[DFM Migration] Migration DFM -> Bridge

examples/diffusion/README.md

@@ -0,0 +1,3 @@
+# Megatron Examples
+
+Recipes and configuration overrides for megatron training.

examples/diffusion/override_configs/README.md

@@ -0,0 +1,3 @@
+# Override Configs
+
+Parallelism configuration overrides for different CP/TP/SP sizes.

examples/diffusion/override_configs/wan_pretrain_sample_data.yaml

@@ -0,0 +1,44 @@
+# WAN Pretrain Mock Data Test Configuration
+# Converted from L2_Function_Tests_GPU_Wan_Mock_Data.sh
+
+model:
+  tensor_model_parallel_size: 1
+  pipeline_model_parallel_size: 1
+  context_parallel_size: 1
+  crossattn_emb_size: 1536
+  hidden_size: 1536
+  ffn_hidden_size: 8960
+  num_attention_heads: 12
+  num_layers: 3
+  qkv_format: thd
+  seq_length: 2048
+
+train:
+  eval_iters: 0
+  train_iters: 10
+  global_batch_size: 2
+  micro_batch_size: 1
+
+optimizer:
+  lr: 5.0e-6
+  min_lr: 5.0e-6
+
+scheduler:
+  lr_decay_style: constant
+  lr_warmup_iters: 0
+
+checkpoint:
+  save: ${oc.env:CHECKPOINT_DIR,null}
+  load: ${oc.env:CHECKPOINT_DIR,null}
+  load_optim: false
+  save_interval: 200
+
+dataset:
+  path: ${oc.env:DATASET_PATH,null}
+  seq_length: 2048
+  global_batch_size: 2
+  micro_batch_size: 1
+  packing_buffer_size: 50
+
+logger:
+  log_interval: 1

examples/diffusion/recipes/README.md

@@ -0,0 +1,3 @@
+# Recipe
+
+Training recipes for Wan2.1 pretraining, finetuning, and weight verification.

examples/diffusion/recipes/flux/conf/flux_pretrain_override_example.yaml

@@ -0,0 +1,56 @@
+# Copyright (c) 2025, NVIDIA CORPORATION.  All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+# Example FLUX pretrain configuration override file
+# This file shows common overrides for FLUX pretraining
+
+# Model configuration
+model:
+  # Parallelism settings
+  tensor_model_parallel_size: 4
+  pipeline_model_parallel_size: 1
+  context_parallel_size: 1
+  sequence_parallel: false
+
+  # FLUX architecture (FLUX-schnell defaults)
+  num_joint_layers: 19
+  num_single_layers: 38
+  hidden_size: 3072
+  num_attention_heads: 24
+  in_channels: 64
+  context_dim: 4096
+
+  # For FLUX-dev, set guidance_embed: true
+  guidance_embed: false
+  guidance_scale: 3.5
+
+# Training configuration
+train:
+  train_iters: 10000
+  eval_interval: 2000
+  eval_iters: 32
+  global_batch_size: 64
+  micro_batch_size: 1
+
+# Optimizer configuration
+optimizer:
+  lr: 1.0e-4
+
+# Checkpoint configuration
+checkpoint:
+  save_interval: 2000
+
+# Logger configuration
+logger:
+  log_interval: 1