Minor improvements to FLUX workflow.

invokeai/app/invocations/flux_text_to_image.py

@@ -33,7 +33,7 @@ class FluxTextToImageInvocation(BaseInvocation, WithMetadata, WithBoard):
 
     model: TFluxModelKeys = InputField(description="The FLUX model to use for text-to-image generation.")
     use_8bit: bool = InputField(
-        default=False, description="Whether to quantize the T5 model and transformer model to 8-bit precision."
+        default=False, description="Whether to quantize the transformer model to 8-bit precision."
     )
     positive_prompt: str = InputField(description="Positive prompt for text-to-image generation.")
     width: int = InputField(default=1024, multiple_of=16, description="Width of the generated image.")
@@ -56,7 +56,7 @@ class FluxTextToImageInvocation(BaseInvocation, WithMetadata, WithBoard):
         return ImageOutput.build(image_dto)
 
     def _encode_prompt(self, context: InvocationContext, flux_model_dir: Path) -> tuple[torch.Tensor, torch.Tensor]:
-        # Determine the T5 max sequence lenght based on the model.
+        # Determine the T5 max sequence length based on the model.
         if self.model == "flux-schnell":
             max_seq_len = 256
         # elif self.model == "flux-dev":
@@ -118,7 +118,9 @@ class FluxTextToImageInvocation(BaseInvocation, WithMetadata, WithBoard):
     ):
         scheduler = FlowMatchEulerDiscreteScheduler.from_pretrained(flux_model_dir / "scheduler", local_files_only=True)
 
-        # HACK(ryand): Manually empty the cache.
+        # HACK(ryand): Manually empty the cache. Currently we don't check the size of the model before loading it from
+        # disk. Since the transformer model is large (24GB), there's a good chance that it will OOM on 32GB RAM systems
+        # if the cache is not empty.
         context.models._services.model_manager.load.ram_cache.make_room(24 * 2**30)
 
         transformer_path = flux_model_dir / "transformer"
@@ -137,7 +139,7 @@ class FluxTextToImageInvocation(BaseInvocation, WithMetadata, WithBoard):
                 transformer=transformer,
             )
 
-            return flux_pipeline_with_transformer(
+            latents = flux_pipeline_with_transformer(
                 height=self.height,
                 width=self.width,
                 num_inference_steps=self.num_steps,
@@ -149,6 +151,9 @@ class FluxTextToImageInvocation(BaseInvocation, WithMetadata, WithBoard):
                 return_dict=False,
             )[0]
 
+        assert isinstance(latents, torch.Tensor)
+        return latents
+
     def _run_vae_decoding(
         self,
         context: InvocationContext,
@@ -201,9 +206,14 @@ class FluxTextToImageInvocation(BaseInvocation, WithMetadata, WithBoard):
             model_8bit_map_path = model_8bit_path / "quantization_map.json"
             if model_8bit_path.exists():
                 # The quantized model exists, load it.
-                with torch.device("meta"):
-                    model = FluxTransformer2DModel.from_pretrained(path, local_files_only=True)
+                # TODO(ryand): Make loading from quantized model work properly.
+                # Reference: https://gist.github.com/AmericanPresidentJimmyCarter/873985638e1f3541ba8b00137e7dacd9?permalink_comment_id=5141210#gistcomment-5141210
+                model = FluxTransformer2DModel.from_pretrained(
+                    path,
+                    local_files_only=True,
+                )
                 assert isinstance(model, FluxTransformer2DModel)
+                model = model.to(device=torch.device("meta"))
 
                 state_dict = load_file(model_8bit_weights_path)
                 with open(model_8bit_map_path, "r") as f:
@@ -211,6 +221,9 @@ class FluxTextToImageInvocation(BaseInvocation, WithMetadata, WithBoard):
                 requantize(model=model, state_dict=state_dict, quantization_map=quant_map)
             else:
                 # The quantized model does not exist yet, quantize and save it.
+                # TODO(ryand): Loading in float16 and then quantizing seems to result in NaNs. In order to run this on
+                # GPUs that don't support bfloat16, we would need to host the quantized model instead of generating it
+                # here.
                 model = FluxTransformer2DModel.from_pretrained(path, local_files_only=True, torch_dtype=torch.bfloat16)
                 assert isinstance(model, FluxTransformer2DModel)
 
@@ -222,9 +235,7 @@ class FluxTextToImageInvocation(BaseInvocation, WithMetadata, WithBoard):
                 with open(model_8bit_map_path, "w") as f:
                     json.dump(quantization_map(model), f)
         else:
-            model = FluxTransformer2DModel.from_pretrained(
-                path, local_files_only=True, torch_dtype=TorchDevice.choose_torch_dtype()
-            )
+            model = FluxTransformer2DModel.from_pretrained(path, local_files_only=True, torch_dtype=torch.bfloat16)
 
         assert isinstance(model, FluxTransformer2DModel)
         return model