0718-2225

backup_diffusion.ipynb

@@ -0,0 +1,953 @@
+{
+ "cells": [
+  {
+   "attachments": {},
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## 改編ContextUnet及相關代碼，使其首先對二維的情況適用。並於diffusers.Unet2DModel作比較並加以優化。最後再改寫爲3維的情形。\n",
+    "- 經試用diffusers的Unet2DModel，發現loss從0.3降到0.2但仍然很高，説明存在非Unet2DModel的問題可以優化\n",
+    "- 改用diffusers的DDMPScheduler和DDPMPipeline后，loss降低至0.1以下，有時甚至可以低至0.004，可見我的代碼問題主要出在DDPM部分。DDPMScheduler部分比較簡短，似乎沒有問題，所以問題應該在DDPMPipeline裏某一部分代碼是我代碼欠缺的。\n",
+    "- 我在DDPMScheduler部分有一個typo，導致beta_t一直很小，修正后loss從0.2能降低至0.02, 維持在0.1以下\n",
+    "- 用diffusers的DDPMScheduler似乎效果要好一些，loss總是比我的DDPMScheduler要小一點。儅epoch為19時，前者的loss約0.02，後者loss約0.07。而且前者還支持3維圖像的加噪，不如直接用別人的輪子。但我想知道爲什麽我的loss會高一些。\n",
+    "- 我意識到別人的DDPMScheduler在sample函數中沒有兼容輸入參數，所以歸根結底還是需要我的DDPMscheduler。不過我可以先用別人的來debug我的ContextUnet.\n",
+    "- 我需要將我的ContextUnet擴展兼容不同維度的照片，畢竟我本身也需要和原文獻對比完了再拓展到三維的情形\n",
+    "- 我已將我的ContextUnet轉成了2維的模式，與diffusers.Unet2DModel的loss=0.037相比，我的Unet的loss=0.07。同時我的Unet生成的圖像看上去很奇怪，説明我的Unet也有問題。我需要將代碼退回原Unet，並檢查問題所在。\n",
+    "- 我將紅移方向的像素的數量限制在了64.以此比較兩個Unet的差別。經比較：\\\n",
+    "Unet2DModel loss：0.03, 0.0655, 0.05, 0.02, 0.05\\\n",
+    "ContextUnet loss: 0.1, 0.16, 0.1, 0.2186, 0.06\n",
+    "- 我把ContextUnet退回到了原作者的版本，結果loss=0.05，輸出的照片也不錯。我主要的改動是改回了他原用的normalization函數，其中還有個參數swish。有時間我可以研究一下具體是哪裏影響了訓練的結果。另外我發現了要想tensorboard的圖綫獨立美觀，需要把他們放在不同的文件夾下\n",
+    "- 經過驗證，GroupNorm比batchNorm效果要好\n",
+    "- 已擴展爲接受不同維度的情形\n",
+    "- 融合cond, guide_w, drop_out這些參數\n",
+    "- 生成的21cm圖像該暗的地方不夠暗，似乎換成MNIST的數字圖像就沒問題\n",
+    "- 我用diffusion模型生成MNIST的數字時發現，儘管生成的數據的範圍也存在負數數值，如-0.1,但畫出來的圖像卻是理想的黑色。數據的分佈與21cm的結果的分佈沒多大差別，我現在打算把代碼退回到21cm的情形\n",
+    "- 我統一了ddpm21cm這個module，能統一實現訓練和生成樣本，但目前有個bug， sample時總是會cuda out of memory，然而單獨resume model並sample就不會。\n",
+    "- 解決了，問題出在我忘了寫with torch.no_grad():\n",
+    "- 接下來就是生成800個lightcones，與此同時研究如何計算global signal以及power spectrum\n",
+    "- 儅訓練圖片的數量達到5000時，生成的圖片與檢測數據的相似程度很高\n",
+    "- it takes 62 mins to generated 8 images with shape of (64,64,64), which is even slower than simulation, which takes ~5 mins for each image. Besides, the batch_size during training and num of images to be generated are limited to be 2 and 8, respectively.\n",
+    "- the slowerness can be solved by using multi-GPUs, and the limited-num-of-images can be solved by multi-accuracy, multi-GPUs.\n",
+    "- In addtion, the performance of DDPM can looks better compared to computation-intensive simulations. "
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 31,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from dataclasses import dataclass\n",
+    "import h5py\n",
+    "import torch\n",
+    "import torch.nn as nn\n",
+    "from torch.utils.data import DataLoader, Dataset\n",
+    "# from datasets import Dataset\n",
+    "import matplotlib.pyplot as plt\n",
+    "import numpy as np\n",
+    "import random\n",
+    "# from abc import ABC, abstractmethod\n",
+    "import torch.nn.functional as F\n",
+    "import math\n",
+    "# from PIL import Image\n",
+    "import os\n",
+    "from torch.utils.tensorboard import SummaryWriter\n",
+    "import copy\n",
+    "from tqdm.auto import tqdm\n",
+    "# from torchvision import transforms\n",
+    "# from diffusers import UNet2DModel#, UNet3DConditionModel\n",
+    "# from diffusers import DDPMScheduler\n",
+    "from diffusers.utils import make_image_grid\n",
+    "import datetime\n",
+    "from pathlib import Path\n",
+    "from diffusers.optimization import get_cosine_schedule_with_warmup\n",
+    "from accelerate import notebook_launcher, Accelerator\n",
+    "from huggingface_hub import create_repo, upload_folder\n",
+    "from load_h5 import Dataset4h5\n",
+    "\n",
+    "from context_unet import ContextUnet\n",
+    "\n",
+    "from huggingface_hub import notebook_login\n",
+    "\n",
+    "import torch.multiprocessing as mp\n",
+    "from torch.utils.data.distributed import DistributedSampler\n",
+    "from torch.nn.parallel import DistributedDataParallel as DDP\n",
+    "from torch.distributed import init_process_group, destroy_process_group"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 32,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "def ddp_setup(rank: int, world_size: int):\n",
+    "  \"\"\"\n",
+    "  Args:\n",
+    "      rank: Unique identifier of each process\n",
+    "     world_size: Total number of processes\n",
+    "  \"\"\"\n",
+    "  os.environ[\"MASTER_ADDR\"] = \"localhost\"\n",
+    "  os.environ[\"MASTER_PORT\"] = \"12355\"\n",
+    "  torch.cuda.set_device(rank)\n",
+    "  init_process_group(backend=\"nccl\", rank=rank, world_size=world_size)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 34,
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "application/vnd.jupyter.widget-view+json": {
+       "model_id": "9bbf7e9db9ce426d9c59d6f6d8e8df29",
+       "version_major": 2,
+       "version_minor": 0
+      },
+      "text/plain": [
+       "VBox(children=(HTML(value='<center> <img\\nsrc=https://huggingface.co/front/assets/huggingface_logo-noborder.sv…"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    }
+   ],
+   "source": [
+    "notebook_login()"
+   ]
+  },
+  {
+   "attachments": {},
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "# Add noise:\n",
+    "\n",
+    "\\begin{align*}\n",
+    "x_t &\\sim \\mathcal N\\left(\\sqrt{1-\\beta_t}\\ x_{t-1},\\ \\beta_t \\right) \\\\\n",
+    "x_t &\\equiv \\sqrt{1-\\beta_t}\\ x_{t-1} + \\sqrt{\\beta_t}\\ \\epsilon\\\\\n",
+    "\\epsilon &\\sim \\mathcal N(0,1)\\\\\n",
+    "\\alpha_t & \\equiv 1 - \\beta_t\\\\\n",
+    "& ...\\\\\n",
+    "x_t &= \\sqrt{\\bar {\\alpha_t}} x_0 + \\epsilon\\ \\sqrt{1 - \\bar{\\alpha_t}}\\\\\n",
+    "\\bar {\\alpha_t} &\\equiv \\prod_{i=1}^t \\alpha_i\\\\\n",
+    "&= \\exp\\left({\\ln{\\prod_{i=1}^t \\alpha_i}}\\right)\\\\\n",
+    "&= \\exp\\left({\\sum_{i=1}^t\\ln{ \\alpha_i}}\\right)\n",
+    "\\end{align*}"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 4,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "class DDPMScheduler(nn.Module):\n",
+    "    def __init__(self, betas: tuple, num_timesteps: int, img_shape: list, device='cpu'):\n",
+    "        super().__init__()\n",
+    "        \n",
+    "        beta_1, beta_T = betas\n",
+    "        assert 0 < beta_1 <= beta_T <= 1, \"ensure 0 < beta_1 <= beta_T <= 1\"\n",
+    "        self.device = device\n",
+    "        self.num_timesteps = num_timesteps\n",
+    "        self.img_shape = img_shape\n",
+    "        self.beta_t = torch.linspace(beta_1, beta_T, self.num_timesteps) #* (beta_T-beta_1) + beta_1\n",
+    "        self.beta_t = self.beta_t.to(self.device)\n",
+    "\n",
+    "        # self.drop_prob = drop_prob\n",
+    "        # self.cond = cond\n",
+    "        self.alpha_t = 1 - self.beta_t\n",
+    "        # self.bar_alpha_t = torch.exp(torch.cumsum(torch.log(self.alpha_t), dim=0))\n",
+    "        self.bar_alpha_t = torch.cumprod(self.alpha_t, dim=0)\n",
+    "\n",
+    "    def add_noise(self, clean_images):\n",
+    "        shape = clean_images.shape\n",
+    "        expand = torch.ones(len(shape)-1, dtype=int)\n",
+    "        # ts_expand = ts.view(ts.shape[0], *expand.tolist())\n",
+    "        # expand = [1 for i in range(len(shape)-1)]\n",
+    "\n",
+    "        noise = torch.randn_like(clean_images).to(self.device)\n",
+    "        ts = torch.randint(0, self.num_timesteps, (shape[0],)).to(self.device)\n",
+    "                \n",
+    "        # test_expand = test.view(test.shape[0],*expand)\n",
+    "        # extend_dim = [None for i in range(shape.dim()-1)]\n",
+    "        noisy_images = (\n",
+    "            clean_images * torch.sqrt(self.bar_alpha_t[ts]).view(shape[0], *expand.tolist())\n",
+    "            + noise * torch.sqrt(1-self.bar_alpha_t[ts]).view(shape[0], *expand.tolist())\n",
+    "            )\n",
+    "        # print(x_t.shape)\n",
+    "\n",
+    "        return noisy_images, noise, ts\n",
+    "\n",
+    "    def sample(self, nn_model, params, device, guide_w = 0):\n",
+    "        n_sample = len(params) #params.shape[0]\n",
+    "        # print(\"params.shape[0], len(params)\", params.shape[0], len(params))\n",
+    "        x_i = torch.randn(n_sample, *self.img_shape).to(device)\n",
+    "        # print(\"x_i.shape =\", x_i.shape)\n",
+    "        # print(\"x_i.shape =\", x_i.shape)\n",
+    "        if guide_w != -1:\n",
+    "            c_i = params\n",
+    "            uncond_tokens = torch.zeros(int(n_sample), params.shape[1]).to(device)\n",
+    "            # uncond_tokens = torch.tensor(np.float32(np.array([0,0]))).to(device)\n",
+    "            # uncond_tokens = uncond_tokens.repeat(int(n_sample),1)\n",
+    "            c_i = torch.cat((c_i, uncond_tokens), 0)\n",
+    "\n",
+    "        x_i_entire = [] # keep track of generated steps in case want to plot something\n",
+    "        # print(\"self.num_timesteps =\", self.num_timesteps)\n",
+    "        # for i in range(self.num_timesteps, 0, -1):\n",
+    "        # print(f'sampling!!!')\n",
+    "        pbar_sample = tqdm(total=self.num_timesteps)\n",
+    "        pbar_sample.set_description(\"Sampling\")\n",
+    "        for i in reversed(range(0, self.num_timesteps)):\n",
+    "            # print(f'sampling timestep {i:4d}',end='\\r')\n",
+    "            t_is = torch.tensor([i]).to(device)\n",
+    "            t_is = t_is.repeat(n_sample)\n",
+    "\n",
+    "            z = torch.randn(n_sample, *self.img_shape).to(device) if i > 0 else 0\n",
+    "\n",
+    "            if guide_w == -1:\n",
+    "                # eps = nn_model(x_i, t_is, return_dict=False)[0]\n",
+    "                eps = nn_model(x_i, t_is)\n",
+    "                # x_i = 1/torch.sqrt(self.alpha_t[i])*(x_i-eps*self.beta_t[i]/torch.sqrt(1-self.bar_alpha_t[i])) + torch.sqrt(self.beta_t[i])*z\n",
+    "            else:\n",
+    "                # double batch\n",
+    "                x_i = x_i.repeat(2, *torch.ones(len(self.img_shape), dtype=int).tolist())\n",
+    "                t_is = t_is.repeat(2)\n",
+    "\n",
+    "                # split predictions and compute weighting\n",
+    "                # print(\"nn_model input shape\", x_i.shape, t_is.shape, c_i.shape)\n",
+    "                eps = nn_model(x_i, t_is, c_i)\n",
+    "                eps1 = eps[:n_sample]\n",
+    "                eps2 = eps[n_sample:]\n",
+    "                eps = eps1 + guide_w*(eps1 - eps2)\n",
+    "                # eps = (1+guide_w)*eps1 - guide_w*eps2\n",
+    "                x_i = x_i[:n_sample]\n",
+    "                # x_i = 1/torch.sqrt(self.alpha_t[i])*(x_i-eps*self.beta_t[i]/torch.sqrt(1-self.bar_alpha_t[i])) + torch.sqrt(self.beta_t[i])*z\n",
+    "            \n",
+    "            # print(\"x_i.shape =\", x_i.shape)\n",
+    "            x_i = 1/torch.sqrt(self.alpha_t[i])*(x_i-eps*self.beta_t[i]/torch.sqrt(1-self.bar_alpha_t[i])) + torch.sqrt(self.beta_t[i])*z\n",
+    "            \n",
+    "            pbar_sample.update(1)\n",
+    "            # pbar_sample.set_postfix(step=i)\n",
+    "            \n",
+    "            # print(\"x_i.shape =\", x_i.shape)\n",
+    "            # store only part of the intermediate steps\n",
+    "            if i%20==0:# or i==0:# or i<8:\n",
+    "                x_i_entire.append(x_i.detach().cpu().numpy())\n",
+    "        x_i = x_i.detach().cpu().numpy()\n",
+    "        x_i_entire = np.array(x_i_entire)\n",
+    "        return x_i, x_i_entire\n",
+    "\n",
+    "\n",
+    "# ddpm_scheduler = DDPMScheduler((1e-4,0.02),10)\n",
+    "# noisy_images, noise, ts = ddpm_scheduler.add_noise(images)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 5,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "class EMA:\n",
+    "    def __init__(self, beta):\n",
+    "        super().__init__()\n",
+    "        self.beta = beta\n",
+    "        self.step = 0\n",
+    "\n",
+    "    def update_model_average(self, ma_model, current_model):\n",
+    "        for current_params, ma_params in zip(current_model.parameters(), ma_model.parameters()):\n",
+    "            old_weight, up_weight = ma_params.data, current_params.data\n",
+    "            ma_params.data = self.update_average(old_weight, up_weight)\n",
+    "\n",
+    "    def update_average(self, old, new):\n",
+    "        if old is None:\n",
+    "            return new\n",
+    "        return old * self.beta + (1 - self.beta) * new\n",
+    "\n",
+    "    def step_ema(self, ema_model, model):\n",
+    "        self.update_model_average(ema_model, model)\n",
+    "        self.step += 1\n",
+    "\n",
+    "    def reset_parameters(self, ema_model, model):\n",
+    "        ema_model.load_state_dict(model.state_dict())\n",
+    "        "
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 6,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "@dataclass\n",
+    "class TrainConfig:\n",
+    "    ###########################\n",
+    "    ## hardcoding these here ##\n",
+    "    ###########################\n",
+    "    push_to_hub = True\n",
+    "    hub_model_id = \"Xsmos/ml21cm\"\n",
+    "    hub_private_repo = False\n",
+    "    dataset_name = \"/storage/home/hcoda1/3/bxia34/scratch/LEN128-DIM64-CUB8.h5\"\n",
+    "    device = 'cuda' if torch.cuda.is_available() else 'cpu'\n",
+    "    # world_size = torch.cuda.device_count()\n",
+    "    # repeat = 2\n",
+    "\n",
+    "    # dim = 2\n",
+    "    dim = 3\n",
+    "    stride = (2,2) if dim == 2 else (2,2,1)\n",
+    "    num_image = 2000#32000#20000#15000#7000#25600#3000#10000#1000#10000#5000#2560#800#2560\n",
+    "    batch_size = 2#2#50#20#2#100 # 10\n",
+    "    n_epoch = 10#50#20#20#2#5#25 # 120\n",
+    "    HII_DIM = 28#64\n",
+    "    num_redshift = 4#128#64#512#256#256#64#512#128\n",
+    "    channel = 1\n",
+    "    img_shape = (channel, HII_DIM, num_redshift) if dim == 2 else (channel, HII_DIM, HII_DIM, num_redshift)\n",
+    "\n",
+    "    ranges_dict = dict(\n",
+    "        params = {\n",
+    "            0: [4, 6], # ION_Tvir_MIN\n",
+    "            1: [10, 250], # HII_EFF_FACTOR\n",
+    "            },\n",
+    "        images = {\n",
+    "            0: [0, 80], # brightness_temp\n",
+    "            }\n",
+    "        )\n",
+    "\n",
+    "    num_timesteps = 1000#1000 # 1000, 500; DDPM time steps\n",
+    "    # n_sample = 24 # 64, the number of samples in sampling process\n",
+    "    n_param = 2\n",
+    "    guide_w = 0#-1#0#-1#0#-1#0.1#[0,0.1] #[0,0.5,2] strength of generative guidance\n",
+    "    drop_prob = 0#0.28 # only takes effect when guide_w != -1\n",
+    "    ema=True # whether to use ema\n",
+    "    ema_rate=0.995\n",
+    "\n",
+    "    # seed = 0\n",
+    "    # save_dir = './outputs/'\n",
+    "\n",
+    "    save_freq = 0#.1 # the period of sampling\n",
+    "    # general parameters for the name and logger    \n",
+    "    # device = \"cuda\" if torch.cuda.is_available() else \"cpu\"\n",
+    "    lrate = 1e-4\n",
+    "    lr_warmup_steps = 0#5#00\n",
+    "    output_dir = \"./outputs/\"\n",
+    "    save_name = os.path.join(output_dir, 'model_state')\n",
+    "    # save_freq = 1 #10 # the period of saving model\n",
+    "    # cond = True # if training using the conditional information\n",
+    "    # lr_decay = False #True# if using the learning rate decay\n",
+    "    resume = save_name # if resume from the trained checkpoints\n",
+    "    # params_single = torch.tensor([0.2,0.80000023])\n",
+    "    # params = torch.tile(params_single,(n_sample,1)).to(device)\n",
+    "    # params =  params\n",
+    "    # data_dir = './data' # data directory\n",
+    "\n",
+    "\n",
+    "    mixed_precision = \"fp16\"\n",
+    "    gradient_accumulation_steps = 1\n",
+    "\n",
+    "    # date = datetime.datetime.now().strftime(\"%m%d-%H%M\")\n",
+    "    # run_name = f'{date}' # the unique name of each experiment\n",
+    "\n",
+    "# config = TrainConfig()\n",
+    "# print(\"device =\", config.device)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 7,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# import os\n",
+    "# print(os.cpu_count())\n",
+    "# print(len(os.sched_getaffinity(0)))\n",
+    "# import torch\n",
+    "# data = torch.randn((64,64))\n",
+    "# print(data.dtype)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 8,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# @dataclass\n",
+    "class DDPM21CM:\n",
+    "    def __init__(self, config):\n",
+    "        # config = TrainConfig()\n",
+    "        # date = datetime.datetime.now().strftime(\"%m%d-%H%M\")\n",
+    "        config.run_name = datetime.datetime.now().strftime(\"%m%d-%H%M\") # the unique name of each experiment\n",
+    "        self.config = config\n",
+    "        # dataset = Dataset4h5(config.dataset_name, num_image=config.num_image, HII_DIM=config.HII_DIM, num_redshift=config.num_redshift, drop_prob=config.drop_prob, dim=config.dim)\n",
+    "        # # self.shape_loaded = dataset.images.shape\n",
+    "        # # print(\"shape_loaded =\", self.shape_loaded)\n",
+    "        # self.dataloader = DataLoader(dataset, batch_size=config.batch_size, shuffle=True)\n",
+    "        # del dataset\n",
+    "        self.ddpm = DDPMScheduler(betas=(1e-4, 0.02), num_timesteps=config.num_timesteps, img_shape=config.img_shape, device=config.device)\n",
+    "\n",
+    "        # initialize the unet\n",
+    "        self.nn_model = ContextUnet(n_param=config.n_param, image_size=config.HII_DIM, dim=config.dim, stride=config.stride)\n",
+    "\n",
+    "        if config.resume and os.path.exists(config.resume):\n",
+    "            # resume_file = os.path.join(config.output_dir, f\"{config.resume}\")\n",
+    "            self.nn_model.load_state_dict(torch.load(config.resume)['unet_state_dict'])\n",
+    "            print(f\"resumed nn_model from {config.resume}\")\n",
+    "        # nn_model = ContextUnet(n_param=1, image_size=28)\n",
+    "        self.nn_model.train()\n",
+    "        self.nn_model.to(self.ddpm.device)\n",
+    "        # print(\"nn_model.device =\", ddpm.device)\n",
+    "        # number of parameters to be trained\n",
+    "        self.number_of_params = sum(x.numel() for x in self.nn_model.parameters())\n",
+    "        print(f\"Number of parameters for nn_model: {self.number_of_params}\")\n",
+    "\n",
+    "        # whether to use ema\n",
+    "        if config.ema:\n",
+    "            self.ema = EMA(config.ema_rate)\n",
+    "            if config.resume and os.path.exists(config.resume):\n",
+    "                self.ema_model = ContextUnet(n_param=config.n_param, image_size=config.HII_DIM, dim=config.dim, stride=config.stride).to(config.device)\n",
+    "                self.ema_model.load_state_dict(torch.load(config.resume)['ema_unet_state_dict'])\n",
+    "                print(f\"resumed ema_model from {config.resume}\")\n",
+    "            else:\n",
+    "                self.ema_model = copy.deepcopy(self.nn_model).eval().requires_grad_(False)\n",
+    "\n",
+    "        self.optimizer = torch.optim.AdamW(self.nn_model.parameters(), lr=config.lrate)\n",
+    "        self.lr_scheduler = get_cosine_schedule_with_warmup(\n",
+    "            optimizer=self.optimizer,\n",
+    "            num_warmup_steps=config.lr_warmup_steps,\n",
+    "            num_training_steps=(int(config.num_image/config.batch_size) * config.n_epoch),\n",
+    "            # num_training_steps=(len(self.dataloader) * config.n_epoch),\n",
+    "        )\n",
+    "\n",
+    "        self.ranges_dict = config.ranges_dict\n",
+    "\n",
+    "    def load(self):\n",
+    "        dataset = Dataset4h5(self.config.dataset_name, num_image=self.config.num_image, HII_DIM=self.config.HII_DIM, num_redshift=self.config.num_redshift, drop_prob=self.config.drop_prob, dim=self.config.dim, ranges_dict=self.ranges_dict)\n",
+    "        # self.shape_loaded = dataset.images.shape\n",
+    "        # print(\"shape_loaded =\", self.shape_loaded)\n",
+    "        self.dataloader = DataLoader(dataset, batch_size=self.config.batch_size, shuffle=True, num_workers=len(os.sched_getaffinity(0)), pin_memory=True)\n",
+    "        # del dataset\n",
+    "        # self.accelerate(self.config)\n",
+    "        del dataset\n",
+    "\n",
+    "    # def accelerate(self):\n",
+    "\n",
+    "    def train(self):\n",
+    "        ###################      \n",
+    "        ## training loop ##\n",
+    "        ###################\n",
+    "        # plot_unet = True\n",
+    "\n",
+    "        self.load()\n",
+    "        self.accelerator = Accelerator(\n",
+    "            mixed_precision=self.config.mixed_precision,\n",
+    "            gradient_accumulation_steps=self.config.gradient_accumulation_steps,\n",
+    "            log_with=\"tensorboard\",\n",
+    "            project_dir=os.path.join(self.config.output_dir, \"logs\"),\n",
+    "        )\n",
+    "        print(\"self.accelerator.is_main_process:\", self.accelerator.is_main_process)\n",
+    "        if self.accelerator.is_main_process:\n",
+    "            if self.config.output_dir is not None:\n",
+    "                os.makedirs(self.config.output_dir, exist_ok=True)\n",
+    "            if self.config.push_to_hub:\n",
+    "                self.repo_id = create_repo(\n",
+    "                    repo_id=self.config.hub_model_id or Path(self.config.output_dir).name, exist_ok=True\n",
+    "                ).repo_id\n",
+    "            self.accelerator.init_trackers(f\"{self.config.run_name}\")\n",
+    "\n",
+    "        self.nn_model, self.optimizer, self.dataloader, self.lr_scheduler = \\\n",
+    "            self.accelerator.prepare(\n",
+    "            self.nn_model, self.optimizer, self.dataloader, self.lr_scheduler\n",
+    "            )\n",
+    "            \n",
+    "        global_step = 0\n",
+    "        for ep in range(self.config.n_epoch):\n",
+    "            self.ddpm.train()\n",
+    "\n",
+    "            pbar_train = tqdm(total=len(self.dataloader), disable=not self.accelerator.is_local_main_process)\n",
+    "            pbar_train.set_description(f\"Epoch {ep}\")\n",
+    "            for i, (x, c) in enumerate(self.dataloader):\n",
+    "                with self.accelerator.accumulate(self.nn_model):\n",
+    "                    x = x.to(self.config.device)\n",
+    "                    xt, noise, ts = self.ddpm.add_noise(x)\n",
+    "                    \n",
+    "                    if self.config.guide_w == -1:\n",
+    "                        noise_pred = self.nn_model(xt, ts)\n",
+    "                    else:\n",
+    "                        c = c.to(self.config.device)\n",
+    "                        noise_pred = self.nn_model(xt, ts, c)\n",
+    "                    \n",
+    "                    loss = F.mse_loss(noise, noise_pred)\n",
+    "                    self.accelerator.backward(loss)\n",
+    "                    self.accelerator.clip_grad_norm_(self.nn_model.parameters(), 1)\n",
+    "                    self.optimizer.step()\n",
+    "                    self.lr_scheduler.step()\n",
+    "                    self.optimizer.zero_grad()\n",
+    "\n",
+    "                # ema update\n",
+    "                if self.config.ema:\n",
+    "                    self.ema.step_ema(self.ema_model, self.nn_model)\n",
+    "\n",
+    "                pbar_train.update(1)\n",
+    "                logs = dict(\n",
+    "                    loss=loss.detach().item(),\n",
+    "                    lr=self.optimizer.param_groups[0]['lr'],\n",
+    "                    step=global_step\n",
+    "                )\n",
+    "                pbar_train.set_postfix(**logs)\n",
+    "\n",
+    "                self.accelerator.log(logs, step=global_step)\n",
+    "                global_step += 1\n",
+    "\n",
+    "            # if ep == config.n_epoch-1 or (ep+1)*config.save_freq==1:\n",
+    "            self.save(ep)\n",
+    "\n",
+    "        del self.nn_model\n",
+    "        if self.config.ema:\n",
+    "            del self.ema_model\n",
+    "        torch.cuda.empty_cache()\n",
+    "\n",
+    "    def save(self, ep):\n",
+    "        # save model\n",
+    "        if self.accelerator.is_main_process:\n",
+    "            if ep == self.config.n_epoch-1 or (ep+1)*self.config.save_freq==1:\n",
+    "                self.nn_model.eval()\n",
+    "                with torch.no_grad():\n",
+    "                    if self.config.push_to_hub:\n",
+    "                        upload_folder(\n",
+    "                            repo_id = self.repo_id,\n",
+    "                            folder_path = \".\",#config.output_dir,\n",
+    "                            commit_message = f\"{self.config.run_name}\",\n",
+    "                            ignore_patterns = [\"step_*\", \"epoch_*\", \"*.npy\", \"__pycache__\"],\n",
+    "                            )\n",
+    "                    if self.config.save_name:\n",
+    "                        model_state = {\n",
+    "                            'epoch': ep,\n",
+    "                            'unet_state_dict': self.nn_model.state_dict(),\n",
+    "                            'ema_unet_state_dict': self.ema_model.state_dict(),\n",
+    "                            }\n",
+    "                        torch.save(model_state, self.config.save_name+f\"-N{self.config.num_image}\")\n",
+    "                        print('saved model at ' + self.config.save_name+f\"-N{self.config.num_image}\")\n",
+    "                        # print('saved model at ' + config.save_dir + f\"model_epoch_{ep}_test_{config.run_name}.pth\")\n",
+    "\n",
+    "    # def rescale(self, value, type='params', to_ranges=[0,1]):\n",
+    "    #     for i, from_ranges in self.ranges_dict[type].items():\n",
+    "    #         value[i] = (value[i] - from_ranges[0])/(from_ranges[1]-from_ranges[0]) # normalize\n",
+    "    #         value[i] = \n",
+    "    def rescale(self, value, ranges, to: list):\n",
+    "        if value.ndim == 1:\n",
+    "            value = value.view(-1,len(value))\n",
+    "            \n",
+    "        for i in range(np.shape(value)[1]):\n",
+    "            value[:,i] = (value[:,i] - ranges[i][0]) / (ranges[i][1]-ranges[i][0])\n",
+    "            # print(f\"i = {i}, value.min = {value[:,i].min()}, value.max = {value[:,i].max()}\")\n",
+    "        value = value * (to[1]-to[0]) + to[0]\n",
+    "        return value \n",
+    "\n",
+    "    def sample(self, file, params:torch.tensor=None, repeat=192, ema=False, entire=False):\n",
+    "        # n_sample = params.shape[0]\n",
+    "        \n",
+    "        if params is None:\n",
+    "            params = torch.tensor([0.20000000000000018, 0.5055875000000001])\n",
+    "            params_backup = params.numpy().copy()\n",
+    "        else:\n",
+    "            params_backup = params.numpy().copy()\n",
+    "            params = self.rescale(params, self.ranges_dict['params'], to=[0,1])\n",
+    "\n",
+    "        print(f\"sampling {repeat} images with normalized params = {params}\")\n",
+    "        params = params.repeat(repeat,1)\n",
+    "        assert params.dim() == 2, \"params must be a 2D torch.tensor\"\n",
+    "        # print(\"params =\", params)\n",
+    "        # print(\"params =\", params)\n",
+    "        # print(\"len(params) =\", len(params))\n",
+    "        # model = self.ema_model if ema else self.nn_model\n",
+    "        # del self.ema_model, self.nn\n",
+    "        # params = torch.tile(params, (n_sample,1)).to(device)\n",
+    "\n",
+    "        nn_model = ContextUnet(n_param=self.config.n_param, image_size=self.config.HII_DIM, dim=self.config.dim, stride=self.config.stride).to(self.config.device)\n",
+    "        if ema:\n",
+    "            nn_model.load_state_dict(torch.load(file)['ema_unet_state_dict'])\n",
+    "        else:\n",
+    "            nn_model.load_state_dict(torch.load(file)['unet_state_dict'])\n",
+    "        print(f\"nn_model resumed from {file}\")\n",
+    "        # nn_model = ContextUnet(n_param=1, image_size=28)\n",
+    "        # nn_model.train()\n",
+    "        nn_model.to(self.ddpm.device)\n",
+    "        nn_model.eval()\n",
+    "\n",
+    "        # self.ema_model = ContextUnet(n_param=config.n_param, image_size=config.HII_DIM, dim=config.dim, stride=config.stride).to(config.device)\n",
+    "        # self.ema_model.load_state_dict(torch.load(os.path.join(config.output_dir, f\"{config.resume}\"))['ema_unet_state_dict'])\n",
+    "        # print(f\"resumed ema_model from {config.resume}\")\n",
+    "\n",
+    "        with torch.no_grad():\n",
+    "            x_last, x_entire = self.ddpm.sample(\n",
+    "                nn_model=nn_model, \n",
+    "                params=params.to(self.config.device), \n",
+    "                device=self.config.device, \n",
+    "                guide_w=self.config.guide_w\n",
+    "                )\n",
+    "\n",
+    "        # np.save(os.path.join(self.config.output_dir, f\"{self.config.run_name}{'ema' if ema else ''}.npy\"), x_last)\n",
+    "        np.save(os.path.join(self.config.output_dir, f\"Tvir{params_backup[0]}-zeta{params_backup[1]}-N{self.config.num_image}{'ema' if ema else ''}.npy\"), x_last)\n",
+    "\n",
+    "        if entire:\n",
+    "            np.save(os.path.join(self.config.output_dir, f\"Tvir{params_backup[0]}-zeta{params_backup[1]}-N{self.config.num_image}{'ema' if ema else ''}_entire.npy\"), x_last)\n",
+    "# print(\"device =\", config.device)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 12,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Traceback (most recent call last):\n",
+      "  File \"<string>\", line 1, in <module>\n",
+      "  File \"/storage/home/hcoda1/3/bxia34/.conda/envs/diffusers/lib/python3.9/multiprocessing/spawn.py\", line 116, in spawn_main\n",
+      "    exitcode = _main(fd, parent_sentinel)\n",
+      "  File \"/storage/home/hcoda1/3/bxia34/.conda/envs/diffusers/lib/python3.9/multiprocessing/spawn.py\", line 126, in _main\n",
+      "    self = reduction.pickle.load(from_parent)\n",
+      "AttributeError: Can't get attribute 'single_main' on <module '__main__' (built-in)>\n"
+     ]
+    },
+    {
+     "ename": "ProcessExitedException",
+     "evalue": "process 0 terminated with exit code 1",
+     "output_type": "error",
+     "traceback": [
+      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
+      "\u001b[0;31mProcessExitedException\u001b[0m                    Traceback (most recent call last)",
+      "Cell \u001b[0;32mIn[12], line 21\u001b[0m\n\u001b[1;32m     16\u001b[0m \u001b[39mif\u001b[39;00m \u001b[39m__name__\u001b[39m \u001b[39m==\u001b[39m \u001b[39m\"\u001b[39m\u001b[39m__main__\u001b[39m\u001b[39m\"\u001b[39m:\n\u001b[1;32m     17\u001b[0m     \u001b[39m# torch.multiprocessing.set_start_method(\"spawn\")\u001b[39;00m\n\u001b[1;32m     18\u001b[0m     \u001b[39m# args = (config, nn_model, ddpm, optimizer, dataloader, lr_scheduler)\u001b[39;00m\n\u001b[1;32m     19\u001b[0m     world_size \u001b[39m=\u001b[39m \u001b[39m1\u001b[39m\u001b[39m#torch.cuda.device_count()\u001b[39;00m\n\u001b[0;32m---> 21\u001b[0m     mp\u001b[39m.\u001b[39;49mspawn(single_main, args\u001b[39m=\u001b[39;49m(world_size,), nprocs\u001b[39m=\u001b[39;49mworld_size)\n\u001b[1;32m     22\u001b[0m     \u001b[39m# notebook_launcher(ddpm21cm.train, num_processes=1, mixed_precision='fp16')\u001b[39;00m\n",
+      "File \u001b[0;32m/usr/local/pace-apps/manual/packages/pytorch/1.12.0/lib/python3.9/site-packages/torch/multiprocessing/spawn.py:240\u001b[0m, in \u001b[0;36mspawn\u001b[0;34m(fn, args, nprocs, join, daemon, start_method)\u001b[0m\n\u001b[1;32m    236\u001b[0m     msg \u001b[39m=\u001b[39m (\u001b[39m'\u001b[39m\u001b[39mThis method only supports start_method=spawn (got: \u001b[39m\u001b[39m%s\u001b[39;00m\u001b[39m).\u001b[39m\u001b[39m\\n\u001b[39;00m\u001b[39m'\u001b[39m\n\u001b[1;32m    237\u001b[0m            \u001b[39m'\u001b[39m\u001b[39mTo use a different start_method use:\u001b[39m\u001b[39m\\n\u001b[39;00m\u001b[39m\\t\u001b[39;00m\u001b[39m\\t\u001b[39;00m\u001b[39m'\u001b[39m\n\u001b[1;32m    238\u001b[0m            \u001b[39m'\u001b[39m\u001b[39m torch.multiprocessing.start_processes(...)\u001b[39m\u001b[39m'\u001b[39m \u001b[39m%\u001b[39m start_method)\n\u001b[1;32m    239\u001b[0m     warnings\u001b[39m.\u001b[39mwarn(msg)\n\u001b[0;32m--> 240\u001b[0m \u001b[39mreturn\u001b[39;00m start_processes(fn, args, nprocs, join, daemon, start_method\u001b[39m=\u001b[39;49m\u001b[39m'\u001b[39;49m\u001b[39mspawn\u001b[39;49m\u001b[39m'\u001b[39;49m)\n",
+      "File \u001b[0;32m/usr/local/pace-apps/manual/packages/pytorch/1.12.0/lib/python3.9/site-packages/torch/multiprocessing/spawn.py:198\u001b[0m, in \u001b[0;36mstart_processes\u001b[0;34m(fn, args, nprocs, join, daemon, start_method)\u001b[0m\n\u001b[1;32m    195\u001b[0m     \u001b[39mreturn\u001b[39;00m context\n\u001b[1;32m    197\u001b[0m \u001b[39m# Loop on join until it returns True or raises an exception.\u001b[39;00m\n\u001b[0;32m--> 198\u001b[0m \u001b[39mwhile\u001b[39;00m \u001b[39mnot\u001b[39;00m context\u001b[39m.\u001b[39;49mjoin():\n\u001b[1;32m    199\u001b[0m     \u001b[39mpass\u001b[39;00m\n",
+      "File \u001b[0;32m/usr/local/pace-apps/manual/packages/pytorch/1.12.0/lib/python3.9/site-packages/torch/multiprocessing/spawn.py:149\u001b[0m, in \u001b[0;36mProcessContext.join\u001b[0;34m(self, timeout)\u001b[0m\n\u001b[1;32m    140\u001b[0m         \u001b[39mraise\u001b[39;00m ProcessExitedException(\n\u001b[1;32m    141\u001b[0m             \u001b[39m\"\u001b[39m\u001b[39mprocess \u001b[39m\u001b[39m%d\u001b[39;00m\u001b[39m terminated with signal \u001b[39m\u001b[39m%s\u001b[39;00m\u001b[39m\"\u001b[39m \u001b[39m%\u001b[39m\n\u001b[1;32m    142\u001b[0m             (error_index, name),\n\u001b[0;32m   (...)\u001b[0m\n\u001b[1;32m    146\u001b[0m             signal_name\u001b[39m=\u001b[39mname\n\u001b[1;32m    147\u001b[0m         )\n\u001b[1;32m    148\u001b[0m     \u001b[39melse\u001b[39;00m:\n\u001b[0;32m--> 149\u001b[0m         \u001b[39mraise\u001b[39;00m ProcessExitedException(\n\u001b[1;32m    150\u001b[0m             \u001b[39m\"\u001b[39m\u001b[39mprocess \u001b[39m\u001b[39m%d\u001b[39;00m\u001b[39m terminated with exit code \u001b[39m\u001b[39m%d\u001b[39;00m\u001b[39m\"\u001b[39m \u001b[39m%\u001b[39m\n\u001b[1;32m    151\u001b[0m             (error_index, exitcode),\n\u001b[1;32m    152\u001b[0m             error_index\u001b[39m=\u001b[39merror_index,\n\u001b[1;32m    153\u001b[0m             error_pid\u001b[39m=\u001b[39mfailed_process\u001b[39m.\u001b[39mpid,\n\u001b[1;32m    154\u001b[0m             exit_code\u001b[39m=\u001b[39mexitcode\n\u001b[1;32m    155\u001b[0m         )\n\u001b[1;32m    157\u001b[0m original_trace \u001b[39m=\u001b[39m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39merror_queues[error_index]\u001b[39m.\u001b[39mget()\n\u001b[1;32m    158\u001b[0m msg \u001b[39m=\u001b[39m \u001b[39m\"\u001b[39m\u001b[39m\\n\u001b[39;00m\u001b[39m\\n\u001b[39;00m\u001b[39m-- Process \u001b[39m\u001b[39m%d\u001b[39;00m\u001b[39m terminated with the following error:\u001b[39m\u001b[39m\\n\u001b[39;00m\u001b[39m\"\u001b[39m \u001b[39m%\u001b[39m error_index\n",
+      "\u001b[0;31mProcessExitedException\u001b[0m: process 0 terminated with exit code 1"
+     ]
+    }
+   ],
+   "source": [
+    "def single_main(rank, world_size):\n",
+    "    config = TrainConfig()\n",
+    "    ddp_setup(rank, world_size)\n",
+    "    \n",
+    "    num_image_list = [100]#[200]#[1600,3200,6400,12800,25600]\n",
+    "    for i, num_image in enumerate(num_image_list):\n",
+    "        config.num_image = num_image\n",
+    "        # config.world_size = world_size\n",
+    "        \n",
+    "        ddpm21cm = DDPM21CM(config)\n",
+    "        print(f\" num_image = {ddpm21cm.config.num_image} \".center(50, '-'))\n",
+    "        print(f\"run_name = {ddpm21cm.config.run_name}\")\n",
+    "        ddpm21cm.train()\n",
+    "\n",
+    "        \n",
+    "if __name__ == \"__main__\":\n",
+    "    # torch.multiprocessing.set_start_method(\"spawn\")\n",
+    "    # args = (config, nn_model, ddpm, optimizer, dataloader, lr_scheduler)\n",
+    "    world_size = 1#torch.cuda.device_count()\n",
+    "\n",
+    "    mp.spawn(single_main, args=(world_size,), nprocs=world_size)\n",
+    "    # notebook_launcher(ddpm21cm.train, num_processes=1, mixed_precision='fp16')"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# torch.cuda.set_device(0)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "True\n",
+      "2\n",
+      "['__name__', '__doc__', '__package__', '__loader__', '__spec__', '__path__', '__file__', '__cached__', '__builtins__', '__annotations__', 'contextlib', 'os', 'torch', 'Device', 'traceback', 'warnings', 'threading', 'List', 'Optional', 'Tuple', 'Union', 'Any', '_utils', '_get_device_index', '_dummy_type', 'classproperty', 'graphs', 'CUDAGraph', 'graph_pool_handle', 'graph', 'make_graphed_callables', 'is_current_stream_capturing', 'streams', 'ExternalStream', 'Stream', 'Event', '_device', '_cudart', '_initialized', '_tls', '_initialization_lock', '_queued_calls', '_is_in_bad_fork', '_device_t', '_LazySeedTracker', '_lazy_seed_tracker', '_CudaDeviceProperties', 'has_magma', 'has_half', 'default_generators', 'is_available', 'is_bf16_supported', '_sleep', '_check_capability', '_check_cubins', 'is_initialized', '_lazy_call', 'DeferredCudaCallError', 'init', '_lazy_init', 'cudart', 'cudaStatus', 'CudaError', 'check_error', 'device', 'device_of', 'set_device', 'get_device_name', 'get_device_capability', 'get_device_properties', 'can_device_access_peer', 'StreamContext', 'stream', 'set_stream', 'device_count', 'get_arch_list', 'get_gencode_flags', 'current_device', 'synchronize', 'ipc_collect', 'current_stream', 'default_stream', 'current_blas_handle', 'set_sync_debug_mode', 'get_sync_debug_mode', 'memory_usage', 'utilization', 'memory', 'caching_allocator_alloc', 'caching_allocator_delete', 'set_per_process_memory_fraction', 'empty_cache', 'memory_stats', 'memory_stats_as_nested_dict', 'reset_accumulated_memory_stats', 'reset_peak_memory_stats', 'reset_max_memory_allocated', 'reset_max_memory_cached', 'memory_allocated', 'max_memory_allocated', 'memory_reserved', 'max_memory_reserved', 'memory_cached', 'max_memory_cached', 'memory_snapshot', 'memory_summary', 'list_gpu_processes', 'mem_get_info', 'random', 'get_rng_state', 'get_rng_state_all', 'set_rng_state', 'set_rng_state_all', 'manual_seed', 'manual_seed_all', 'seed', 'seed_all', 'initial_seed', '_lazy_new', '_CudaBase', 'ByteStorage', 'DoubleStorage', 'FloatStorage', 'HalfStorage', 'LongStorage', 'IntStorage', 'ShortStorage', 'CharStorage', 'BoolStorage', 'BFloat16Storage', 'ComplexDoubleStorage', 'ComplexFloatStorage', 'sparse', 'profiler', 'nvtx', 'amp', 'jiterator', 'ByteTensor', 'CharTensor', 'DoubleTensor', 'FloatTensor', 'IntTensor', 'LongTensor', 'ShortTensor', 'HalfTensor', 'BoolTensor', 'BFloat16Tensor', 'nccl', '_get_device_properties']\n"
+     ]
+    }
+   ],
+   "source": [
+    "print(torch.cuda.is_available())\n",
+    "print(torch.cuda.device_count())\n",
+    "print(torch.cuda.__dir__())"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 17,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "True\n",
+      "<class 'torch.cuda.device'>\n",
+      "Quadro RTX 6000\n",
+      "0\n",
+      "(7, 5)\n",
+      "_CudaDeviceProperties(name='Quadro RTX 6000', major=7, minor=5, total_memory=24212MB, multi_processor_count=72)\n"
+     ]
+    }
+   ],
+   "source": [
+    "print(torch.cuda.is_initialized())\n",
+    "print(torch.cuda.device)\n",
+    "print(torch.cuda.get_device_name())\n",
+    "print(torch.cuda.current_device())\n",
+    "print(torch.cuda.get_device_capability())\n",
+    "print(torch.cuda.get_device_properties(torch.cuda.device))\n",
+    "# print('here')\n",
+    "# print(torch.cuda.memory_usage())\n",
+    "# print(torch.cuda.utilization())\n",
+    "# print(torch.cuda.memory())\n",
+    "# print('here')\n",
+    "# print(torch.cuda.memory_summary())"
+   ]
+  },
+  {
+   "attachments": {},
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "# Sampling"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "if __name__ == \"__main__\":\n",
+    "    # num_image_list = [1600,3200,6400,12800,25600]\n",
+    "    num_image_list = [1000]\n",
+    "    # num_image_list = [3200,6400,12800,25600]\n",
+    "    # args = (config, nn_model, ddpm, optimizer, dataloader, lr_scheduler)\n",
+    "    repeat = 2\n",
+    "    config = TrainConfig()\n",
+    "    for i, num_image in enumerate(num_image_list):\n",
+    "        config.num_image = num_image\n",
+    "        ddpm21cm = DDPM21CM(config)\n",
+    "\n",
+    "        ddpm21cm.sample(f\"./outputs/model_state-N{num_image}\", params=torch.tensor([4.4, 131.341]), repeat=repeat)\n",
+    "\n",
+    "        # ddpm21cm.sample(f\"./outputs/model_state-N{num_image}\", params=torch.tensor((5.6, 19.037)), repeat=repeat)\n",
+    "\n",
+    "        # ddpm21cm.sample(f\"./outputs/model_state-N{num_image}\", params=torch.tensor((4.699, 30)), repeat=repeat)\n",
+    "\n",
+    "        # ddpm21cm.sample(f\"./outputs/model_state-N{num_image}\", params=torch.tensor((5.477, 200)), repeat=repeat)\n",
+    "\n",
+    "        # ddpm21cm.sample(f\"./outputs/model_state-N{num_image}\", params=torch.tensor((4.8, 131.341)), repeat=repeat)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 13,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# ls -lth outputs | head"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "def plot_grid(samples, c=None, row=1, col=2):\n",
+    "    print(\"samples.shape =\", samples.shape)\n",
+    "    for j in range(samples.shape[4]):\n",
+    "        plt.figure(figsize = (12,6), dpi=400)\n",
+    "        for i in range(len(samples)):\n",
+    "            plt.subplot(row,col,i+1)\n",
+    "            plt.imshow(samples[i,0,:,:,j], cmap='gray')#, vmin=-1, vmax=1)\n",
+    "            plt.xticks([])\n",
+    "            plt.yticks([])\n",
+    "        # plt.suptitle(f\"ION_Tvir_MIN = {c[0][0]}, HII_EFF_FACTOR = {c[0][1]}\")\n",
+    "            # plt.show()\n",
+    "        # plt.suptitle('simulations')\n",
+    "        plt.tight_layout()\n",
+    "        plt.subplots_adjust(wspace=0, hspace=0)\n",
+    "        plt.savefig(f\"test3D-{j:03d}.png\")\n",
+    "        plt.close()\n",
+    "        # plt.show()\n",
+    "    \n",
+    "data = np.load(\"outputs/Tvir4.400000095367432-zeta131.34100341796875-N1000.npy\")\n",
+    "# print(data.shape)\n",
+    "plot_grid(data)\n",
+    "# plt.imshow(data)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# config = TrainConfig()\n",
+    "# def plot(filename, row=4, col=6):\n",
+    "#     samples = np.load(filename)\n",
+    "#     params = filename.split('guide_w')[-1][:-4]\n",
+    "#     print(\"plotting\", samples.shape, params)\n",
+    "#     plt.figure(figsize = (8,8))\n",
+    "#     for i in range(24):\n",
+    "#         plt.subplot(row,col,i+1)\n",
+    "#         plt.imshow(samples[i,0,:,:], cmap='gray')#, vmin=-1, vmax=1)\n",
+    "#         plt.xticks([])\n",
+    "#         plt.yticks([])\n",
+    "#         # plt.show()\n",
+    "#     plt.suptitle(params)\n",
+    "#     plt.tight_layout()\n",
+    "#     plt.subplots_adjust(wspace=0, hspace=0) \n",
+    "#     plt.show()\n",
+    "#     # plt.savefig('outputs/'+params+'.png')\n",
+    "#     # plt.close()\n",
+    "#     # plt.imshow(images[0,0])\n",
+    "#     # plt.show()"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 1,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# import torch\n",
+    "# print(torch.__version__)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 9,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# import torch\n",
+    "# import os\n",
+    "\n",
+    "# def compare_models(num_gpus):\n",
+    "#     model_states = []\n",
+    "    \n",
+    "#     for gpu_id in range(num_gpus):\n",
+    "#         filename = f\"outputs/model_state-N40-device{gpu_id}\"\n",
+    "#         if os.path.exists(filename):\n",
+    "#             state_dict = torch.load(filename, map_location='cpu')\n",
+    "#             model_states.append(state_dict)\n",
+    "#             print(filename)\n",
+    "#         else:\n",
+    "#             print(f\"File {filename} not found!\")\n",
+    "#             return False\n",
+    "    \n",
+    "#     # Compare all model state_dicts\n",
+    "#     print(\"len(model_states) =\", len(model_states))\n",
+    "#     base_state = model_states[0]\n",
+    "#     for state in model_states[1:]:\n",
+    "#         for key in base_state.keys():\n",
+    "#             # print(key, base_state[key], state[key])\n",
+    "#             print(key)\n",
+    "#         print(\"epoch\", base_state['epoch'], state['epoch'])\n",
+    "\n",
+    "#         print(base_state['unet_state_dict'].keys())\n",
+    "#         for key in base_state['unet_state_dict']:\n",
+    "#             # print(key)\n",
+    "#             if not torch.equal(base_state['unet_state_dict'][key], state['unet_state_dict'][key]):\n",
+    "#                 print(\"different\")\n",
+    "#                 return \n",
+    "#             # else:\n",
+    "#         print(\"exactly same\")\n",
+    "\n",
+    "#             # if key == 'epoch':\n",
+    "#             #     print(base_state[key], state[key])\n",
+    "#             # else:\n",
+    "#             #     print(base_state[key], state[key])\n",
+    "#             #     if not torch.equal(base_state[key], state[key]):\n",
+    "#             #     # if not (base_state[key] == state[key]):\n",
+    "#             #         print(f\"Mismatch found in parameter {key}\")\n",
+    "#             #         return False\n",
+    "    \n",
+    "#     # print(\"All models are identical!\")\n",
+    "#     # return True\n",
+    "\n",
+    "# if __name__ == \"__main__\":\n",
+    "#     # epoch_to_check = 0  # specify the epoch you want to check\n",
+    "#     num_gpus = torch.cuda.device_count()  # specify the number of GPUs used in training\n",
+    "#     compare_models(num_gpus)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 6,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import numpy as np\n",
+    "test = np.random.normal(0,1,(800,1,64,64,512))"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 7,
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "12.5"
+      ]
+     },
+     "execution_count": 7,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "(test.itemsize*test.size) / 1024/1024/1024"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 8,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "del test"
+   ]
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "base",
+   "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": "3.9.19"
+  },
+  "orig_nbformat": 4
+ },
+ "nbformat": 4,
+ "nbformat_minor": 2
+}

diffusion.py

@@ -239,7 +239,7 @@ class TrainConfig:
     stride = (2,2) if dim == 2 else (2,2,4)
     num_image = 1000#2000#20000#15000#7000#25600#3000#10000#1000#10000#5000#2560#800#2560
     batch_size = 10#1#2#50#20#2#100 # 10
-    n_epoch = 5#4# 10#50#20#20#2#5#25 # 120
+    n_epoch = 10#5#4# 10#50#20#20#2#5#25 # 120
     HII_DIM = 64
     num_redshift = 512#128#64#512#256#256#64#512#128
     channel = 1
@@ -266,7 +266,7 @@ class TrainConfig:
     # seed = 0
     # save_dir = './outputs/'
 
-    save_period = 1#np.infty#.1 # the period of sampling
+    save_period = np.infty#.1 # the period of sampling
     # general parameters for the name and logger    
     # device = "cuda" if torch.cuda.is_available() else "cpu"
     lrate = 1e-4
@@ -580,13 +580,16 @@ class DDPM21CM:
         # else:
         return x_last
 # %%
+
+num_train_image_list = [5000]
+
 def train(rank, world_size):
     config = TrainConfig()
     config.world_size = world_size
 
     ddp_setup(rank, world_size)
     
-    num_train_image_list = [200]#[3200]#[200]#[1600,3200,6400,12800,25600]
+    #[3200]#[200]#[1600,3200,6400,12800,25600]
     for i, num_image in enumerate(num_train_image_list):
         config.num_image = num_image
         # config.world_size = world_size
@@ -607,20 +610,6 @@ if __name__ == "__main__":
     mp.spawn(train, args=(world_size,), nprocs=world_size, join=True)
     # notebook_launcher(ddpm21cm.train, num_processes=1, mixed_precision='fp16')
 
-# %%
-# print("torch.cuda.is_initialized() =", torch.cuda.is_initialized())
-# print("torch.cuda.get_device_name() =", torch.cuda.get_device_name())
-# print("torch.cuda.current_device() =", torch.cuda.current_device())
-# print("torch.cuda.get_device_capability() =", torch.cuda.get_device_capability())
-# print("torch.cuda.get_device_properties(torch.cuda.device) =", torch.cuda.get_device_properties(torch.cuda.device))
-# print('here')
-# print(torch.cuda.memory_usage())
-# print(torch.cuda.utilization())
-# print(torch.cuda.memory())
-# print('here')
-# print(torch.cuda.memory_summary())
-# %% [markdown]
-# # Sampling
 
 # %%
 
@@ -677,7 +666,7 @@ if __name__ == "__main__":
     world_size = torch.cuda.device_count()
     print(f" sampling, world_size = {world_size} ".center(100,'-'))
     # num_train_image_list = [1600,3200,6400,12800,25600]
-    num_train_image_list = [200]
+    # num_train_image_list = [5000]
     num_new_img_per_gpu = 40
     max_num_img_per_gpu = 20