Diffusers, txt2img and img2img, make sure guidance scale defaults to …

…0 when num steps <=4 Signed-off-by: Raphael Glon <oOraph@users.noreply.github.com>
huggingface · Oct 25, 2024 · 18a07ce · 18a07ce
1 parent 4a49933
commit 18a07ce
Show file tree

Hide file tree

Showing 3 changed files with 16 additions and 3 deletions.
diff --git a/docker_images/diffusers/app/lora.py b/docker_images/diffusers/app/lora.py
@@ -101,9 +101,11 @@ def _load_textual_embeddings(self, adapter, model_data):
         if self._is_pivotal_tuning_lora(model_data):
             embedding_path = self._hub_repo_file(
                 repo_id=adapter,
-                filename="embeddings.safetensors"
-                if self._is_safetensors_pivotal(model_data)
-                else "embeddings.pti",
+                filename=(
+                    "embeddings.safetensors"
+                    if self._is_safetensors_pivotal(model_data)
+                    else "embeddings.pti"
+                ),
                 repo_type="model",
             )
 

diff --git a/docker_images/diffusers/app/pipelines/image_to_image.py b/docker_images/diffusers/app/pipelines/image_to_image.py
@@ -233,6 +233,12 @@ def _process_req(self, image, prompt, **kwargs):
                 "negative_prompt": kwargs.get("negative_prompt", None),
                 "guidance_scale": kwargs.get("guidance_scale", 7),
             }
+            if "guidance_scale" not in kwargs:
+                default_guidance_scale = os.getenv("DEFAULT_GUIDANCE_SCALE")
+                if default_guidance_scale:
+                    kwargs["guidance_scale"] = float(default_guidance_scale)
+                    prior_args["guidance_scale"] = float(default_guidance_scale)
+                # Else, don't specify anything, leave the default behaviour
             image_emb, zero_image_emb = self.prior(prompt, **prior_args).to_tuple()
             images = self.ldm(
                 prompt,

diff --git a/docker_images/diffusers/app/pipelines/text_to_image.py b/docker_images/diffusers/app/pipelines/text_to_image.py
@@ -169,6 +169,11 @@ def _process_req(self, inputs, **kwargs):
                 kwargs["num_inference_steps"] = 20
             # Else, don't specify anything, leave the default behaviour
 
+        if "guidance_scale" not in kwargs:
+            default_guidance_scale = os.getenv("DEFAULT_GUIDANCE_SCALE")
+            if default_guidance_scale:
+                kwargs["guidance_scale"] = float(default_guidance_scale)
+            # Else, don't specify anything, leave the default behaviour
         if "seed" in kwargs:
             seed = int(kwargs["seed"])
             generator = torch.Generator().manual_seed(seed)