more warnings.

Author: sayakpaul

src/diffusers/models/attention_flax.py

@@ -19,6 +19,11 @@
 import jax
 import jax.numpy as jnp
 
+from ..utils import logging
+
+
+logger = logging.get_logger(__name__)
+
 
 def _query_chunk_attention(query, key, value, precision, key_chunk_size: int = 4096):
     """Multi-head dot product attention with a limited number of queries."""
@@ -151,6 +156,11 @@ class FlaxAttention(nn.Module):
     dtype: jnp.dtype = jnp.float32
 
     def setup(self):
+        logger.warning(
+            "Flax classes are deprecated and will be removed in Diffusers v1. We "
+            "recommend migrating to PyTorch classes or pinning your version of Diffusers."
+        )
+
         inner_dim = self.dim_head * self.heads
         self.scale = self.dim_head**-0.5
 
@@ -277,6 +287,11 @@ class FlaxBasicTransformerBlock(nn.Module):
     split_head_dim: bool = False
 
     def setup(self):
+        logger.warning(
+            "Flax classes are deprecated and will be removed in Diffusers v1. We "
+            "recommend migrating to PyTorch classes or pinning your version of Diffusers."
+        )
+
         # self attention (or cross_attention if only_cross_attention is True)
         self.attn1 = FlaxAttention(
             self.dim,
@@ -365,6 +380,11 @@ class FlaxTransformer2DModel(nn.Module):
     split_head_dim: bool = False
 
     def setup(self):
+        logger.warning(
+            "Flax classes are deprecated and will be removed in Diffusers v1. We "
+            "recommend migrating to PyTorch classes or pinning your version of Diffusers."
+        )
+
         self.norm = nn.GroupNorm(num_groups=32, epsilon=1e-5)
 
         inner_dim = self.n_heads * self.d_head
@@ -454,6 +474,11 @@ class FlaxFeedForward(nn.Module):
     dtype: jnp.dtype = jnp.float32
 
     def setup(self):
+        logger.warning(
+            "Flax classes are deprecated and will be removed in Diffusers v1. We "
+            "recommend migrating to PyTorch classes or pinning your version of Diffusers."
+        )
+
         # The second linear layer needs to be called
         # net_2 for now to match the index of the Sequential layer
         self.net_0 = FlaxGEGLU(self.dim, self.dropout, self.dtype)
@@ -484,6 +509,11 @@ class FlaxGEGLU(nn.Module):
     dtype: jnp.dtype = jnp.float32
 
     def setup(self):
+        logger.warning(
+            "Flax classes are deprecated and will be removed in Diffusers v1. We "
+            "recommend migrating to PyTorch classes or pinning your version of Diffusers."
+        )
+
         inner_dim = self.dim * 4
         self.proj = nn.Dense(inner_dim * 2, dtype=self.dtype)
         self.dropout_layer = nn.Dropout(rate=self.dropout)

src/diffusers/models/controlnets/controlnet_flax.py

@@ -20,7 +20,7 @@
 from flax.core.frozen_dict import FrozenDict
 
 from ...configuration_utils import ConfigMixin, flax_register_to_config
-from ...utils import BaseOutput
+from ...utils import BaseOutput, logging
 from ..embeddings_flax import FlaxTimestepEmbedding, FlaxTimesteps
 from ..modeling_flax_utils import FlaxModelMixin
 from ..unets.unet_2d_blocks_flax import (
@@ -30,6 +30,9 @@
 )
 
 
+logger = logging.get_logger(__name__)
+
+
 @flax.struct.dataclass
 class FlaxControlNetOutput(BaseOutput):
     """
@@ -50,6 +53,11 @@ class FlaxControlNetConditioningEmbedding(nn.Module):
     dtype: jnp.dtype = jnp.float32
 
     def setup(self) -> None:
+        logger.warning(
+            "Flax classes are deprecated and will be removed in Diffusers v1. We "
+            "recommend migrating to PyTorch classes or pinning your version of Diffusers."
+        )
+
         self.conv_in = nn.Conv(
             self.block_out_channels[0],
             kernel_size=(3, 3),
@@ -184,6 +192,11 @@ def init_weights(self, rng: jax.Array) -> FrozenDict:
         return self.init(rngs, sample, timesteps, encoder_hidden_states, controlnet_cond)["params"]
 
     def setup(self) -> None:
+        logger.warning(
+            "Flax classes are deprecated and will be removed in Diffusers v1. We "
+            "recommend migrating to PyTorch classes or pinning your version of Diffusers."
+        )
+
         block_out_channels = self.block_out_channels
         time_embed_dim = block_out_channels[0] * 4
 

src/diffusers/models/embeddings_flax.py

@@ -16,6 +16,11 @@
 import flax.linen as nn
 import jax.numpy as jnp
 
+from ..utils import logging
+
+
+logger = logging.get_logger(__name__)
+
 
 def get_sinusoidal_embeddings(
     timesteps: jnp.ndarray,
@@ -76,6 +81,11 @@ class FlaxTimestepEmbedding(nn.Module):
             The data type for the embedding parameters.
     """
 
+    logger.warning(
+        "Flax classes are deprecated and will be removed in Diffusers v1. We "
+        "recommend migrating to PyTorch classes or pinning your version of Diffusers."
+    )
+
     time_embed_dim: int = 32
     dtype: jnp.dtype = jnp.float32
 
@@ -104,6 +114,11 @@ class FlaxTimesteps(nn.Module):
     flip_sin_to_cos: bool = False
     freq_shift: float = 1
 
+    logger.warning(
+        "Flax classes are deprecated and will be removed in Diffusers v1. We "
+        "recommend migrating to PyTorch classes or pinning your version of Diffusers."
+    )
+
     @nn.compact
     def __call__(self, timesteps):
         return get_sinusoidal_embeddings(

src/diffusers/models/resnet_flax.py

@@ -15,12 +15,22 @@
 import jax
 import jax.numpy as jnp
 
+from ..utils import logging
+
+
+logger = logging.get_logger(__name__)
+
 
 class FlaxUpsample2D(nn.Module):
     out_channels: int
     dtype: jnp.dtype = jnp.float32
 
     def setup(self):
+        logger.warning(
+            "Flax classes are deprecated and will be removed in Diffusers v1. We "
+            "recommend migrating to PyTorch classes or pinning your version of Diffusers."
+        )
+
         self.conv = nn.Conv(
             self.out_channels,
             kernel_size=(3, 3),
@@ -45,6 +55,11 @@ class FlaxDownsample2D(nn.Module):
     dtype: jnp.dtype = jnp.float32
 
     def setup(self):
+        logger.warning(
+            "Flax classes are deprecated and will be removed in Diffusers v1. We "
+            "recommend migrating to PyTorch classes or pinning your version of Diffusers."
+        )
+
         self.conv = nn.Conv(
             self.out_channels,
             kernel_size=(3, 3),
@@ -68,6 +83,11 @@ class FlaxResnetBlock2D(nn.Module):
     dtype: jnp.dtype = jnp.float32
 
     def setup(self):
+        logger.warning(
+            "Flax classes are deprecated and will be removed in Diffusers v1. We "
+            "recommend migrating to PyTorch classes or pinning your version of Diffusers."
+        )
+
         out_channels = self.in_channels if self.out_channels is None else self.out_channels
 
         self.norm1 = nn.GroupNorm(num_groups=32, epsilon=1e-5)

src/diffusers/models/unets/unet_2d_blocks_flax.py

@@ -15,10 +15,14 @@
 import flax.linen as nn
 import jax.numpy as jnp
 
+from ...utils import logging
 from ..attention_flax import FlaxTransformer2DModel
 from ..resnet_flax import FlaxDownsample2D, FlaxResnetBlock2D, FlaxUpsample2D
 
 
+logger = logging.get_logger(__name__)
+
+
 class FlaxCrossAttnDownBlock2D(nn.Module):
     r"""
     Cross Attention 2D Downsizing block - original architecture from Unet transformers:
@@ -60,6 +64,11 @@ class FlaxCrossAttnDownBlock2D(nn.Module):
     transformer_layers_per_block: int = 1
 
     def setup(self):
+        logger.warning(
+            "Flax classes are deprecated and will be removed in Diffusers v1. We "
+            "recommend migrating to PyTorch classes or pinning your version of Diffusers."
+        )
+
         resnets = []
         attentions = []
 
@@ -135,6 +144,11 @@ class FlaxDownBlock2D(nn.Module):
     dtype: jnp.dtype = jnp.float32
 
     def setup(self):
+        logger.warning(
+            "Flax classes are deprecated and will be removed in Diffusers v1. We "
+            "recommend migrating to PyTorch classes or pinning your version of Diffusers."
+        )
+
         resnets = []
 
         for i in range(self.num_layers):
@@ -208,6 +222,11 @@ class FlaxCrossAttnUpBlock2D(nn.Module):
     transformer_layers_per_block: int = 1
 
     def setup(self):
+        logger.warning(
+            "Flax classes are deprecated and will be removed in Diffusers v1. We "
+            "recommend migrating to PyTorch classes or pinning your version of Diffusers."
+        )
+
         resnets = []
         attentions = []
 
@@ -288,6 +307,11 @@ class FlaxUpBlock2D(nn.Module):
     dtype: jnp.dtype = jnp.float32
 
     def setup(self):
+        logger.warning(
+            "Flax classes are deprecated and will be removed in Diffusers v1. We "
+            "recommend migrating to PyTorch classes or pinning your version of Diffusers."
+        )
+
         resnets = []
 
         for i in range(self.num_layers):
@@ -356,6 +380,11 @@ class FlaxUNetMidBlock2DCrossAttn(nn.Module):
     transformer_layers_per_block: int = 1
 
     def setup(self):
+        logger.warning(
+            "Flax classes are deprecated and will be removed in Diffusers v1. We "
+            "recommend migrating to PyTorch classes or pinning your version of Diffusers."
+        )
+
         # there is always at least one resnet
         resnets = [
             FlaxResnetBlock2D(

src/diffusers/models/unets/unet_2d_condition_flax.py

@@ -20,7 +20,7 @@
 from flax.core.frozen_dict import FrozenDict
 
 from ...configuration_utils import ConfigMixin, flax_register_to_config
-from ...utils import BaseOutput
+from ...utils import BaseOutput, logging
 from ..embeddings_flax import FlaxTimestepEmbedding, FlaxTimesteps
 from ..modeling_flax_utils import FlaxModelMixin
 from .unet_2d_blocks_flax import (
@@ -32,6 +32,9 @@
 )
 
 
+logger = logging.get_logger(__name__)
+
+
 @flax.struct.dataclass
 class FlaxUNet2DConditionOutput(BaseOutput):
     """
@@ -163,6 +166,11 @@ def init_weights(self, rng: jax.Array) -> FrozenDict:
         return self.init(rngs, sample, timesteps, encoder_hidden_states, added_cond_kwargs)["params"]
 
     def setup(self) -> None:
+        logger.warning(
+            "Flax classes are deprecated and will be removed in Diffusers v1. We "
+            "recommend migrating to PyTorch classes or pinning your version of Diffusers."
+        )
+
         block_out_channels = self.block_out_channels
         time_embed_dim = block_out_channels[0] * 4