ford442 commited on
Commit
f8d4d9e
Β·
1 Parent(s): 68bb8b3

Update app.py

Browse files
Files changed (1) hide show
  1. app.py +12 -10
app.py CHANGED
@@ -18,17 +18,19 @@ from typing import Tuple
18
  from transformers import AutoTokenizer, AutoModelForCausalLM
19
  import paramiko
20
 
21
- os.system("chmod +x ./magma.sh")
22
- os.system("./magma.sh")
 
 
23
 
24
- torch.backends.cuda.matmul.allow_tf32 = True
25
- torch.backends.cuda.matmul.allow_bf16_reduced_precision_reduction = True
26
  torch.backends.cuda.matmul.allow_fp16_reduced_precision_reduction = False
27
- torch.backends.cudnn.allow_tf32 = True
28
  torch.backends.cudnn.deterministic = False
29
- torch.backends.cudnn.benchmark = True
30
  torch.backends.cuda.preferred_blas_library="cublas"
31
- torch.backends.cuda.preferred_linalg_library="magma"
32
 
33
  torch.set_float32_matmul_precision("highest")
34
 
@@ -123,7 +125,7 @@ def load_and_prepare_model(model_id):
123
  # Preload and compile both models
124
  models = {key: load_and_prepare_model(value) for key, value in MODEL_OPTIONS.items()}
125
 
126
- MAX_SEED = np.iinfo(np.int32).max
127
 
128
  def upload_to_ftp(filename):
129
  try:
@@ -168,7 +170,7 @@ def generate(
168
  global models
169
  pipe = models[model_choice]
170
  seed = int(randomize_seed_fn(seed, randomize_seed))
171
- generator = torch.Generator(device=device).manual_seed(seed)
172
 
173
  prompt, negative_prompt = apply_style(style_selection, prompt, negative_prompt)
174
 
@@ -220,7 +222,7 @@ def generate_cpu(
220
  pipe.to("cpu")
221
 
222
  seed = int(randomize_seed_fn(seed, randomize_seed))
223
- generator = torch.Generator(device=device).manual_seed(seed)
224
 
225
  prompt, negative_prompt = apply_style(style_selection, prompt, negative_prompt)
226
 
 
18
  from transformers import AutoTokenizer, AutoModelForCausalLM
19
  import paramiko
20
 
21
+ os.system("chmod +x ./cusparselt.sh")
22
+ os.system("./cusparselt.sh")
23
+ os.system("chmod +x ./cudnn.sh")
24
+ os.system("./cudnn.sh")
25
 
26
+ torch.backends.cuda.matmul.allow_tf32 = False
27
+ torch.backends.cuda.matmul.allow_bf16_reduced_precision_reduction = False
28
  torch.backends.cuda.matmul.allow_fp16_reduced_precision_reduction = False
29
+ torch.backends.cudnn.allow_tf32 = False
30
  torch.backends.cudnn.deterministic = False
31
+ torch.backends.cudnn.benchmark = False
32
  torch.backends.cuda.preferred_blas_library="cublas"
33
+ torch.backends.cuda.preferred_linalg_library="cusolver"
34
 
35
  torch.set_float32_matmul_precision("highest")
36
 
 
125
  # Preload and compile both models
126
  models = {key: load_and_prepare_model(value) for key, value in MODEL_OPTIONS.items()}
127
 
128
+ MAX_SEED = np.iinfo(np.int64).max
129
 
130
  def upload_to_ftp(filename):
131
  try:
 
170
  global models
171
  pipe = models[model_choice]
172
  seed = int(randomize_seed_fn(seed, randomize_seed))
173
+ generator = torch.Generator(device='cpu').manual_seed(seed)
174
 
175
  prompt, negative_prompt = apply_style(style_selection, prompt, negative_prompt)
176
 
 
222
  pipe.to("cpu")
223
 
224
  seed = int(randomize_seed_fn(seed, randomize_seed))
225
+ generator = torch.Generator(device='cpu').manual_seed(seed)
226
 
227
  prompt, negative_prompt = apply_style(style_selection, prompt, negative_prompt)
228