Upload folder using huggingface_hub

exp_hcai/latent_diffusion/imagenet_256/dc_ae_f32c32_in_1.0_256px/dit_xl_1/SiTSampler_ODE_heun2_30/mxfp4_e2m1_plus_cs/checkpoint.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9d11a3de8376139f69c734fe556f9ca21363379b6f226ec011471d00a9975b9b
+size 10798884178

exp_hcai/latent_diffusion/imagenet_256/dc_ae_f32c32_in_1.0_256px/dit_xl_1/SiTSampler_ODE_heun2_30/mxfp4_e2m1_plus_cs/checkpoint_.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9d11a3de8376139f69c734fe556f9ca21363379b6f226ec011471d00a9975b9b
+size 10798884178

exp_hcai/latent_diffusion/imagenet_256/dc_ae_f32c32_in_1.0_256px/dit_xl_1/SiTSampler_ODE_heun2_30/mxfp4_e2m1_plus_cs/config.yaml

@@ -0,0 +1,271 @@
+run_dir: exp_hcai/latent_diffusion/imagenet_256/dc_ae_f32c32_in_1.0_256px/dit_xl_1/SiTSampler_ODE_heun2_30/mxfp4_e2m1_plus_cs
+seed: 0
+allow_tf32: true
+timeout: null
+resolution: 256
+amp: bf16
+cfg_scale: 1.0
+evaluate_split: test
+eval_dir_name: null
+num_save_images: 64
+save_all_images: false
+save_image_format: jpg
+save_images_at_all_procs: false
+save_latent_samples: false
+latent_samples_dir: null
+evaluate_dataset: sample_class
+sample_class:
+  name: SampleClass
+  batch_size: 128
+  n_worker: 8
+  drop_last: false
+  seed: 0
+  shuffle: false
+  num_classes: 1000
+  num_samples: 50000
+autoencoder:
+  num_settings: 1
+  name: dc-ae-f32c32-in-1.0-256px
+  scaling_factor: 0.3285
+  latent_channels: null
+autoencoder_dtype: fp32
+eval_autoencoder_setting_list: null
+model: fp8coat_dit
+dit:
+  name: DiT
+  in_channels: 32
+  input_size: 8
+  cfg_channels: null
+  pretrained_path: null
+  pretrained_source: dc-ae
+  train_scheduler: SiTSampler
+  eval_scheduler: ODE_heun2
+  num_inference_steps: 30
+  flow_shift: 3.0
+  reverse_time: false
+  use_cads: false
+  cads_noise_scale: 0.1
+  cads_mixing_factor: 1.0
+  cads_tau_min: 0.2
+  cads_tau_max: 0.9
+  use_guidance_interval: false
+  guidance_t_min: 0.2
+  guidance_t_max: 0.8
+  count_nfe: false
+  patch_size: 1
+  hidden_size: 1152
+  depth: 28
+  num_heads: 16
+  mlp_ratio: 4.0
+  post_norm: false
+  class_dropout_prob: 0.1
+  num_classes: 1000
+  learn_sigma: false
+  unconditional: false
+  use_checkpoint: true
+  adaptive_channel: false
+  adaptive_channel_share_weights: true
+  only_load_backbone: false
+  freeze_backbone: false
+uvit:
+  name: UViT
+  in_channels: 4
+  input_size: 32
+  cfg_channels: null
+  pretrained_path: null
+  pretrained_source: dc-ae
+  train_scheduler: DPM_Solver
+  eval_scheduler: DPM_Solver
+  num_inference_steps: 30
+  flow_shift: 3.0
+  reverse_time: false
+  use_cads: false
+  cads_noise_scale: 0.1
+  cads_mixing_factor: 1.0
+  cads_tau_min: 0.2
+  cads_tau_max: 0.9
+  use_guidance_interval: false
+  guidance_t_min: 0.2
+  guidance_t_max: 0.8
+  count_nfe: false
+  patch_size: 2
+  hidden_size: 1152
+  depth: 28
+  num_heads: 16
+  mlp_ratio: 4.0
+  mlp_time_embed: false
+  qkv_bias: false
+  act_layer: gelu
+  use_checkpoint: true
+  class_dropout_prob: 0.1
+  num_classes: 1000
+  attn_mode: null
+sana_cls:
+  name: SanaCls
+  in_channels: 4
+  input_size: 32
+  cfg_channels: null
+  pretrained_path: null
+  pretrained_source: dc-ae
+  train_scheduler: SanaScheduler
+  eval_scheduler: SanaScheduler
+  num_inference_steps: 250
+  flow_shift: 3.0
+  reverse_time: false
+  use_cads: false
+  cads_noise_scale: 0.1
+  cads_mixing_factor: 1.0
+  cads_tau_min: 0.2
+  cads_tau_max: 0.9
+  use_guidance_interval: false
+  guidance_t_min: 0.2
+  guidance_t_max: 0.8
+  count_nfe: false
+  patch_size: 2
+  hidden_size: 1152
+  depth: 28
+  num_heads: 16
+  mlp_ratio: 4.0
+  post_norm: false
+  class_dropout_prob: 0.1
+  num_classes: 1000
+  unconditional: false
+  use_checkpoint: true
+  only_load_backbone: false
+  freeze_backbone: false
+  learn_sigma: false
+usana_cls:
+  name: USanaCls
+  in_channels: 4
+  input_size: 32
+  cfg_channels: null
+  pretrained_path: null
+  pretrained_source: dc-ae
+  train_scheduler: DPM_Solver
+  eval_scheduler: DPM_Solver
+  num_inference_steps: 30
+  flow_shift: 3.0
+  reverse_time: false
+  use_cads: false
+  cads_noise_scale: 0.1
+  cads_mixing_factor: 1.0
+  cads_tau_min: 0.2
+  cads_tau_max: 0.9
+  use_guidance_interval: false
+  guidance_t_min: 0.2
+  guidance_t_max: 0.8
+  count_nfe: false
+  patch_size: 2
+  hidden_size: 1152
+  depth: 28
+  num_heads: 16
+  mlp_ratio: 4.0
+  mlp_time_embed: false
+  qkv_bias: false
+  act_layer: gelu
+  use_checkpoint: true
+  class_dropout_prob: 0.1
+  num_classes: 1000
+  num_training_steps: 1000
+fp8:
+  name: FP8DiT
+fp8coat:
+  name: FP8COATDiT
+  qchoice: linear
+  symm: true
+  row_blocksize: -1
+  col_blocksize: -1
+  linear_row_blocksize: 1
+  linear_col_blocksize: 32
+  min_blockunit_row: -1
+  min_blockunit_col: -1
+  fabit: MXE2M1_plus
+  fwbit: MXE2M1_plus
+  babit: MXE2M1_plus
+  bwbit: MXE2M1_plus
+  bobit: MXE2M1_plus
+  epsilon: 1.0e-08
+compute_fid: true
+fid:
+  save_path: null
+  ref_path: assets/data/fid/imagenet_train_256.npz
+  precision_recall_ref_path: assets/data/precision_recall/VIRTUAL_imagenet256.npy
+compute_inception_score: true
+inception_score: {}
+compute_cmmd: true
+cmmd:
+  save_path: null
+  ref_path: assets/data/cmmd/VIRTUAL_imagenet256.npy
+verbose: false
+train_dataset: latent_imagenet
+latent_imagenet:
+  name: LatentImageNet
+  batch_size: 128
+  n_worker: 8
+  drop_last: true
+  seed: 0
+  shuffle: true
+  data_dir: assets/data/latent/dc_ae_f32c32_in_1.0_256px/imagenet_256
+latent_mjhq:
+  name: LatentMJHQ
+  batch_size: 32
+  n_worker: 8
+  drop_last: true
+  seed: 0
+  shuffle: true
+  data_dir: assets/data/latent/dc_ae_f32c32/mjhq_1024
+latent_ffhq:
+  name: LatentFFHQ
+  batch_size: 32
+  n_worker: 8
+  drop_last: true
+  seed: 0
+  shuffle: true
+  data_dir: assets/data/latent/dc_ae_f32c32/ffhq_1024
+latent_mapillary_vistas:
+  name: LatentMapillaryVistas
+  batch_size: 32
+  n_worker: 8
+  drop_last: true
+  seed: 0
+  shuffle: true
+  data_dir: assets/data/latent/dc_ae_f32c32/mapillary_vistas_2048
+latent_multiple_channel_imagenet:
+  name: LatentMultipleChannelImageNet
+  batch_size: 32
+  n_worker: 8
+  drop_last: true
+  seed: 0
+  shuffle: true
+  dataset_sample_ratio: null
+  num_channels_list: null
+  data_dirs:
+  - assets/data/latent/dc_ae_f32c32/imagenet_512
+resume: true
+resume_path: null
+resume_schedule: true
+num_epochs: null
+max_steps: 500000
+clip_grad: null
+num_store_images: 64
+save_checkpoint_steps: 1000
+eval_steps: 20000
+save_eval_checkpoint_steps: 100000
+optimizer:
+  name: adamw
+  lr: 0.0001
+  warmup_lr: 0.0
+  weight_decay: 0.0
+  no_wd_keys: []
+  betas:
+  - 0.9
+  - 0.999
+lr_scheduler:
+  name: constant
+  warmup_steps: 1000
+log: true
+wandb_entity: han2024
+wandb_project: dc_ae_diffusion
+ema_decay: 0.9999
+ema_warmup_steps: 2000
+eval_ema: true

exp_hcai/latent_diffusion/imagenet_256/dc_ae_f32c32_in_1.0_256px/dit_xl_1/SiTSampler_ODE_heun2_30/mxfp4_e2m1_plus_cs/eval_results.csv

@@ -0,0 +1,14 @@
+,fid,precision,recall,inception_score_mean,inception_score_std,cmmd
+step_100000_autoencoder_setting_0_cfg_1.0,22.82251239617085,0.6495599746704102,0.5740000009536743,54.4717929909383,1.3851383102860535,0.573277473449707
+step_120000_autoencoder_setting_0_cfg_1.0,23.09154068529824,0.639739990234375,0.5774999856948853,56.29451314562423,1.3199133010401618,0.6319284439086914
+step_140000_autoencoder_setting_0_cfg_1.0,20.76137721606244,0.6392399668693542,0.5882999897003174,61.48154406857825,1.6071809236386536,0.6085634231567383
+step_160000_autoencoder_setting_0_cfg_1.0,19.763616799984906,0.6626799702644348,0.5631999969482422,66.3743937296074,1.3981266954044251,0.5801916122436523
+step_180000_autoencoder_setting_0_cfg_1.0,18.79058651979068,0.6640200018882751,0.5453000068664551,68.40362149122765,1.412380260108231,0.4642009735107422
+step_200000_autoencoder_setting_0_cfg_1.0,20.954522054968265,0.653499960899353,0.5582000017166138,63.14416956965454,1.0557505979988988,0.5255937576293945
+step_20000_autoencoder_setting_0_cfg_1.0,57.72364422441791,0.4287199974060058,0.5460999608039856,19.90047475034472,0.3800645592686704,1.1063814163208008
+step_220000_autoencoder_setting_0_cfg_1.0,18.47325050523102,0.6658399701118469,0.5615000128746033,69.24973211027222,1.3201464970260688,0.4718303680419922
+step_240000_autoencoder_setting_0_cfg_1.0,21.973751038787213,0.6630799770355225,0.5539000034332275,59.74237377356919,1.562433067266785,0.5408525466918945
+step_260000_autoencoder_setting_0_cfg_1.0,17.706635191026123,0.680079996585846,0.5595999956130981,71.48858761147187,1.564728001100873,0.48232078552246094
+step_40000_autoencoder_setting_0_cfg_1.0,44.55932573190052,0.5211399793624878,0.5742999911308289,26.626259701696387,0.5180268710510688,0.9006261825561523
+step_60000_autoencoder_setting_0_cfg_1.0,36.75938892258444,0.5808199644088745,0.5612999796867371,34.614852400498975,0.9469187747487444,0.8183717727661133
+step_80000_autoencoder_setting_0_cfg_1.0,30.842797217782447,0.608299970626831,0.5734999775886536,42.12004014484396,1.0538750580332945,0.7462501525878906

exp_hcai/latent_diffusion/imagenet_256/dc_ae_f32c32_in_1.0_256px/dit_xl_1/SiTSampler_ODE_heun2_30/mxfp4_e2m1_plus_cs/model.txt

@@ -0,0 +1,56 @@
+FP8COATDiT(
+  (x_embedder): PatchEmbed(
+    (proj): Conv2d(32, 1152, kernel_size=(1, 1), stride=(1, 1))
+    (norm): Identity()
+  )
+  (t_embedder): TimestepEmbedder(
+    (mlp): Sequential(
+      (0): Linear(in_features=256, out_features=1152, bias=True)
+      (1): SiLU()
+      (2): Linear(in_features=1152, out_features=1152, bias=True)
+    )
+  )
+  (y_embedder): LabelEmbedder(
+    (embedding_table): Embedding(1001, 1152)
+  )
+  (blocks): ModuleList(
+    (0-27): 28 x FP8COATDiTBlock(
+      (norm1): LayerNorm((1152,), eps=1e-06, elementwise_affine=False)
+      (attn): FP8COATAttention(
+        (qkv): QAct_Linear(in_features=1152, out_features=3456, bias=True)
+        (q_norm): Identity()
+        (k_norm): Identity()
+        (attn_drop): Dropout(p=0.0, inplace=False)
+        (proj): QAct_Linear(in_features=1152, out_features=1152, bias=True)
+        (proj_drop): Dropout(p=0.0, inplace=False)
+        (qact_q_quantize): QAct_Quantize()
+        (qact_k_quantize): QAct_Quantize()
+      )
+      (norm2): LayerNorm((1152,), eps=1e-06, elementwise_affine=False)
+      (mlp): FP8COATMlp(
+        (fc1): QAct_Linear(in_features=1152, out_features=4608, bias=True)
+        (act): GELU(approximate='tanh')
+        (drop1): Dropout(p=0, inplace=False)
+        (norm): Identity()
+        (fc2): QAct_Linear(in_features=4608, out_features=1152, bias=True)
+        (drop2): Dropout(p=0, inplace=False)
+        (qact_quantize_gelu): QAct_Quantize()
+        (qact_quantize_norm): QAct_Quantize()
+      )
+      (adaLN_modulation): Sequential(
+        (0): SiLU()
+        (1): Linear(in_features=1152, out_features=6912, bias=True)
+      )
+      (qact_quantize_norm1): QAct_Quantize()
+      (qact_quantize_norm2): QAct_Quantize()
+    )
+  )
+  (final_layer): FinalLayer(
+    (norm_final): LayerNorm((1152,), eps=1e-06, elementwise_affine=False)
+    (linear): Linear(in_features=1152, out_features=32, bias=True)
+    (adaLN_modulation): Sequential(
+      (0): SiLU()
+      (1): Linear(in_features=1152, out_features=2304, bias=True)
+    )
+  )
+)

exp_hcai/latent_diffusion/imagenet_256/dc_ae_f32c32_in_1.0_256px/dit_xl_1/SiTSampler_ODE_heun2_30/mxfp4_e2m1_plus_cs/slurm.sh

@@ -0,0 +1,22 @@
+#!/bin/bash
+#SBATCH -A nvr_elm_llm                 #account
+#SBATCH -p grizzly,polar,polar3,polar4 #partition
+#SBATCH -t 04:00:00                    #wall time limit, hr:min:sec
+#SBATCH -N 1                           #number of nodes
+#SBATCH -J train_diffusion_dit         #job name
+#SBATCH --array=1-40%1
+#SBATCH --output=exp_hcai/latent_diffusion/imagenet_256/dc_ae_f32c32_in_1.0_256px/dit_xl_1/SiTSampler_ODE_heun2_30/mxfp4_e2m1_plus_cs/slurm_out/%A_%a.out
+#SBATCH --gpus-per-node 8
+#SBATCH --exclusive
+
+export LOGLEVEL=INFO
+export PATH="/home/hcai/workspace/anaconda3/envs/efficientvit/bin:$PATH"
+
+cd /lustre/fs12/portfolios/nvr/users/hcai/workspace/code/fp8-dit
+
+read -r -d '' cmd <<EOF
+torchrun --nnodes=1 --nproc_per_node=8 \
+-m efficientvit.diffusioncore.trainer yaml=exp_hcai/latent_diffusion/imagenet_256/dc_ae_f32c32_in_1.0_256px/dit_xl_1/SiTSampler_ODE_heun2_30/mxfp4_e2m1_plus_cs/config.yaml
+EOF
+
+srun bash -c "${cmd}"

exp_hcai/latent_diffusion/imagenet_256/dc_ae_f32c32_in_1.0_256px/dit_xl_1/SiTSampler_ODE_heun2_30/mxfp4_e2m1_plus_cs/step_100000.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dd43ba952a645c837ac580c184650a755dbf70ae7f7e8985eed8207667132b89
+size 10798884178

exp_hcai/latent_diffusion/imagenet_256/dc_ae_f32c32_in_1.0_256px/dit_xl_1/SiTSampler_ODE_heun2_30/mxfp4_e2m1_plus_cs/step_200000.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e8c5376cbd89dae41145a00b7a6d007763b8697b3db993d6f07f27e54eba1bb4
+size 10798884178