made stable diffusin buildable

Author: divamgupta

.DS_Store

@@ -1,3 +1,7 @@
+HF_weights/
+models/
+outputs/
+
 # Byte-compiled / optimized / DLL files
 __pycache__/
 *.py[cod]

stable-diffusion/.gitignore

@@ -0,0 +1,9 @@
+pyinstaller txt2img.py --collect-all huggingface_hub --noconfirm --clean --collect-all tqdm --collect-all regex --collect-all requests --collect-all packaging --collect-all filelock --collect-all numpy --collect-all tokenizers --collect-all cv2
+
+mkdir dist/txt2img/clip
+cp src/clip/clip/bpe_simple_vocab_16e6.txt.gz dist/txt2img/clip/bpe_simple_vocab_16e6.txt.gz
+cp -r configs dist/txt2img/configs
+cp -r HF_weights dist/txt2img/HF_weights
+mkdir dist/txt2img/models
+
+echo "Right now you gotta copy the weights"

stable-diffusion/build.sh

@@ -7,6 +7,7 @@
 
 from ldm.modules.diffusionmodules.util import make_ddim_sampling_parameters, make_ddim_timesteps, noise_like, \
  extract_into_tensor
+from ...utils.stdin_input import is_avail, get_input
 
 
 class DDIMSampler(object):
@@ -144,6 +145,13 @@ def ddim_sampling(self, cond, shape,
  iterator = tqdm(time_range, desc='DDIM Sampler', total=total_steps)
 
  for i, step in enumerate(iterator):
+
+ if is_avail():
+ if "__stop__" in get_input():
+ return None , None
+ else:
+ print("boooo")
+ 
  index = total_steps - i - 1
  ts = torch.full((b,), step, device=device, dtype=torch.long)
 

stable-diffusion/ldm/models/diffusion/ddim.py

@@ -6,6 +6,7 @@
 from functools import partial
 
 from ldm.modules.diffusionmodules.util import make_ddim_sampling_parameters, make_ddim_timesteps, noise_like
+from ...utils.stdin_input import is_avail, get_input
 
 
 class PLMSSampler(object):
@@ -146,6 +147,14 @@ def plms_sampling(self, cond, shape,
  old_eps = []
 
  for i, step in enumerate(iterator):
+
+ if is_avail():
+ if "__stop__" in get_input():
+ return None , None
+ else:
+ print("boooo")
+
+
  index = total_steps - i - 1
  ts = torch.full((b,), step, device=device, dtype=torch.long)
  ts_next = torch.full((b,), time_range[min(i + 1, len(time_range) - 1)], device=device, dtype=torch.long)

stable-diffusion/ldm/models/diffusion/plms.py

@@ -4,8 +4,8 @@
 import clip
 from einops import rearrange, repeat
 from transformers import CLIPTokenizer, CLIPTextModel
-import kornia
-
+# import kornia
+import os
 from ldm.modules.x_transformer import Encoder, TransformerWrapper # TODO: can we directly rely on lucidrains code and simply add this as a reuirement? --> test
 
 
@@ -147,8 +147,21 @@ class FrozenCLIPEmbedder(AbstractEncoder):
  """Uses the CLIP transformer encoder for text (from Hugging Face)"""
  def __init__(self, version="openai/clip-vit-large-patch14", device=get_default_device_type(), max_length=77):
  super().__init__()
- self.tokenizer = CLIPTokenizer.from_pretrained(version)
- self.transformer = CLIPTextModel.from_pretrained(version)
+
+ tokenizer_path = os.path.abspath(
+ os.path.join(
+ os.path.dirname(__file__),
+ ".." , '..' , '..' , 'HF_weights',
+ "clip_tokenizer"))
+
+ transformer_path = os.path.abspath(
+ os.path.join(
+ os.path.dirname(__file__),
+ ".." , '..' , '..' , 'HF_weights',
+ "clip_transformer"))
+
+ self.tokenizer = CLIPTokenizer.from_pretrained(tokenizer_path) #dg load from HF
+ self.transformer = CLIPTextModel.from_pretrained(transformer_path) #dg load from HF
  self.device = device
  self.max_length = max_length
  self.freeze()

stable-diffusion/ldm/modules/encoders/modules.py

@@ -0,0 +1,63 @@
+import threading
+import queue
+import sys
+import platform
+import select
+
+
+if platform.system() == "Windows":
+
+ # Due to limitations of windows, we spawn a new thread for stdin. The thread collects the stdin inputs and appends them in list. 
+
+ input_queue = queue.Queue()
+ evt = threading.Event()
+ print("starting the input thread!")
+
+ def input_collector_worker(input_queue):
+ while True:
+ input_queue.put(input())
+ evt.set()
+
+ input_thread = threading.Thread(
+ target=input_collector_worker, args=(
+ input_queue,))
+ input_thread.daemon = True
+ input_thread.start()
+
+ def is_avail():
+ """To check in a non blocking way if there is some input available.
+ Returns:
+ bool: If input is avaialble
+ """
+ return not input_queue.empty()
+
+ def get_input():
+ """To get the input in a blocking way.
+ Returns:
+ str: The input
+ """
+ evt.clear()
+
+ if is_avail():
+ return input_queue.get()
+
+ evt.wait()
+
+ assert is_avail()
+ return get_input()
+
+
+else:
+
+ # for unix we use select to see if stdin is there or not
+
+ print("non threaded input for unix systems!")
+
+ def get_input():
+ return input()
+
+ def is_avail():
+ if select.select([sys.stdin, ], [], [], 0.0)[0]:
+ return True
+ else:
+ return False