ai-simple-engine-runtime-diffusers 0.0.4__tar.gz → 0.0.6__tar.gz

{ai_simple_engine_runtime_diffusers-0.0.4 → ai_simple_engine_runtime_diffusers-0.0.6}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: ai-simple-engine-runtime-diffusers
-Version: 0.0.4
+Version: 0.0.6
 Summary: AI Simple Engine Runtime Diffusers Module
 License-File: LICENSE
 Author: danialcala94
@@ -11,7 +11,7 @@ Classifier: Programming Language :: Python :: 3.10
 Classifier: Programming Language :: Python :: 3.11
 Requires-Dist: ai-simple-engine (>=0.2.9,<1.0.0)
 Requires-Dist: ai-simple-engine-common (>=0.0.3,<1.0.0)
-Requires-Dist: ai-simple-engine-diffusion (>=0.0.4,<1.0.0)
+Requires-Dist: ai-simple-engine-diffusion (>=0.0.8,<1.0.0)
 Requires-Dist: diffusers (>=0.38.0,<9999.0.0)
 Requires-Dist: transformers (>=5.13.0,<9999.0.0)
 Description-Content-Type: text/markdown

{ai_simple_engine_runtime_diffusers-0.0.4 → ai_simple_engine_runtime_diffusers-0.0.6}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "ai-simple-engine-runtime-diffusers"
-version = "0.0.4"
+version = "0.0.6"
 description = "AI Simple Engine Runtime Diffusers Module"
 authors = [
     {name = "danialcala94", email = "danielalcalavalera@gmail.com"}
@@ -12,7 +12,7 @@ requires-python = ">=3.10,<3.12"
 # Mandatory
 ai-simple-engine = { version = ">=0.2.9,<1.0.0", optional = false }
 ai-simple-engine-common = { version = ">=0.0.3,<1.0.0", optional = false }
-ai-simple-engine-diffusion = { version = ">=0.0.4,<1.0.0", optional = false }
+ai-simple-engine-diffusion = { version = ">=0.0.8,<1.0.0", optional = false }
 transformers = { version = ">=5.13.0,<9999.0.0", optional = false }
 diffusers = { version = ">=0.38.0,<9999.0.0", optional = false }
 # Optional

{ai_simple_engine_runtime_diffusers-0.0.4 → ai_simple_engine_runtime_diffusers-0.0.6}/src/ai_simple_engine_runtime_diffusers/models/executor/diffusers_latent_diffusion_model_executor.py

@@ -26,7 +26,7 @@ class DiffusersLatentDiffusionModelExecutor (
         self,
         context: ExecutionContext,
         model: LoadedModel,
-    ):
+    ) -> DiffusersScheduler:
         scheduler_registry = context.services.get(DiffusersSchedulerRegistry)
 
         scheduler_class = scheduler_registry.resolve(model.info.scheduler.identifier)
@@ -47,7 +47,7 @@ class DiffusersLatentDiffusionModelExecutor (
         embeddings: PromptEmbeddings,
         timestep,
         guidance_scale
-    ):
+    ) -> torch.Tensor:
         runtime_model = model.instance
         unet = runtime_model.unet
         latent_input = torch.cat([latents, latents])
@@ -112,21 +112,66 @@ class DiffusersLatentDiffusionModelExecutor (
         *,
         model: LoadedModel,
         latents
-    ) -> Image:
+    ) -> torch.Tensor:
         runtime_model = model.instance
 
         vae = runtime_model.vae
 
+        # Scale based on the vae (mandatory)
         latents = latents / vae.config.scaling_factor
 
+        # Decode the image (mandatory)
         image = vae.decode(latents).sample
         """
         TODO: I don't know if this must be done always
         or depending or what, but I should pay attention
         and refactor it.
         """
-        image = (image / 2 + 0.5).clamp(0, 1)
+
+        """
+        The VAE of Stable Diffusion return in the [-1, 1]
+        so we need to transform to [0, 1]. This is
+        specific from the Stable Diffusion.
+        """
+        image = self._normalize_image(
+            image = image,
+            model = model
+        )
+        # image = (image / 2 + 0.5).clamp(0, 1)
+
+        return image
+
+        """
+        TODO: This is to make the image ready be read as a
+        real image and should be done here.
+        """
         image = image.cpu().permute(0, 2, 3, 1).float().detach().numpy()
         image = Image(image)
 
-        return image
+        return image
+    
+    # TODO: Maybe rename to '_normalize_tensor' (?)
+    def _normalize_image(
+        self,
+        image: torch.Tensor,
+        model: LoadedModel
+    ) -> torch.Tensor:
+        """
+        *For internal use only*
+
+        Normalize the `image` provided according to
+        the specifications in the information of the
+        `model` given.
+        """
+        minimum, maximum = model.info.tensor_output_spec.value_range
+
+        if (minimum, maximum) == (-1.0, 1.0):
+            image = (image + 1.0) / 2.0
+            # Previously:
+            # image = (image / 2 + 0.5)
+        elif (minimum, maximum) == (0.0, 1.0):
+            pass
+        else:
+            image = (image - minimum) / (maximum - minimum)
+
+        return image.clamp(0.0, 1.0)

{ai_simple_engine_runtime_diffusers-0.0.4 → ai_simple_engine_runtime_diffusers-0.0.6}/src/ai_simple_engine_runtime_diffusers/models/executor/noise/pytorch_noise_generator.py

@@ -6,8 +6,11 @@ import torch
 class TorchNoiseGenerator(
     NoiseGenerator
 ):
+    """
+    Class to generate random noise using `torch`.
+    """
 
-    def generate(
+    async def generate(
         self,
         *,
         shape,

{ai_simple_engine_runtime_diffusers-0.0.4 → ai_simple_engine_runtime_diffusers-0.0.6}/src/ai_simple_engine_runtime_diffusers/models/loader/latent_diffusion_model_loader.py

@@ -1,6 +1,7 @@
 from ai_simple_engine_runtime_diffusers.models.runtime.diffusers_latent_diffusion_model import DiffusersLatentDiffusionModel
 from ai_simple_engine_runtime_diffusers.models.runtime.utils import get_torch_dtype_for
 from ai_simple_engine_diffusion.model.info.diffusion_model_info import DiffusionModelInfo
+from ai_simple_engine_diffusion.model.info.tensor_output_spec import TensorOutputSpec
 from ai_simple_engine_diffusion.scheduler.spec.base import SchedulerSpec
 from ai_simple_engine.device.base import Device
 from ai_simple_engine.models.loaders.abstract import ModelLoader
@@ -84,7 +85,11 @@ class LatentDiffusionModelLoader(
             latent_channels = unet.config.in_channels,
             vae_scale_factor = 2 ** (len(vae.config.block_out_channels) - 1),
             # TODO: Make dynamic depending on the 'installed_model'
-            scheduler = self._default_scheduler(installed_model)
+            scheduler = self._default_scheduler(installed_model),
+            # TODO: Is this ok (?)
+            tensor_output_spec = TensorOutputSpec(
+                value_range = (-1.0, 1.0)
+            )
         )
 
         return LoadedModel(