Minor patches

seed_all.py

@@ -0,0 +1,19 @@
+# Zippy's seed_all function
+from numpy.random import MT19937, RandomState, SeedSequence
+import random, numpy as np, time
+import torch
+
+def seed_all(seed_value=None):
+    new_state = RandomState(MT19937(SeedSequence(int(time.time_ns()) % (2**32 - 1))))
+    np.random.set_state(new_state.get_state())
+    if seed_value is None:
+        seed_value = np.random.randint(1, 2**32 - 1)
+    random.seed(seed_value)  # Python
+    np.random.seed(seed_value)  # cpu vars
+    torch.manual_seed(seed_value)  # cpu  vars
+    if torch.cuda.is_available():
+        torch.cuda.manual_seed(seed_value)
+        torch.cuda.manual_seed_all(seed_value)  # gpu vars
+        torch.backends.cudnn.deterministic = True  # needed
+        torch.backends.cudnn.benchmark = False
+    return seed_value

simulacra.py

@@ -1064,12 +1064,12 @@ async def add(interaction: nextcord.Interaction):
         if (' ' + word.lower()) in interaction.message.content.lower():
             await interaction.send("'{}' is not an allowed word by the content filter".format(word))
             return
-    prompt = interaction.message.content.split(".add")[1].strip()
-    if len(prompt) > 246: # Must be able to fit _grid.png on end.
+    prompt = re.sub("\.add\s+", "", interaction.message.content).strip()
+    seed = generations.get_next_seed()
+    if len(prompt) > (255 - 12 - len(str(seed))): # Must be able to fit <seed>_<prompt>_grid.png so inference doesn't crash
         await interaction.send("The length of the prompt wouldn't fit on the "
-                               "filesystem. Please shorten it and try again.")
+                               "filesystem, current max length is {}. Please shorten it and try again.".format((255 - 12 - len(str(seed)))))
         return
-    seed = generations.get_next_seed()
     job = Job(prompt=prompt,
               cloob_checkpoint='cloob_laion_400m_vit_b_16_16_epochs',
               scale=secrets.choice([7,8,9,10]),

simulacra_imagen_sample.py

@@ -1,13 +1,13 @@
 import argparse, os, sys, glob
 import torch
-from pytorch_lightning import seed_everything
 import numpy as np
 from omegaconf import OmegaConf
 from PIL import Image
 from tqdm import tqdm, trange
 from itertools import islice
 from einops import rearrange
 from torchvision.utils import make_grid
+from seed_all import seed_all
 
 from ldm.util import instantiate_from_config
 from ldm.models.diffusion.ddim import DDIMSampler
@@ -148,16 +148,22 @@ def main(opt):
         default="logs/f8-kl-clip-encoder-256x256-run1/checkpoints/last.ckpt",
         help="path to checkpoint of model",
     )
-    #opt = parser.parse_args()
+
+    parser.add_argument(
+        "--seed",
+        type=int,
+        help="seed for seed_all",
+    )
+    # opt = parser.parse_args()
+    seed_all(opt.seed)
+
     opt.n_rows = 1
     opt.config = "stable-diffusion/configs/stable-diffusion/txt2img-multinode-clip-encoder-f16-768-laion-hr-inference.yaml"
     opt.ckpt = "f16-33k+12k-hr_pruned.ckpt"
     opt.dyn = None
     opt.C = 16
     opt.f = 16
 
-    seed_everything(opt.seed)
-
     config = OmegaConf.load(f"{opt.config}")
     model = load_model_from_config(config, f"{opt.ckpt}")
 
@@ -207,11 +213,11 @@ def main(opt):
                 outs = all_samples[0]
                 for index, out in enumerate(outs):
                     print(out.shape)
-                    outpath = str(opt.seed) + "_" + opt.prompt.replace(" ", "_").replace("/","_") + "_" + str(index + 1) + ".png"
+                    outpath = str(opt.scale) + "_" + str(opt.seed) + "_" + opt.prompt.replace(" ", "_").replace("/","_") + "_" + str(index + 1) + ".png"
                     out = 255. * rearrange(out.cpu().numpy(), 'c h w -> h w c')
                     Image.fromarray(out.astype(np.uint8)).save(outpath)
 
-                gridpath = str(opt.seed) + "_" + opt.prompt.replace(" ", "_").replace("/","_") + "_grid" + ".png"
+                gridpath = str(opt.scale) + "_" + opt.str(opt.seed) + "_" + opt.prompt.replace(" ", "_").replace("/","_") + "_grid" + ".png"
                 grid = torch.stack(all_samples, 0)
                 grid = rearrange(grid, 'n b c h w -> (n b) c h w')
                 grid = make_grid(grid, nrow=opt.n_samples)