Flax: avoid recompilation when params change (huggingface#1096)

* Do not recompile when guidance_scale changes.

* Remove debug for simplicity.

* make style

* Make guidance_scale an array.

* Make DEBUG a constant to avoid passing it down.

* Add comments for clarification.

Author: pcuenca

src/diffusers/pipelines/stable_diffusion/pipeline_flax_stable_diffusion.py

@@ -42,6 +42,9 @@
 
 logger = logging.get_logger(__name__)  # pylint: disable=invalid-name
 
+# Set to True to use python for loop instead of jax.fori_loop for easier debugging
+DEBUG = False
+
 
 class FlaxStableDiffusionPipeline(FlaxDiffusionPipeline):
     r"""
@@ -187,7 +190,6 @@ def _generate(
         width: Optional[int] = None,
         guidance_scale: float = 7.5,
         latents: Optional[jnp.array] = None,
-        debug: bool = False,
         neg_prompt_ids: jnp.array = None,
     ):
         # 0. Default height and width to unet
@@ -260,8 +262,7 @@ def loop_body(step, args):
 
         # scale the initial noise by the standard deviation required by the scheduler
         latents = latents * self.scheduler.init_noise_sigma
-
-        if debug:
+        if DEBUG:
             # run with python for loop
             for i in range(num_inference_steps):
                 latents, scheduler_state = loop_body(i, (latents, scheduler_state))
@@ -283,11 +284,10 @@ def __call__(
         num_inference_steps: int = 50,
         height: Optional[int] = None,
         width: Optional[int] = None,
-        guidance_scale: float = 7.5,
+        guidance_scale: Union[float, jnp.array] = 7.5,
         latents: jnp.array = None,
         return_dict: bool = True,
         jit: bool = False,
-        debug: bool = False,
         neg_prompt_ids: jnp.array = None,
     ):
         r"""
@@ -334,6 +334,14 @@ def __call__(
         height = height or self.unet.config.sample_size * self.vae_scale_factor
         width = width or self.unet.config.sample_size * self.vae_scale_factor
 
+        if isinstance(guidance_scale, float):
+            # Convert to a tensor so each device gets a copy. Follow the prompt_ids for
+            # shape information, as they may be sharded (when `jit` is `True`), or not.
+            guidance_scale = jnp.array([guidance_scale] * prompt_ids.shape[0])
+            if len(prompt_ids.shape) > 2:
+                # Assume sharded
+                guidance_scale = guidance_scale.reshape(prompt_ids.shape[:2])
+
         if jit:
             images = _p_generate(
                 self,
@@ -345,7 +353,6 @@ def __call__(
                 width,
                 guidance_scale,
                 latents,
-                debug,
                 neg_prompt_ids,
             )
         else:
@@ -358,7 +365,6 @@ def __call__(
                 width,
                 guidance_scale,
                 latents,
-                debug,
                 neg_prompt_ids,
             )
 
@@ -388,8 +394,13 @@ def __call__(
         return FlaxStableDiffusionPipelineOutput(images=images, nsfw_content_detected=has_nsfw_concept)
 
 
-# TODO: maybe use a config dict instead of so many static argnums
-@partial(jax.pmap, static_broadcasted_argnums=(0, 4, 5, 6, 7, 9))
+# Static argnums are pipe, num_inference_steps, height, width. A change would trigger recompilation.
+# Non-static args are (sharded) input tensors mapped over their first dimension (hence, `0`).
+@partial(
+    jax.pmap,
+    in_axes=(None, 0, 0, 0, None, None, None, 0, 0, 0),
+    static_broadcasted_argnums=(0, 4, 5, 6),
+)
 def _p_generate(
     pipe,
     prompt_ids,
@@ -400,7 +411,6 @@ def _p_generate(
     width,
     guidance_scale,
     latents,
-    debug,
     neg_prompt_ids,
 ):
     return pipe._generate(
@@ -412,7 +422,6 @@ def _p_generate(
         width,
         guidance_scale,
         latents,
-        debug,
         neg_prompt_ids,
     )