-1, -1, -1, -1], recover=True)\n",
")\n",
"ipdb> print(self.mha.candidate)\n",
"*** AttributeError: 'SuperSelfAttention' object has no attribute 'candidate'\n",
"ipdb> print(self.mha.abstract_candidate)\n",
"*** AttributeError: 'SuperSelfAttention' object has no attribute 'abstract_candidate'\n",
"ipdb> print(self.mha._abstract_child)\n",
"None\n",
"ipdb> print(self.abstract_child)\n",
"None\n",
"ipdb> print(self.abstract_child.abstract_child)\n",
"*** AttributeError: 'NoneType' object has no attribute 'abstract_child'\n",
"ipdb> print(self.mha)\n",
"SuperSelfAttention(\n",
" input_dim=Categorical(candidates=[36, 72, 100], default_index=None), proj_dim=Categorical(candidates=[36, 72, 100], default_index=None), num_heads=Categorical(candidates=[2, 4, 6], default_index=None), mask=False, infinity=1000000000.0\n",
" (q_fc): SuperLinear(in_features=Categorical(candidates=[36, 72, 100], default_index=None), out_features=Categorical(candidates=[36, 72, 100], default_index=None), bias=False)\n",
" (k_fc): SuperLinear(in_features=Categorical(candidates=[36, 72, 100], default_index=None), out_features=Categorical(candidates=[36, 72, 100], default_index=None), bias=False)\n",
" (v_fc): SuperLinear(in_features=Categorical(candidates=[36, 72, 100], default_index=None), out_features=Categorical(candidates=[36, 72, 100], default_index=None), bias=False)\n",
" (attn_drop): SuperDrop(p=0.0, dims=[-1, -1, -1, -1], recover=True)\n",
")\n"
]
}
],
"source": [
"inputs = torch.rand(batch, seq_dim, input_dim)\n",
"outputs = model(inputs)"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "05332b98",
"metadata": {},
"outputs": [],
"source": [
"abstract_space = model.abstract_search_space\n",
"abstract_space.clean_last()\n",
"abstract_child = abstract_space.random(reuse_last=True)\n",
"# print(\"The abstract child program is:\\n{:}\".format(abstract_child))\n",
"model.enable_candidate()\n",
"model.set_super_run_type(super_core.SuperRunMode.Candidate)\n",
"model.apply_candidate(abstract_child)\n",
"outputs = model(inputs)"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "3289f938",
"metadata": {},
"outputs": [],
"source": [
"print(outputs.shape)"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "36951cdf",
"metadata": {},
"outputs": [],
"source": []
}
],
"metadata": {
"kernelspec": {
"display_name": "Python 3",
"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": "3.8.8"
}
},
"nbformat": 4,
"nbformat_minor": 5
}