Flax pipeline pndm (#583)

* WIP: flax FlaxDiffusionPipeline & FlaxStableDiffusionPipeline

* todo comment

* Fix imports

* Fix imports

* add dummies

* Fix empty init

* make pipeline work

* up

* Allow dtype to be overridden on model load.

This may be a temporary solution until #567 is addressed.

* Convert params to bfloat16 or fp16 after loading.

This deals with the weights, not the model.

* Use Flax schedulers (typing, docstring)

* PNDM: replace control flow with jax functions.

Otherwise jitting/parallelization don't work properly as they don't know
how to deal with traced objects.

I temporarily removed `step_prk`.

* Pass latents shape to scheduler set_timesteps()

PNDMScheduler uses it to reserve space, other schedulers will just
ignore it.

* Wrap model imports inside availability checks.

* Optionally return state in from_config.

Useful for Flax schedulers.

* Do not convert model weights to dtype.

* Re-enable PRK steps with functional implementation.

Values returned still not verified for correctness.

* Remove left over has_state var.

* make style

* Apply suggestion list -> tuple

Co-authored-by: Suraj Patil <[email protected]>

* Apply suggestion list -> tuple

Co-authored-by: Suraj Patil <[email protected]>

* Remove unused comments.

* Use zeros instead of empty.

Co-authored-by: Mishig Davaadorj <[email protected]>
Co-authored-by: Mishig Davaadorj <[email protected]>
Co-authored-by: Patrick von Platen <[email protected]>
Co-authored-by: Suraj Patil <[email protected]>

Author: 5 people

src/diffusers/pipelines/stable_diffusion/__init__.py

@@ -56,5 +56,6 @@ class FlaxStableDiffusionPipelineOutput(BaseOutput):
         images: Union[List[PIL.Image.Image], np.ndarray]
         nsfw_content_detected: List[bool]
 
+    from ...schedulers.scheduling_pndm_flax import PNDMSchedulerState
     from .pipeline_flax_stable_diffusion import FlaxStableDiffusionPipeline
     from .safety_checker_flax import FlaxStableDiffusionSafetyChecker

src/diffusers/pipelines/stable_diffusion/pipeline_flax_stable_diffusion.py

@@ -186,7 +186,9 @@ def loop_body(step, args):
             latents, scheduler_state = self.scheduler.step(scheduler_state, noise_pred, t, latents).to_tuple()
             return latents, scheduler_state
 
-        scheduler_state = self.scheduler.set_timesteps(params["scheduler"], num_inference_steps=num_inference_steps)
+        scheduler_state = self.scheduler.set_timesteps(
+            params["scheduler"], num_inference_steps=num_inference_steps, shape=latents.shape
+        )
 
         if debug:
             # run with python for loop

src/diffusers/schedulers/scheduling_pndm_flax.py

@@ -19,6 +19,7 @@
 from typing import Optional, Tuple, Union
 
 import flax
+import jax
 import jax.numpy as jnp
 
 from ..configuration_utils import ConfigMixin, register_to_config
@@ -155,7 +156,7 @@ def __init__(
     def create_state(self):
         return PNDMSchedulerState.create(num_train_timesteps=self.config.num_train_timesteps)
 
-    def set_timesteps(self, state: PNDMSchedulerState, num_inference_steps: int) -> PNDMSchedulerState:
+    def set_timesteps(self, state: PNDMSchedulerState, shape: Tuple, num_inference_steps: int) -> PNDMSchedulerState:
         """
         Sets the discrete timesteps used for the diffusion chain. Supporting function to be run before inference.
 
@@ -196,8 +197,11 @@ def set_timesteps(self, state: PNDMSchedulerState, num_inference_steps: int) ->
 
         return state.replace(
             timesteps=jnp.concatenate([state.prk_timesteps, state.plms_timesteps]).astype(jnp.int64),
-            ets=jnp.array([]),
             counter=0,
+            # Reserve space for the state variables
+            cur_model_output=jnp.zeros(shape),
+            cur_sample=jnp.zeros(shape),
+            ets=jnp.zeros((4,) + shape),
         )
 
     def step(
@@ -227,22 +231,32 @@ def step(
             When returning a tuple, the first element is the sample tensor.
 
         """
-        if state.counter < len(state.prk_timesteps) and not self.config.skip_prk_steps:
-            return self.step_prk(
-                state=state, model_output=model_output, timestep=timestep, sample=sample, return_dict=return_dict
+        if self.config.skip_prk_steps:
+            prev_sample, state = self.step_plms(
+                state=state, model_output=model_output, timestep=timestep, sample=sample
             )
         else:
-            return self.step_plms(
-                state=state, model_output=model_output, timestep=timestep, sample=sample, return_dict=return_dict
+            prev_sample, state = jax.lax.switch(
+                jnp.where(state.counter < len(state.prk_timesteps), 0, 1),
+                (self.step_prk, self.step_plms),
+                # Args to either branch
+                state,
+                model_output,
+                timestep,
+                sample,
             )
 
+        if not return_dict:
+            return (prev_sample, state)
+
+        return FlaxSchedulerOutput(prev_sample=prev_sample, state=state)
+
     def step_prk(
         self,
         state: PNDMSchedulerState,
         model_output: jnp.ndarray,
         timestep: int,
         sample: jnp.ndarray,
-        return_dict: bool = True,
     ) -> Union[FlaxSchedulerOutput, Tuple]:
         """
         Step function propagating the sample with the Runge-Kutta method. RK takes 4 forward passes to approximate the
@@ -266,42 +280,53 @@ def step_prk(
                 "Number of inference steps is 'None', you need to run 'set_timesteps' after creating the scheduler"
             )
 
-        diff_to_prev = 0 if state.counter % 2 else self.config.num_train_timesteps // state.num_inference_steps // 2
+        diff_to_prev = jnp.where(
+            state.counter % 2, 0, self.config.num_train_timesteps // state.num_inference_steps // 2
+        )
         prev_timestep = timestep - diff_to_prev
         timestep = state.prk_timesteps[state.counter // 4 * 4]
 
-        if state.counter % 4 == 0:
-            state = state.replace(
-                cur_model_output=state.cur_model_output + 1 / 6 * model_output,
-                ets=state.ets.append(model_output),
-                cur_sample=sample,
+        def remainder_0(state: PNDMSchedulerState, model_output: jnp.ndarray, ets_at: int):
+            return (
+                state.replace(
+                    cur_model_output=state.cur_model_output + 1 / 6 * model_output,
+                    ets=state.ets.at[ets_at].set(model_output),
+                    cur_sample=sample,
+                ),
+                model_output,
             )
-        elif (self.counter - 1) % 4 == 0:
-            state = state.replace(cur_model_output=state.cur_model_output + 1 / 3 * model_output)
-        elif (self.counter - 2) % 4 == 0:
-            state = state.replace(cur_model_output=state.cur_model_output + 1 / 3 * model_output)
-        elif (self.counter - 3) % 4 == 0:
-            model_output = state.cur_model_output + 1 / 6 * model_output
-            state = state.replace(cur_model_output=0)
 
-        # cur_sample should not be `None`
-        cur_sample = state.cur_sample if state.cur_sample is not None else sample
+        def remainder_1(state: PNDMSchedulerState, model_output: jnp.ndarray, ets_at: int):
+            return state.replace(cur_model_output=state.cur_model_output + 1 / 3 * model_output), model_output
 
+        def remainder_2(state: PNDMSchedulerState, model_output: jnp.ndarray, ets_at: int):
+            return state.replace(cur_model_output=state.cur_model_output + 1 / 3 * model_output), model_output
+
+        def remainder_3(state: PNDMSchedulerState, model_output: jnp.ndarray, ets_at: int):
+            model_output = state.cur_model_output + 1 / 6 * model_output
+            return state.replace(cur_model_output=jnp.zeros_like(state.cur_model_output)), model_output
+
+        state, model_output = jax.lax.switch(
+            state.counter % 4,
+            (remainder_0, remainder_1, remainder_2, remainder_3),
+            # Args to either branch
+            state,
+            model_output,
+            state.counter // 4,
+        )
+
+        cur_sample = state.cur_sample
         prev_sample = self._get_prev_sample(cur_sample, timestep, prev_timestep, model_output)
         state = state.replace(counter=state.counter + 1)
 
-        if not return_dict:
-            return (prev_sample, state)
-
-        return FlaxSchedulerOutput(prev_sample=prev_sample, state=state)
+        return (prev_sample, state)
 
     def step_plms(
         self,
         state: PNDMSchedulerState,
         model_output: jnp.ndarray,
         timestep: int,
         sample: jnp.ndarray,
-        return_dict: bool = True,
     ) -> Union[FlaxSchedulerOutput, Tuple]:
         """
         Step function propagating the sample with the linear multi-step method. This has one forward pass with multiple
@@ -334,36 +359,91 @@ def step_plms(
             )
 
         prev_timestep = timestep - self.config.num_train_timesteps // state.num_inference_steps
+        prev_timestep = jnp.where(prev_timestep > 0, prev_timestep, 0)
+
+        # Reference:
+        # if state.counter != 1:
+        #     state.ets.append(model_output)
+        # else:
+        #     prev_timestep = timestep
+        #     timestep = timestep + self.config.num_train_timesteps // state.num_inference_steps
+
+        prev_timestep = jnp.where(state.counter == 1, timestep, prev_timestep)
+        timestep = jnp.where(
+            state.counter == 1, timestep + self.config.num_train_timesteps // state.num_inference_steps, timestep
+        )
 
-        if state.counter != 1:
-            state = state.replace(ets=state.ets.append(model_output))
-        else:
-            prev_timestep = timestep
-            timestep = timestep + self.config.num_train_timesteps // state.num_inference_steps
-
-        if len(state.ets) == 1 and state.counter == 0:
-            model_output = model_output
-            state = state.replace(cur_sample=sample)
-        elif len(state.ets) == 1 and state.counter == 1:
-            model_output = (model_output + state.ets[-1]) / 2
-            sample = state.cur_sample
-            state = state.replace(cur_sample=None)
-        elif len(state.ets) == 2:
-            model_output = (3 * state.ets[-1] - state.ets[-2]) / 2
-        elif len(state.ets) == 3:
-            model_output = (23 * state.ets[-1] - 16 * state.ets[-2] + 5 * state.ets[-3]) / 12
-        else:
-            model_output = (1 / 24) * (
-                55 * state.ets[-1] - 59 * state.ets[-2] + 37 * state.ets[-3] - 9 * state.ets[-4]
+        # Reference:
+        # if len(state.ets) == 1 and state.counter == 0:
+        #     model_output = model_output
+        #     state.cur_sample = sample
+        # elif len(state.ets) == 1 and state.counter == 1:
+        #     model_output = (model_output + state.ets[-1]) / 2
+        #     sample = state.cur_sample
+        #     state.cur_sample = None
+        # elif len(state.ets) == 2:
+        #     model_output = (3 * state.ets[-1] - state.ets[-2]) / 2
+        # elif len(state.ets) == 3:
+        #     model_output = (23 * state.ets[-1] - 16 * state.ets[-2] + 5 * state.ets[-3]) / 12
+        # else:
+        #     model_output = (1 / 24) * (55 * state.ets[-1] - 59 * state.ets[-2] + 37 * state.ets[-3] - 9 * state.ets[-4])
+
+        def counter_0(state: PNDMSchedulerState):
+            ets = state.ets.at[0].set(model_output)
+            return state.replace(
+                ets=ets,
+                cur_sample=sample,
+                cur_model_output=jnp.array(model_output, dtype=jnp.float32),
+            )
+
+        def counter_1(state: PNDMSchedulerState):
+            return state.replace(
+                cur_model_output=(model_output + state.ets[0]) / 2,
             )
 
+        def counter_2(state: PNDMSchedulerState):
+            ets = state.ets.at[1].set(model_output)
+            return state.replace(
+                ets=ets,
+                cur_model_output=(3 * ets[1] - ets[0]) / 2,
+                cur_sample=sample,
+            )
+
+        def counter_3(state: PNDMSchedulerState):
+            ets = state.ets.at[2].set(model_output)
+            return state.replace(
+                ets=ets,
+                cur_model_output=(23 * ets[2] - 16 * ets[1] + 5 * ets[0]) / 12,
+                cur_sample=sample,
+            )
+
+        def counter_other(state: PNDMSchedulerState):
+            ets = state.ets.at[3].set(model_output)
+            next_model_output = (1 / 24) * (55 * ets[3] - 59 * ets[2] + 37 * ets[1] - 9 * ets[0])
+
+            ets = ets.at[0].set(ets[1])
+            ets = ets.at[1].set(ets[2])
+            ets = ets.at[2].set(ets[3])
+
+            return state.replace(
+                ets=ets,
+                cur_model_output=next_model_output,
+                cur_sample=sample,
+            )
+
+        counter = jnp.clip(state.counter, 0, 4)
+        state = jax.lax.switch(
+            counter,
+            [counter_0, counter_1, counter_2, counter_3, counter_other],
+            state,
+        )
+
+        sample = state.cur_sample
+        model_output = state.cur_model_output
         prev_sample = self._get_prev_sample(sample, timestep, prev_timestep, model_output)
         state = state.replace(counter=state.counter + 1)
 
-        if not return_dict:
-            return (prev_sample, state)
-
-        return FlaxSchedulerOutput(prev_sample=prev_sample, state=state)
+        return (prev_sample, state)
 
     def _get_prev_sample(self, sample, timestep, prev_timestep, model_output):
         # See formula (9) of PNDM paper https://arxiv.org/pdf/2202.09778.pdf
@@ -379,7 +459,7 @@ def _get_prev_sample(self, sample, timestep, prev_timestep, model_output):
         # model_output -> e_θ(x_t, t)
         # prev_sample -> x_(t−δ)
         alpha_prod_t = self.alphas_cumprod[timestep]
-        alpha_prod_t_prev = self.alphas_cumprod[prev_timestep] if prev_timestep >= 0 else self.final_alpha_cumprod
+        alpha_prod_t_prev = jnp.where(prev_timestep >= 0, self.alphas_cumprod[prev_timestep], self.final_alpha_cumprod)
         beta_prod_t = 1 - alpha_prod_t
         beta_prod_t_prev = 1 - alpha_prod_t_prev