Switch from no_grad to inference mode for scripts

scripts/gradio/depth2img.py

@@ -64,8 +64,7 @@ def paint(sampler, image, prompt, t_enc, seed, scale, num_samples=1, callback=No
     wm_encoder = WatermarkEncoder()
     wm_encoder.set_watermark('bytes', wm.encode('utf-8'))
 
-    with torch.no_grad(),\
+    with torch.inference_mode(), torch.autocast("cuda"):
-            torch.autocast("cuda"):
         batch = make_batch_sd(
             image, txt=prompt, device=device, num_samples=num_samples)
         z = model.get_first_stage_encoding(model.encode_first_stage(

scripts/gradio/inpainting.py

@@ -81,8 +81,7 @@ def inpaint(sampler, image, mask, prompt, seed, scale, ddim_steps, num_samples=1
     start_code = torch.from_numpy(start_code).to(
         device=device, dtype=torch.float32)
 
-    with torch.no_grad(), \
+    with torch.inference_mode(), torch.autocast("cuda"):
-            torch.autocast("cuda"):
         batch = make_batch_sd(image, mask, txt=prompt,
                               device=device, num_samples=num_samples)
 

scripts/gradio/superresolution.py

@@ -67,8 +67,8 @@ def paint(sampler, image, prompt, seed, scale, h, w, steps, num_samples=1, callb
     wm = "SDV2"
     wm_encoder = WatermarkEncoder()
     wm_encoder.set_watermark('bytes', wm.encode('utf-8'))
-    with torch.no_grad(),\
+    
-            torch.autocast("cuda"):
+    with torch.inference_mode(), torch.autocast("cuda"):
         batch = make_batch_sd(
             image, txt=prompt, device=device, num_samples=num_samples)
         c = model.cond_stage_model.encode(batch["txt"])
@@ -112,8 +112,10 @@ def paint(sampler, image, prompt, seed, scale, h, w, steps, num_samples=1, callb
             x_T=start_code,
             callback=callback
         )
-    with torch.no_grad():
+    
+    with torch.inference_mode():
         x_samples_ddim = model.decode_first_stage(samples)
+    
     result = torch.clamp((x_samples_ddim + 1.0) / 2.0, min=0.0, max=1.0)
     result = result.cpu().numpy().transpose(0, 2, 3, 1) * 255
     return [put_watermark(Image.fromarray(img.astype(np.uint8)), wm_encoder) for img in result]

scripts/img2img.py

@@ -230,7 +230,8 @@ def main():
     print(f"target t_enc is {t_enc} steps")
 
     precision_scope = autocast if opt.precision == "autocast" else nullcontext
-    with torch.no_grad():
+    
+    with torch.inference_mode():
         with precision_scope("cuda"):
             with model.ema_scope():
                 all_samples = list()

scripts/streamlit/depth2img.py

@@ -61,8 +61,7 @@ def paint(sampler, image, prompt, t_enc, seed, scale, num_samples=1, callback=No
     wm_encoder = WatermarkEncoder()
     wm_encoder.set_watermark('bytes', wm.encode('utf-8'))
 
-    with torch.no_grad(),\
+    with torch.inference_mode(), torch.autocast("cuda"):
-            torch.autocast("cuda"):
         batch = make_batch_sd(image, txt=prompt, device=device, num_samples=num_samples)
         z = model.get_first_stage_encoding(model.encode_first_stage(batch[model.first_stage_key]))  # move to latent space
         c = model.cond_stage_model.encode(batch["txt"])

scripts/streamlit/inpainting.py

@@ -79,8 +79,7 @@ def inpaint(sampler, image, mask, prompt, seed, scale, ddim_steps, num_samples=1
     start_code = prng.randn(num_samples, 4, h // 8, w // 8)
     start_code = torch.from_numpy(start_code).to(device=device, dtype=torch.float32)
 
-    with torch.no_grad(), \
+    with torch.inference_mode(), torch.autocast("cuda"):
-            torch.autocast("cuda"):
             batch = make_batch_sd(image, mask, txt=prompt, device=device, num_samples=num_samples)
 
             c = model.cond_stage_model.encode(batch["txt"])

scripts/streamlit/superresolution.py

@@ -64,8 +64,8 @@ def paint(sampler, image, prompt, seed, scale, h, w, steps, num_samples=1, callb
     wm = "SDV2"
     wm_encoder = WatermarkEncoder()
     wm_encoder.set_watermark('bytes', wm.encode('utf-8'))
-    with torch.no_grad(),\
+    
-            torch.autocast("cuda"):
+    with torch.inference_mode(), torch.autocast("cuda"):
         batch = make_batch_sd(image, txt=prompt, device=device, num_samples=num_samples)
         c = model.cond_stage_model.encode(batch["txt"])
         c_cat = list()
@@ -105,8 +105,10 @@ def paint(sampler, image, prompt, seed, scale, h, w, steps, num_samples=1, callb
             x_T=start_code,
             callback=callback
         )
-    with torch.no_grad():
+    
+    with torch.inference_mode():
         x_samples_ddim = model.decode_first_stage(samples)
+    
     result = torch.clamp((x_samples_ddim + 1.0) / 2.0, min=0.0, max=1.0)
     result = result.cpu().numpy().transpose(0, 2, 3, 1) * 255
     st.text(f"upscaled image shape: {result.shape}")

scripts/txt2img.py

@@ -232,9 +232,8 @@ def main(opt):
         start_code = torch.randn([opt.n_samples, opt.C, opt.H // opt.f, opt.W // opt.f], device=device)
 
     precision_scope = autocast if opt.precision == "autocast" else nullcontext
-    with torch.no_grad(), \
+    
-        precision_scope("cuda"), \
+    with torch.inference_mode(), precision_scope("cuda"), model.ema_scope():
-        model.ema_scope():
             all_samples = list()
             for n in trange(opt.n_iter, desc="Sampling"):
                 for prompts in tqdm(data, desc="data"):