cleanup before merge

modules/sd_hijack.py

@@ -42,10 +42,6 @@ def apply_optimizations():
         ldm.modules.attention.CrossAttention.forward = sd_hijack_optimizations.xformers_attention_forward
         ldm.modules.diffusionmodules.model.AttnBlock.forward = sd_hijack_optimizations.xformers_attnblock_forward
         optimization_method = 'xformers'
-    elif cmd_opts.opt_sdp_attention and (hasattr(torch.nn.functional, "scaled_dot_product_attention") and callable(getattr(torch.nn.functional, "scaled_dot_product_attention"))):
-        print("Applying scaled dot product cross attention optimization.")
-        ldm.modules.attention.CrossAttention.forward = sd_hijack_optimizations.scaled_dot_product_attention_forward
-        optimization_method = 'sdp'
     elif cmd_opts.opt_sub_quad_attention:
         print("Applying sub-quadratic cross attention optimization.")
         ldm.modules.attention.CrossAttention.forward = sd_hijack_optimizations.sub_quad_attention_forward

modules/sd_hijack_optimizations.py

@@ -346,48 +346,6 @@ def xformers_attention_forward(self, x, context=None, mask=None):
     out = rearrange(out, 'b n h d -> b n (h d)', h=h)
     return self.to_out(out)
 
-# Based on Diffusers usage of scaled dot product attention from https://github.com/huggingface/diffusers/blob/c7da8fd23359a22d0df2741688b5b4f33c26df21/src/diffusers/models/cross_attention.py
-# The scaled_dot_product_attention_forward function contains parts of code under Apache-2.0 license listed under Scaled Dot Product Attention in the Licenses section of the web UI interface
-def scaled_dot_product_attention_forward(self, x, context=None, mask=None):
-    batch_size, sequence_length, inner_dim = x.shape
-
-    if mask is not None:
-        mask = self.prepare_attention_mask(mask, sequence_length, batch_size)
-        mask = mask.view(batch_size, self.heads, -1, mask.shape[-1])
-
-    h = self.heads
-    q_in = self.to_q(x)
-    context = default(context, x)
-
-    context_k, context_v = hypernetwork.apply_hypernetworks(shared.loaded_hypernetworks, context)
-    k_in = self.to_k(context_k)
-    v_in = self.to_v(context_v)
-
-    head_dim = inner_dim // h
-    q = q_in.view(batch_size, -1, h, head_dim).transpose(1, 2)
-    k = k_in.view(batch_size, -1, h, head_dim).transpose(1, 2)
-    v = v_in.view(batch_size, -1, h, head_dim).transpose(1, 2)
-    
-    del q_in, k_in, v_in
-
-    dtype = q.dtype
-    if shared.opts.upcast_attn:
-        q, k = q.float(), k.float()
-
-    # the output of sdp = (batch, num_heads, seq_len, head_dim)
-    hidden_states = torch.nn.functional.scaled_dot_product_attention(
-        q, k, v, attn_mask=mask, dropout_p=0.0, is_causal=False
-    )
-
-    hidden_states = hidden_states.transpose(1, 2).reshape(batch_size, -1, h * head_dim)
-    hidden_states = hidden_states.to(dtype)
-
-    # linear proj
-    hidden_states = self.to_out[0](hidden_states)
-    # dropout
-    hidden_states = self.to_out[1](hidden_states)
-    return hidden_states
-
 def cross_attention_attnblock_forward(self, x):
         h_ = x
         h_ = self.norm(h_)

requirements_versions.txt

@@ -1,28 +1,30 @@
-accelerate==0.16.0
-basicsr==1.4.2
 blendmodes==2022
-clean-fid==0.1.35
-diffusers==0.12.1
-einops==0.4.1
-fastapi==0.90.1
+transformers==4.25.1
+accelerate==0.12.0
+basicsr==1.4.2
 gfpgan==1.3.8
-GitPython==3.1.27
 gradio==3.16.2
-inflection==0.5.1
-jsonmerge==1.9.0
-kornia==0.6.9
-lark==1.1.5
-numexpr==2.8.4
-omegaconf==2.3.0
-pandas==1.5.3
+numpy==1.23.3
 Pillow==9.4.0
-protobuf==3.20.3
-pytorch_lightning==1.7.7
 realesrgan==0.3.0
+torch
+omegaconf==2.2.3
+pytorch_lightning==1.7.6
+scikit-image==0.19.2
+fonts
+font-roboto
+timm==0.6.7
+piexif==1.1.3
+einops==0.4.1
+jsonmerge==1.8.0
+clean-fid==0.1.29
 resize-right==0.0.2
-safetensors==0.3.0
-scikit-image==0.19.3
-timm==0.6.12
 torchdiffeq==0.2.3
+kornia==0.6.7
+lark==1.1.2
+inflection==0.5.1
+GitPython==3.1.27
 torchsde==0.2.5
-transformers==4.26.1
+safetensors==0.2.7
+httpcore<=0.15
+fastapi==0.90.1

test/basic_features/txt2img_test.py

@@ -0,0 +1,80 @@
+import unittest
+import requests
+
+
+class TestTxt2ImgWorking(unittest.TestCase):
+    def setUp(self):
+        self.url_txt2img = "http://localhost:7860/sdapi/v1/txt2img"
+        self.simple_txt2img = {
+            "enable_hr": False,
+            "denoising_strength": 0,
+            "firstphase_width": 0,
+            "firstphase_height": 0,
+            "prompt": "example prompt",
+            "styles": [],
+            "seed": -1,
+            "subseed": -1,
+            "subseed_strength": 0,
+            "seed_resize_from_h": -1,
+            "seed_resize_from_w": -1,
+            "batch_size": 1,
+            "n_iter": 1,
+            "steps": 3,
+            "cfg_scale": 7,
+            "width": 64,
+            "height": 64,
+            "restore_faces": False,
+            "tiling": False,
+            "negative_prompt": "",
+            "eta": 0,
+            "s_churn": 0,
+            "s_tmax": 0,
+            "s_tmin": 0,
+            "s_noise": 1,
+            "sampler_index": "Euler a"
+        }
+
+    def test_txt2img_simple_performed(self):
+        self.assertEqual(requests.post(self.url_txt2img, json=self.simple_txt2img).status_code, 200)
+
+    def test_txt2img_with_negative_prompt_performed(self):
+        self.simple_txt2img["negative_prompt"] = "example negative prompt"
+        self.assertEqual(requests.post(self.url_txt2img, json=self.simple_txt2img).status_code, 200)
+
+    def test_txt2img_with_complex_prompt_performed(self):
+        self.simple_txt2img["prompt"] = "((emphasis)), (emphasis1:1.1), [to:1], [from::2], [from:to:0.3], [alt|alt1]"
+        self.assertEqual(requests.post(self.url_txt2img, json=self.simple_txt2img).status_code, 200)
+
+    def test_txt2img_not_square_image_performed(self):
+        self.simple_txt2img["height"] = 128
+        self.assertEqual(requests.post(self.url_txt2img, json=self.simple_txt2img).status_code, 200)
+
+    def test_txt2img_with_hrfix_performed(self):
+        self.simple_txt2img["enable_hr"] = True
+        self.assertEqual(requests.post(self.url_txt2img, json=self.simple_txt2img).status_code, 200)
+
+    def test_txt2img_with_tiling_performed(self):
+        self.simple_txt2img["tiling"] = True
+        self.assertEqual(requests.post(self.url_txt2img, json=self.simple_txt2img).status_code, 200)
+
+    def test_txt2img_with_restore_faces_performed(self):
+        self.simple_txt2img["restore_faces"] = True
+        self.assertEqual(requests.post(self.url_txt2img, json=self.simple_txt2img).status_code, 200)
+
+    def test_txt2img_with_vanilla_sampler_performed(self):
+        self.simple_txt2img["sampler_index"] = "PLMS"
+        self.assertEqual(requests.post(self.url_txt2img, json=self.simple_txt2img).status_code, 200)
+        self.simple_txt2img["sampler_index"] = "DDIM"
+        self.assertEqual(requests.post(self.url_txt2img, json=self.simple_txt2img).status_code, 200)
+
+    def test_txt2img_multiple_batches_performed(self):
+        self.simple_txt2img["n_iter"] = 2
+        self.assertEqual(requests.post(self.url_txt2img, json=self.simple_txt2img).status_code, 200)
+
+    def test_txt2img_batch_performed(self):
+        self.simple_txt2img["batch_size"] = 2
+        self.assertEqual(requests.post(self.url_txt2img, json=self.simple_txt2img).status_code, 200)
+
+
+if __name__ == "__main__":
+    unittest.main()