Remove noop spmd_mode check, correct type annotations in attention flax

src/maxdiffusion/models/attention_flax.py

@@ -52,17 +52,17 @@ def _maybe_aqt_einsum(quant: Quant):
 class AttentionOp(nn.Module):
   mesh: Mesh
   attention_kernel: str
-  scale: int
+  scale: float
   heads: int
   dim_head: int
   use_memory_efficient_attention: bool = False
   split_head_dim: bool = False
   float32_qk_product: bool = True
   flash_axis_names: AxisNames = (BATCH, HEAD, LENGTH, D_KV)
   flash_min_seq_length: int = 4096
-  flash_block_sizes: BlockSizes = None
+  flash_block_sizes: BlockSizes | None = None
   dtype: DType = jnp.float32
-  quant: Quant = None
+  quant: Quant | None = None
 
   def setup(self):
     if self.attention_kernel == "cudnn_flash_te":
@@ -79,7 +79,7 @@ def setup(self):
           dtype=self.dtype,
           # float32_logits=self.float32_logits,
           qkv_layout="BSHD_BSHD_BSHD",  # 'BS3HD', 'BSHD_BS2HD' or 'BSHD_BSHD_BSHD'
-          scale_factor=self.scale,
+          scale_factor=float(self.scale),
           transpose_batch_sequence=False,
       )
 
@@ -415,15 +415,15 @@ class FlaxFluxAttention(nn.Module):
   split_head_dim: bool = False
   attention_kernel: str = "dot_product"
   flash_min_seq_length: int = 4096
-  flash_block_sizes: BlockSizes = None
-  mesh: jax.sharding.Mesh = None
+  flash_block_sizes: BlockSizes | None = None
+  mesh: jax.sharding.Mesh | None = None
   dtype: jnp.dtype = jnp.float32
   weights_dtype: jnp.dtype = jnp.float32
   query_axis_names: AxisNames = (BATCH, LENGTH, HEAD)
   key_axis_names: AxisNames = (BATCH, LENGTH, HEAD)
   value_axis_names: AxisNames = (BATCH, LENGTH, HEAD)
   out_axis_names: AxisNames = (BATCH, LENGTH, EMBED)
-  precision: jax.lax.Precision = None
+  precision: jax.lax.Precision | None = None
   qkv_bias: bool = False
 
   def setup(self):
@@ -619,16 +619,16 @@ class FlaxAttention(nn.Module):
   split_head_dim: bool = False
   attention_kernel: str = "dot_product"
   flash_min_seq_length: int = 4096
-  flash_block_sizes: BlockSizes = None
-  mesh: jax.sharding.Mesh = None
+  flash_block_sizes: BlockSizes | None = None
+  mesh: jax.sharding.Mesh | None = None
   dtype: jnp.dtype = jnp.float32
   weights_dtype: jnp.dtype = jnp.float32
   query_axis_names: AxisNames = (BATCH, LENGTH, HEAD)
   key_axis_names: AxisNames = (BATCH, LENGTH, HEAD)
   value_axis_names: AxisNames = (BATCH, LENGTH, HEAD)
   out_axis_names: AxisNames = (BATCH, LENGTH, HEAD)
-  precision: jax.lax.Precision = None
-  quant: Quant = None
+  precision: jax.lax.Precision | None = None
+  quant: Quant | None = None
 
   def setup(self):
 
@@ -762,10 +762,10 @@ class FlaxBasicTransformerBlock(nn.Module):
   split_head_dim: bool = False
   attention_kernel: str = "dot_product"
   flash_min_seq_length: int = 4096
-  flash_block_sizes: BlockSizes = None
-  mesh: jax.sharding.Mesh = None
-  precision: jax.lax.Precision = None
-  quant: Quant = None
+  flash_block_sizes: BlockSizes | None = None
+  mesh: jax.sharding.Mesh | None = None
+  precision: jax.lax.Precision | None = None
+  quant: Quant | None = None
 
   def setup(self):
     # self attention (or cross_attention if only_cross_attention is True)
@@ -890,12 +890,12 @@ class FlaxTransformer2DModel(nn.Module):
   split_head_dim: bool = False
   attention_kernel: str = "dot_product"
   flash_min_seq_length: int = 4096
-  flash_block_sizes: BlockSizes = None
-  mesh: jax.sharding.Mesh = None
+  flash_block_sizes: BlockSizes | None = None
+  mesh: jax.sharding.Mesh | None = None
   norm_num_groups: int = 32
-  precision: jax.lax.Precision = None
+  precision: jax.lax.Precision | None = None
   hidden_state_axis_names: AxisNames = (BATCH, LENGTH, D_KV)
-  quant: Quant = (None,)
+  quant: Quant | tuple[None] = (None,)
 
   def setup(self):
     self.norm = nn.GroupNorm(num_groups=self.norm_num_groups, epsilon=1e-5, dtype=self.dtype, param_dtype=self.weights_dtype)
@@ -1019,7 +1019,7 @@ class FlaxFeedForward(nn.Module):
   dropout: float = 0.0
   dtype: jnp.dtype = jnp.float32
   weights_dtype: jnp.dtype = jnp.float32
-  precision: jax.lax.Precision = None
+  precision: jax.lax.Precision | None = None
 
   def setup(self):
     # The second linear layer needs to be called
@@ -1051,7 +1051,7 @@ class FlaxGEGLU(nn.Module):
   dropout: float = 0.0
   dtype: jnp.dtype = jnp.float32
   weights_dtype: jnp.dtype = jnp.float32
-  precision: jax.lax.Precision = None
+  precision: jax.lax.Precision | None = None
 
   def setup(self):
     inner_dim = self.dim * 4

src/maxdiffusion/train_utils.py

@@ -22,8 +22,7 @@
 
 
 def get_first_step(state):
-  with jax.spmd_mode("allow_all"):
-    return int(state.step)
+  return int(state.step)
 
 
 def load_next_batch(train_iter, example_batch, config):
@@ -101,27 +100,27 @@ def write_metrics(writer, local_metrics_file, running_gcs_metrics, metrics, step
 
 def write_metrics_to_tensorboard(writer, metrics, step, config):
   """Writes metrics to tensorboard"""
-  with jax.spmd_mode("allow_all"):
-    if jax.process_index() == 0:
-      for metric_name in metrics.get("scalar", []):
-        writer.add_scalar(metric_name, np.array(metrics["scalar"][metric_name]), step)
-      for metric_name in metrics.get("scalars", []):
-        writer.add_scalars(metric_name, metrics["scalars"][metric_name], step)
-
-    full_log = step % config.log_period == 0
-    if jax.process_index() == 0:
-      max_logging.log(
-          "completed step: {}, seconds: {:.3f}, TFLOP/s/device: {:.3f}, loss: {:.3f}".format(
-              step,
-              metrics["scalar"]["perf/step_time_seconds"],
-              metrics["scalar"]["perf/per_device_tflops_per_sec"],
-              float(metrics["scalar"]["learning/loss"]),
-          )
-      )
 
-    if full_log and jax.process_index() == 0:
-      max_logging.log(f"To see full metrics 'tensorboard --logdir={config.tensorboard_dir}'")
-      writer.flush()
+  if jax.process_index() == 0:
+    for metric_name in metrics.get("scalar", []):
+      writer.add_scalar(metric_name, np.array(metrics["scalar"][metric_name]), step)
+    for metric_name in metrics.get("scalars", []):
+      writer.add_scalars(metric_name, metrics["scalars"][metric_name], step)
+
+  full_log = step % config.log_period == 0
+  if jax.process_index() == 0:
+    max_logging.log(
+        "completed step: {}, seconds: {:.3f}, TFLOP/s/device: {:.3f}, loss: {:.3f}".format(
+            step,
+            metrics["scalar"]["perf/step_time_seconds"],
+            metrics["scalar"]["perf/per_device_tflops_per_sec"],
+            float(metrics["scalar"]["learning/loss"]),
+        )
+    )
+
+  if full_log and jax.process_index() == 0:
+    max_logging.log(f"To see full metrics 'tensorboard --logdir={config.tensorboard_dir}'")
+    writer.flush()
 
 
 def get_params_to_save(params):

src/maxdiffusion/utils/outputs.py

@@ -16,7 +16,7 @@
 """
 
 from collections import OrderedDict
-from dataclasses import fields, is_dataclass
+from dataclasses import dataclass, fields, is_dataclass
 from typing import Any, Tuple
 
 import numpy as np
@@ -37,6 +37,7 @@ def is_tensor(x):
   return isinstance(x, np.ndarray)
 
 
+@dataclass
 class BaseOutput(OrderedDict):
   """
   Base class for all model outputs as dataclass. Has a `__getitem__` that allows indexing by integer or slice (like a

src/maxdiffusion/utils/pil_utils.py

@@ -16,11 +16,11 @@
   }
 else:
   PIL_INTERPOLATION = {
-      "linear": PIL.Image.LINEAR,
-      "bilinear": PIL.Image.BILINEAR,
-      "bicubic": PIL.Image.BICUBIC,
-      "lanczos": PIL.Image.LANCZOS,
-      "nearest": PIL.Image.NEAREST,
+      "linear": PIL.Image.LINEAR,  # pytype: disable=module-attr
+      "bilinear": PIL.Image.BILINEAR,  # pytype: disable=module-attr
+      "bicubic": PIL.Image.BICUBIC,  # pytype: disable=module-attr
+      "lanczos": PIL.Image.LANCZOS,  # pytype: disable=module-attr
+      "nearest": PIL.Image.NEAREST,  # pytype: disable=module-attr
   }
 
 
@@ -50,7 +50,7 @@ def numpy_to_pil(images):
   return pil_images
 
 
-def make_image_grid(images: List[PIL.Image.Image], rows: int, cols: int, resize: int = None) -> PIL.Image.Image:
+def make_image_grid(images: List[PIL.Image.Image], rows: int, cols: int, resize: int | None = None) -> PIL.Image.Image:
   """
   Prepares a single grid of images. Useful for visualization purposes.
   """