Upload folder using huggingface_hub
Browse files- cogvideox_lora_ckpts/lora_cogvidx5b_fluid_nexus_ball_i2v_fix3samples-10-28-17-06/10000-ema/mp_rank_00_model_states.pt +3 -0
- cogvideox_lora_ckpts/lora_cogvidx5b_fluid_nexus_ball_i2v_fix3samples-10-28-17-06/10000/mp_rank_00_model_states.pt +3 -0
- cogvideox_lora_ckpts/lora_cogvidx5b_fluid_nexus_ball_i2v_fix3samples-10-28-17-06/latest +1 -0
- cogvideox_lora_ckpts/lora_cogvidx5b_fluid_nexus_ball_i2v_fix3samples-10-28-17-06/model_config.json +4 -0
- cogvideox_lora_ckpts/lora_cogvidx5b_fluid_nexus_ball_i2v_fix3samples-10-28-17-06/training_config.yaml +219 -0
- cogvideox_lora_ckpts/lora_cogvidx5b_fluid_nexus_smoke_i2v_fix3samples-10-28-17-06/.DS_Store +0 -0
- cogvideox_lora_ckpts/lora_cogvidx5b_fluid_nexus_smoke_i2v_fix3samples-10-28-17-06/10000-ema/mp_rank_00_model_states.pt +3 -0
- cogvideox_lora_ckpts/lora_cogvidx5b_fluid_nexus_smoke_i2v_fix3samples-10-28-17-06/10000/mp_rank_00_model_states.pt +3 -0
- cogvideox_lora_ckpts/lora_cogvidx5b_fluid_nexus_smoke_i2v_fix3samples-10-28-17-06/latest +1 -0
- cogvideox_lora_ckpts/lora_cogvidx5b_fluid_nexus_smoke_i2v_fix3samples-10-28-17-06/model_config.json +4 -0
- cogvideox_lora_ckpts/lora_cogvidx5b_fluid_nexus_smoke_i2v_fix3samples-10-28-17-06/training_config.yaml +219 -0
- cogvideox_lora_ckpts/lora_cogvidx5b_scalar_flow_i2v_fix3samples-09-09-05-39/10000-ema/mp_rank_00_model_states.pt +3 -0
- cogvideox_lora_ckpts/lora_cogvidx5b_scalar_flow_i2v_fix3samples-09-09-05-39/10000/mp_rank_00_model_states.pt +3 -0
- cogvideox_lora_ckpts/lora_cogvidx5b_scalar_flow_i2v_fix3samples-09-09-05-39/latest +1 -0
- cogvideox_lora_ckpts/lora_cogvidx5b_scalar_flow_i2v_fix3samples-09-09-05-39/model_config.json +4 -0
- cogvideox_lora_ckpts/lora_cogvidx5b_scalar_flow_i2v_fix3samples-09-09-05-39/training_config.yaml +219 -0
- zero123_finetune_logs/2024-04-24T14-04-16_scalar_flow/checkpoints/step=000015499.ckpt +3 -0
- zero123_finetune_logs/2024-10-28T18-26-21_fluid_nexus_ball/checkpoints/step=000079999.ckpt +3 -0
- zero123_finetune_logs/2024-10-30T06-31-08_fluid_nexus_smoke/checkpoints/step=000051999.ckpt +3 -0
cogvideox_lora_ckpts/lora_cogvidx5b_fluid_nexus_ball_i2v_fix3samples-10-28-17-06/10000-ema/mp_rank_00_model_states.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:c7076e10568d7aceb600582ed7e537d7ea921b404f97a95e769b832dfdec7e7f
|
| 3 |
+
size 42463648157
|
cogvideox_lora_ckpts/lora_cogvidx5b_fluid_nexus_ball_i2v_fix3samples-10-28-17-06/10000/mp_rank_00_model_states.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:0f5c3a9db5de4978b30c1d53ea7cd0847a67dbf4b87071d29eb11720133acb9c
|
| 3 |
+
size 42463648157
|
cogvideox_lora_ckpts/lora_cogvidx5b_fluid_nexus_ball_i2v_fix3samples-10-28-17-06/latest
ADDED
|
@@ -0,0 +1 @@
|
|
|
|
|
|
|
| 1 |
+
10000
|
cogvideox_lora_ckpts/lora_cogvidx5b_fluid_nexus_ball_i2v_fix3samples-10-28-17-06/model_config.json
ADDED
|
@@ -0,0 +1,4 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"model_class": "SATVideoDiffusionEngine",
|
| 3 |
+
"model_parallel_size": 1
|
| 4 |
+
}
|
cogvideox_lora_ckpts/lora_cogvidx5b_fluid_nexus_ball_i2v_fix3samples-10-28-17-06/training_config.yaml
ADDED
|
@@ -0,0 +1,219 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
model:
|
| 2 |
+
scale_factor: 0.7
|
| 3 |
+
disable_first_stage_autocast: true
|
| 4 |
+
not_trainable_prefixes:
|
| 5 |
+
- all
|
| 6 |
+
log_keys:
|
| 7 |
+
- txt
|
| 8 |
+
denoiser_config:
|
| 9 |
+
target: sgm.modules.diffusionmodules.denoiser.DiscreteDenoiser
|
| 10 |
+
params:
|
| 11 |
+
num_idx: 1000
|
| 12 |
+
quantize_c_noise: false
|
| 13 |
+
weighting_config:
|
| 14 |
+
target: sgm.modules.diffusionmodules.denoiser_weighting.EpsWeighting
|
| 15 |
+
scaling_config:
|
| 16 |
+
target: sgm.modules.diffusionmodules.denoiser_scaling.VideoScaling
|
| 17 |
+
discretization_config:
|
| 18 |
+
target: sgm.modules.diffusionmodules.discretizer.ZeroSNRDDPMDiscretization
|
| 19 |
+
params:
|
| 20 |
+
shift_scale: 1.0
|
| 21 |
+
network_config:
|
| 22 |
+
target: dit_video_concat.DiffusionTransformer
|
| 23 |
+
params:
|
| 24 |
+
time_embed_dim: 512
|
| 25 |
+
elementwise_affine: true
|
| 26 |
+
num_frames: 49
|
| 27 |
+
time_compressed_rate: 4
|
| 28 |
+
latent_width: 90
|
| 29 |
+
latent_height: 60
|
| 30 |
+
num_layers: 42
|
| 31 |
+
patch_size: 2
|
| 32 |
+
in_channels: 16
|
| 33 |
+
out_channels: 16
|
| 34 |
+
hidden_size: 3072
|
| 35 |
+
adm_in_channels: 256
|
| 36 |
+
num_attention_heads: 48
|
| 37 |
+
transformer_args:
|
| 38 |
+
checkpoint_activations: true
|
| 39 |
+
vocab_size: 1
|
| 40 |
+
max_sequence_length: 64
|
| 41 |
+
layernorm_order: pre
|
| 42 |
+
skip_init: false
|
| 43 |
+
model_parallel_size: 1
|
| 44 |
+
is_decoder: false
|
| 45 |
+
modules:
|
| 46 |
+
pos_embed_config:
|
| 47 |
+
target: dit_video_concat.Rotary3DPositionEmbeddingMixin
|
| 48 |
+
params:
|
| 49 |
+
hidden_size_head: 64
|
| 50 |
+
text_length: 226
|
| 51 |
+
lora_config:
|
| 52 |
+
target: sat.model.finetune.lora2.LoraMixin
|
| 53 |
+
params:
|
| 54 |
+
r: 128
|
| 55 |
+
patch_embed_config:
|
| 56 |
+
target: dit_video_concat.ImagePatchEmbeddingMixin
|
| 57 |
+
params:
|
| 58 |
+
text_hidden_size: 4096
|
| 59 |
+
adaln_layer_config:
|
| 60 |
+
target: dit_video_concat.AdaLNMixin
|
| 61 |
+
params:
|
| 62 |
+
qk_ln: true
|
| 63 |
+
final_layer_config:
|
| 64 |
+
target: dit_video_concat.FinalLayerMixin
|
| 65 |
+
conditioner_config:
|
| 66 |
+
target: sgm.modules.GeneralConditioner
|
| 67 |
+
params:
|
| 68 |
+
emb_models:
|
| 69 |
+
- is_trainable: false
|
| 70 |
+
input_key: txt
|
| 71 |
+
ucg_rate: 0.1
|
| 72 |
+
target: sgm.modules.encoders.modules.FrozenT5Embedder
|
| 73 |
+
params:
|
| 74 |
+
model_dir: /path/to/FluidNexusRoot/cogvideox-sat/2b/t5-v1_1-xxl
|
| 75 |
+
max_length: 226
|
| 76 |
+
first_stage_config:
|
| 77 |
+
target: vae_modules.autoencoder.VideoAutoencoderInferenceWrapper
|
| 78 |
+
params:
|
| 79 |
+
cp_size: 1
|
| 80 |
+
ckpt_path: /path/to/FluidNexusRoot/cogvideox-sat/5b/vae/3d-vae.pt
|
| 81 |
+
ignore_keys:
|
| 82 |
+
- loss
|
| 83 |
+
loss_config:
|
| 84 |
+
target: torch.nn.Identity
|
| 85 |
+
regularizer_config:
|
| 86 |
+
target: vae_modules.regularizers.DiagonalGaussianRegularizer
|
| 87 |
+
encoder_config:
|
| 88 |
+
target: vae_modules.cp_enc_dec.ContextParallelEncoder3D
|
| 89 |
+
params:
|
| 90 |
+
double_z: true
|
| 91 |
+
z_channels: 16
|
| 92 |
+
resolution: 256
|
| 93 |
+
in_channels: 3
|
| 94 |
+
out_ch: 3
|
| 95 |
+
ch: 128
|
| 96 |
+
ch_mult:
|
| 97 |
+
- 1
|
| 98 |
+
- 2
|
| 99 |
+
- 2
|
| 100 |
+
- 4
|
| 101 |
+
attn_resolutions: []
|
| 102 |
+
num_res_blocks: 3
|
| 103 |
+
dropout: 0.0
|
| 104 |
+
gather_norm: true
|
| 105 |
+
decoder_config:
|
| 106 |
+
target: vae_modules.cp_enc_dec.ContextParallelDecoder3D
|
| 107 |
+
params:
|
| 108 |
+
double_z: true
|
| 109 |
+
z_channels: 16
|
| 110 |
+
resolution: 256
|
| 111 |
+
in_channels: 3
|
| 112 |
+
out_ch: 3
|
| 113 |
+
ch: 128
|
| 114 |
+
ch_mult:
|
| 115 |
+
- 1
|
| 116 |
+
- 2
|
| 117 |
+
- 2
|
| 118 |
+
- 4
|
| 119 |
+
attn_resolutions: []
|
| 120 |
+
num_res_blocks: 3
|
| 121 |
+
dropout: 0.0
|
| 122 |
+
gather_norm: false
|
| 123 |
+
loss_fn_config:
|
| 124 |
+
target: sgm.modules.diffusionmodules.loss.VideoDiffusionLoss
|
| 125 |
+
params:
|
| 126 |
+
fixed_frames: 3
|
| 127 |
+
offset_noise_level: 0
|
| 128 |
+
sigma_sampler_config:
|
| 129 |
+
target: sgm.modules.diffusionmodules.sigma_sampling.DiscreteSampling
|
| 130 |
+
params:
|
| 131 |
+
uniform_sampling: true
|
| 132 |
+
num_idx: 1000
|
| 133 |
+
discretization_config:
|
| 134 |
+
target: sgm.modules.diffusionmodules.discretizer.ZeroSNRDDPMDiscretization
|
| 135 |
+
params:
|
| 136 |
+
shift_scale: 1.0
|
| 137 |
+
sampler_config:
|
| 138 |
+
target: sgm.modules.diffusionmodules.sampling.VPSDEDPMPP2MSampler
|
| 139 |
+
params:
|
| 140 |
+
num_steps: 50
|
| 141 |
+
verbose: true
|
| 142 |
+
discretization_config:
|
| 143 |
+
target: sgm.modules.diffusionmodules.discretizer.ZeroSNRDDPMDiscretization
|
| 144 |
+
params:
|
| 145 |
+
shift_scale: 1.0
|
| 146 |
+
guider_config:
|
| 147 |
+
target: sgm.modules.diffusionmodules.guiders.DynamicCFG
|
| 148 |
+
params:
|
| 149 |
+
scale: 6
|
| 150 |
+
exp: 5
|
| 151 |
+
num_steps: 50
|
| 152 |
+
args:
|
| 153 |
+
checkpoint_activations: true
|
| 154 |
+
model_parallel_size: 1
|
| 155 |
+
experiment_name: lora_cogvidx5b_realcapture_blackbluecloudredball_i2v_fix3samples
|
| 156 |
+
mode: finetune
|
| 157 |
+
load: /path/to/FluidNexusRoot/cogvideox-sat/5b/transformer
|
| 158 |
+
no_load_rng: true
|
| 159 |
+
train_iters: 10000
|
| 160 |
+
eval_iters: 1
|
| 161 |
+
eval_interval: 1000
|
| 162 |
+
eval_batch_size: 1
|
| 163 |
+
save: /path/to/FluidNexusRoot/cogvideox_lora_ckpts
|
| 164 |
+
summary_dir: /path/to/FluidNexusRoot/cogvideox_lora_runs
|
| 165 |
+
save_interval: 500
|
| 166 |
+
log_interval: 100
|
| 167 |
+
train_data:
|
| 168 |
+
- /dev/shm/RealCaptureBlackBlueCloudRedBallSetData_cogvideox_dataset
|
| 169 |
+
valid_data:
|
| 170 |
+
- /dev/shm/RealCaptureBlackBlueCloudRedBallSetData_cogvideox_dataset_sub_235
|
| 171 |
+
split: 1,0,0
|
| 172 |
+
num_workers: 8
|
| 173 |
+
force_train: true
|
| 174 |
+
only_log_video_latents: false
|
| 175 |
+
data:
|
| 176 |
+
target: data_video.SFTDataset
|
| 177 |
+
params:
|
| 178 |
+
video_size:
|
| 179 |
+
- 480
|
| 180 |
+
- 720
|
| 181 |
+
fps: 8
|
| 182 |
+
max_num_frames: 49
|
| 183 |
+
skip_frms_num: 0
|
| 184 |
+
deepspeed:
|
| 185 |
+
train_micro_batch_size_per_gpu: 2
|
| 186 |
+
gradient_accumulation_steps: 1
|
| 187 |
+
steps_per_print: 50
|
| 188 |
+
gradient_clipping: 0.1
|
| 189 |
+
zero_optimization:
|
| 190 |
+
stage: 2
|
| 191 |
+
cpu_offload: false
|
| 192 |
+
contiguous_gradients: false
|
| 193 |
+
overlap_comm: true
|
| 194 |
+
reduce_scatter: true
|
| 195 |
+
reduce_bucket_size: 1000000000
|
| 196 |
+
allgather_bucket_size: 1000000000
|
| 197 |
+
load_from_fp32_weights: false
|
| 198 |
+
zero_allow_untested_optimizer: true
|
| 199 |
+
bf16:
|
| 200 |
+
enabled: true
|
| 201 |
+
fp16:
|
| 202 |
+
enabled: false
|
| 203 |
+
loss_scale: 0
|
| 204 |
+
loss_scale_window: 400
|
| 205 |
+
hysteresis: 2
|
| 206 |
+
min_loss_scale: 1
|
| 207 |
+
optimizer:
|
| 208 |
+
type: sat.ops.FusedEmaAdam
|
| 209 |
+
params:
|
| 210 |
+
lr: 0.001
|
| 211 |
+
betas:
|
| 212 |
+
- 0.9
|
| 213 |
+
- 0.95
|
| 214 |
+
eps: 1.0e-08
|
| 215 |
+
weight_decay: 0.0001
|
| 216 |
+
activation_checkpointing:
|
| 217 |
+
partition_activations: false
|
| 218 |
+
contiguous_memory_optimization: false
|
| 219 |
+
wall_clock_breakdown: false
|
cogvideox_lora_ckpts/lora_cogvidx5b_fluid_nexus_smoke_i2v_fix3samples-10-28-17-06/.DS_Store
ADDED
|
Binary file (6.15 kB). View file
|
|
|
cogvideox_lora_ckpts/lora_cogvidx5b_fluid_nexus_smoke_i2v_fix3samples-10-28-17-06/10000-ema/mp_rank_00_model_states.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:576a198a53cb3b73b2549e1da890071c23ecd3ae8f7394f5c96e789351e54e60
|
| 3 |
+
size 42463648157
|
cogvideox_lora_ckpts/lora_cogvidx5b_fluid_nexus_smoke_i2v_fix3samples-10-28-17-06/10000/mp_rank_00_model_states.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:13c8790a90eb62fbb115438a141d87695f452f2d4f8edf518c3621d46c3344e6
|
| 3 |
+
size 42463648157
|
cogvideox_lora_ckpts/lora_cogvidx5b_fluid_nexus_smoke_i2v_fix3samples-10-28-17-06/latest
ADDED
|
@@ -0,0 +1 @@
|
|
|
|
|
|
|
| 1 |
+
10000
|
cogvideox_lora_ckpts/lora_cogvidx5b_fluid_nexus_smoke_i2v_fix3samples-10-28-17-06/model_config.json
ADDED
|
@@ -0,0 +1,4 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"model_class": "SATVideoDiffusionEngine",
|
| 3 |
+
"model_parallel_size": 1
|
| 4 |
+
}
|
cogvideox_lora_ckpts/lora_cogvidx5b_fluid_nexus_smoke_i2v_fix3samples-10-28-17-06/training_config.yaml
ADDED
|
@@ -0,0 +1,219 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
model:
|
| 2 |
+
scale_factor: 0.7
|
| 3 |
+
disable_first_stage_autocast: true
|
| 4 |
+
not_trainable_prefixes:
|
| 5 |
+
- all
|
| 6 |
+
log_keys:
|
| 7 |
+
- txt
|
| 8 |
+
denoiser_config:
|
| 9 |
+
target: sgm.modules.diffusionmodules.denoiser.DiscreteDenoiser
|
| 10 |
+
params:
|
| 11 |
+
num_idx: 1000
|
| 12 |
+
quantize_c_noise: false
|
| 13 |
+
weighting_config:
|
| 14 |
+
target: sgm.modules.diffusionmodules.denoiser_weighting.EpsWeighting
|
| 15 |
+
scaling_config:
|
| 16 |
+
target: sgm.modules.diffusionmodules.denoiser_scaling.VideoScaling
|
| 17 |
+
discretization_config:
|
| 18 |
+
target: sgm.modules.diffusionmodules.discretizer.ZeroSNRDDPMDiscretization
|
| 19 |
+
params:
|
| 20 |
+
shift_scale: 1.0
|
| 21 |
+
network_config:
|
| 22 |
+
target: dit_video_concat.DiffusionTransformer
|
| 23 |
+
params:
|
| 24 |
+
time_embed_dim: 512
|
| 25 |
+
elementwise_affine: true
|
| 26 |
+
num_frames: 49
|
| 27 |
+
time_compressed_rate: 4
|
| 28 |
+
latent_width: 90
|
| 29 |
+
latent_height: 60
|
| 30 |
+
num_layers: 42
|
| 31 |
+
patch_size: 2
|
| 32 |
+
in_channels: 16
|
| 33 |
+
out_channels: 16
|
| 34 |
+
hidden_size: 3072
|
| 35 |
+
adm_in_channels: 256
|
| 36 |
+
num_attention_heads: 48
|
| 37 |
+
transformer_args:
|
| 38 |
+
checkpoint_activations: true
|
| 39 |
+
vocab_size: 1
|
| 40 |
+
max_sequence_length: 64
|
| 41 |
+
layernorm_order: pre
|
| 42 |
+
skip_init: false
|
| 43 |
+
model_parallel_size: 1
|
| 44 |
+
is_decoder: false
|
| 45 |
+
modules:
|
| 46 |
+
pos_embed_config:
|
| 47 |
+
target: dit_video_concat.Rotary3DPositionEmbeddingMixin
|
| 48 |
+
params:
|
| 49 |
+
hidden_size_head: 64
|
| 50 |
+
text_length: 226
|
| 51 |
+
lora_config:
|
| 52 |
+
target: sat.model.finetune.lora2.LoraMixin
|
| 53 |
+
params:
|
| 54 |
+
r: 128
|
| 55 |
+
patch_embed_config:
|
| 56 |
+
target: dit_video_concat.ImagePatchEmbeddingMixin
|
| 57 |
+
params:
|
| 58 |
+
text_hidden_size: 4096
|
| 59 |
+
adaln_layer_config:
|
| 60 |
+
target: dit_video_concat.AdaLNMixin
|
| 61 |
+
params:
|
| 62 |
+
qk_ln: true
|
| 63 |
+
final_layer_config:
|
| 64 |
+
target: dit_video_concat.FinalLayerMixin
|
| 65 |
+
conditioner_config:
|
| 66 |
+
target: sgm.modules.GeneralConditioner
|
| 67 |
+
params:
|
| 68 |
+
emb_models:
|
| 69 |
+
- is_trainable: false
|
| 70 |
+
input_key: txt
|
| 71 |
+
ucg_rate: 0.1
|
| 72 |
+
target: sgm.modules.encoders.modules.FrozenT5Embedder
|
| 73 |
+
params:
|
| 74 |
+
model_dir: /path/to/FluidNexusRoot/cogvideox-sat/2b/t5-v1_1-xxl
|
| 75 |
+
max_length: 226
|
| 76 |
+
first_stage_config:
|
| 77 |
+
target: vae_modules.autoencoder.VideoAutoencoderInferenceWrapper
|
| 78 |
+
params:
|
| 79 |
+
cp_size: 1
|
| 80 |
+
ckpt_path: /path/to/FluidNexusRoot/cogvideox-sat/5b/vae/3d-vae.pt
|
| 81 |
+
ignore_keys:
|
| 82 |
+
- loss
|
| 83 |
+
loss_config:
|
| 84 |
+
target: torch.nn.Identity
|
| 85 |
+
regularizer_config:
|
| 86 |
+
target: vae_modules.regularizers.DiagonalGaussianRegularizer
|
| 87 |
+
encoder_config:
|
| 88 |
+
target: vae_modules.cp_enc_dec.ContextParallelEncoder3D
|
| 89 |
+
params:
|
| 90 |
+
double_z: true
|
| 91 |
+
z_channels: 16
|
| 92 |
+
resolution: 256
|
| 93 |
+
in_channels: 3
|
| 94 |
+
out_ch: 3
|
| 95 |
+
ch: 128
|
| 96 |
+
ch_mult:
|
| 97 |
+
- 1
|
| 98 |
+
- 2
|
| 99 |
+
- 2
|
| 100 |
+
- 4
|
| 101 |
+
attn_resolutions: []
|
| 102 |
+
num_res_blocks: 3
|
| 103 |
+
dropout: 0.0
|
| 104 |
+
gather_norm: true
|
| 105 |
+
decoder_config:
|
| 106 |
+
target: vae_modules.cp_enc_dec.ContextParallelDecoder3D
|
| 107 |
+
params:
|
| 108 |
+
double_z: true
|
| 109 |
+
z_channels: 16
|
| 110 |
+
resolution: 256
|
| 111 |
+
in_channels: 3
|
| 112 |
+
out_ch: 3
|
| 113 |
+
ch: 128
|
| 114 |
+
ch_mult:
|
| 115 |
+
- 1
|
| 116 |
+
- 2
|
| 117 |
+
- 2
|
| 118 |
+
- 4
|
| 119 |
+
attn_resolutions: []
|
| 120 |
+
num_res_blocks: 3
|
| 121 |
+
dropout: 0.0
|
| 122 |
+
gather_norm: false
|
| 123 |
+
loss_fn_config:
|
| 124 |
+
target: sgm.modules.diffusionmodules.loss.VideoDiffusionLoss
|
| 125 |
+
params:
|
| 126 |
+
fixed_frames: 3
|
| 127 |
+
offset_noise_level: 0
|
| 128 |
+
sigma_sampler_config:
|
| 129 |
+
target: sgm.modules.diffusionmodules.sigma_sampling.DiscreteSampling
|
| 130 |
+
params:
|
| 131 |
+
uniform_sampling: true
|
| 132 |
+
num_idx: 1000
|
| 133 |
+
discretization_config:
|
| 134 |
+
target: sgm.modules.diffusionmodules.discretizer.ZeroSNRDDPMDiscretization
|
| 135 |
+
params:
|
| 136 |
+
shift_scale: 1.0
|
| 137 |
+
sampler_config:
|
| 138 |
+
target: sgm.modules.diffusionmodules.sampling.VPSDEDPMPP2MSampler
|
| 139 |
+
params:
|
| 140 |
+
num_steps: 50
|
| 141 |
+
verbose: true
|
| 142 |
+
discretization_config:
|
| 143 |
+
target: sgm.modules.diffusionmodules.discretizer.ZeroSNRDDPMDiscretization
|
| 144 |
+
params:
|
| 145 |
+
shift_scale: 1.0
|
| 146 |
+
guider_config:
|
| 147 |
+
target: sgm.modules.diffusionmodules.guiders.DynamicCFG
|
| 148 |
+
params:
|
| 149 |
+
scale: 6
|
| 150 |
+
exp: 5
|
| 151 |
+
num_steps: 50
|
| 152 |
+
args:
|
| 153 |
+
checkpoint_activations: true
|
| 154 |
+
model_parallel_size: 1
|
| 155 |
+
experiment_name: lora_cogvidx5b_realcapture_blackbluecloud_i2v_fix3samples
|
| 156 |
+
mode: finetune
|
| 157 |
+
load: /path/to/FluidNexusRoot/cogvideox-sat/5b/transformer
|
| 158 |
+
no_load_rng: true
|
| 159 |
+
train_iters: 10000
|
| 160 |
+
eval_iters: 1
|
| 161 |
+
eval_interval: 1000
|
| 162 |
+
eval_batch_size: 1
|
| 163 |
+
save: /path/to/FluidNexusRoot/cogvideox_lora_ckpts
|
| 164 |
+
summary_dir: /path/to/FluidNexusRoot/cogvideox_lora_runs
|
| 165 |
+
save_interval: 500
|
| 166 |
+
log_interval: 100
|
| 167 |
+
train_data:
|
| 168 |
+
- /dev/shm/RealCaptureBlackBlueCloudSetData_cogvideox_dataset
|
| 169 |
+
valid_data:
|
| 170 |
+
- /dev/shm/RealCaptureBlackBlueCloudSetData_cogvideox_dataset_sub_235
|
| 171 |
+
split: 1,0,0
|
| 172 |
+
num_workers: 8
|
| 173 |
+
force_train: true
|
| 174 |
+
only_log_video_latents: false
|
| 175 |
+
data:
|
| 176 |
+
target: data_video.SFTDataset
|
| 177 |
+
params:
|
| 178 |
+
video_size:
|
| 179 |
+
- 480
|
| 180 |
+
- 720
|
| 181 |
+
fps: 8
|
| 182 |
+
max_num_frames: 49
|
| 183 |
+
skip_frms_num: 0
|
| 184 |
+
deepspeed:
|
| 185 |
+
train_micro_batch_size_per_gpu: 2
|
| 186 |
+
gradient_accumulation_steps: 1
|
| 187 |
+
steps_per_print: 50
|
| 188 |
+
gradient_clipping: 0.1
|
| 189 |
+
zero_optimization:
|
| 190 |
+
stage: 2
|
| 191 |
+
cpu_offload: false
|
| 192 |
+
contiguous_gradients: false
|
| 193 |
+
overlap_comm: true
|
| 194 |
+
reduce_scatter: true
|
| 195 |
+
reduce_bucket_size: 1000000000
|
| 196 |
+
allgather_bucket_size: 1000000000
|
| 197 |
+
load_from_fp32_weights: false
|
| 198 |
+
zero_allow_untested_optimizer: true
|
| 199 |
+
bf16:
|
| 200 |
+
enabled: true
|
| 201 |
+
fp16:
|
| 202 |
+
enabled: false
|
| 203 |
+
loss_scale: 0
|
| 204 |
+
loss_scale_window: 400
|
| 205 |
+
hysteresis: 2
|
| 206 |
+
min_loss_scale: 1
|
| 207 |
+
optimizer:
|
| 208 |
+
type: sat.ops.FusedEmaAdam
|
| 209 |
+
params:
|
| 210 |
+
lr: 0.001
|
| 211 |
+
betas:
|
| 212 |
+
- 0.9
|
| 213 |
+
- 0.95
|
| 214 |
+
eps: 1.0e-08
|
| 215 |
+
weight_decay: 0.0001
|
| 216 |
+
activation_checkpointing:
|
| 217 |
+
partition_activations: false
|
| 218 |
+
contiguous_memory_optimization: false
|
| 219 |
+
wall_clock_breakdown: false
|
cogvideox_lora_ckpts/lora_cogvidx5b_scalar_flow_i2v_fix3samples-09-09-05-39/10000-ema/mp_rank_00_model_states.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:ca32a23c842129b42bbbd15021e6487c342fa22dddd7d9af9a58bceb3e1df378
|
| 3 |
+
size 42463648157
|
cogvideox_lora_ckpts/lora_cogvidx5b_scalar_flow_i2v_fix3samples-09-09-05-39/10000/mp_rank_00_model_states.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:a484a53f48e0ea059a7ab4e5df3ca5330fcf6f1adcdf3111593e3d55aea0416c
|
| 3 |
+
size 42463648157
|
cogvideox_lora_ckpts/lora_cogvidx5b_scalar_flow_i2v_fix3samples-09-09-05-39/latest
ADDED
|
@@ -0,0 +1 @@
|
|
|
|
|
|
|
| 1 |
+
10000
|
cogvideox_lora_ckpts/lora_cogvidx5b_scalar_flow_i2v_fix3samples-09-09-05-39/model_config.json
ADDED
|
@@ -0,0 +1,4 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"model_class": "SATVideoDiffusionEngine",
|
| 3 |
+
"model_parallel_size": 1
|
| 4 |
+
}
|
cogvideox_lora_ckpts/lora_cogvidx5b_scalar_flow_i2v_fix3samples-09-09-05-39/training_config.yaml
ADDED
|
@@ -0,0 +1,219 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
model:
|
| 2 |
+
scale_factor: 0.7
|
| 3 |
+
disable_first_stage_autocast: true
|
| 4 |
+
not_trainable_prefixes:
|
| 5 |
+
- all
|
| 6 |
+
log_keys:
|
| 7 |
+
- txt
|
| 8 |
+
denoiser_config:
|
| 9 |
+
target: sgm.modules.diffusionmodules.denoiser.DiscreteDenoiser
|
| 10 |
+
params:
|
| 11 |
+
num_idx: 1000
|
| 12 |
+
quantize_c_noise: false
|
| 13 |
+
weighting_config:
|
| 14 |
+
target: sgm.modules.diffusionmodules.denoiser_weighting.EpsWeighting
|
| 15 |
+
scaling_config:
|
| 16 |
+
target: sgm.modules.diffusionmodules.denoiser_scaling.VideoScaling
|
| 17 |
+
discretization_config:
|
| 18 |
+
target: sgm.modules.diffusionmodules.discretizer.ZeroSNRDDPMDiscretization
|
| 19 |
+
params:
|
| 20 |
+
shift_scale: 1.0
|
| 21 |
+
network_config:
|
| 22 |
+
target: dit_video_concat.DiffusionTransformer
|
| 23 |
+
params:
|
| 24 |
+
time_embed_dim: 512
|
| 25 |
+
elementwise_affine: true
|
| 26 |
+
num_frames: 49
|
| 27 |
+
time_compressed_rate: 4
|
| 28 |
+
latent_width: 90
|
| 29 |
+
latent_height: 60
|
| 30 |
+
num_layers: 42
|
| 31 |
+
patch_size: 2
|
| 32 |
+
in_channels: 16
|
| 33 |
+
out_channels: 16
|
| 34 |
+
hidden_size: 3072
|
| 35 |
+
adm_in_channels: 256
|
| 36 |
+
num_attention_heads: 48
|
| 37 |
+
transformer_args:
|
| 38 |
+
checkpoint_activations: true
|
| 39 |
+
vocab_size: 1
|
| 40 |
+
max_sequence_length: 64
|
| 41 |
+
layernorm_order: pre
|
| 42 |
+
skip_init: false
|
| 43 |
+
model_parallel_size: 1
|
| 44 |
+
is_decoder: false
|
| 45 |
+
modules:
|
| 46 |
+
pos_embed_config:
|
| 47 |
+
target: dit_video_concat.Rotary3DPositionEmbeddingMixin
|
| 48 |
+
params:
|
| 49 |
+
hidden_size_head: 64
|
| 50 |
+
text_length: 226
|
| 51 |
+
lora_config:
|
| 52 |
+
target: sat.model.finetune.lora2.LoraMixin
|
| 53 |
+
params:
|
| 54 |
+
r: 128
|
| 55 |
+
patch_embed_config:
|
| 56 |
+
target: dit_video_concat.ImagePatchEmbeddingMixin
|
| 57 |
+
params:
|
| 58 |
+
text_hidden_size: 4096
|
| 59 |
+
adaln_layer_config:
|
| 60 |
+
target: dit_video_concat.AdaLNMixin
|
| 61 |
+
params:
|
| 62 |
+
qk_ln: true
|
| 63 |
+
final_layer_config:
|
| 64 |
+
target: dit_video_concat.FinalLayerMixin
|
| 65 |
+
conditioner_config:
|
| 66 |
+
target: sgm.modules.GeneralConditioner
|
| 67 |
+
params:
|
| 68 |
+
emb_models:
|
| 69 |
+
- is_trainable: false
|
| 70 |
+
input_key: txt
|
| 71 |
+
ucg_rate: 0.1
|
| 72 |
+
target: sgm.modules.encoders.modules.FrozenT5Embedder
|
| 73 |
+
params:
|
| 74 |
+
model_dir: /path/to/FluidNexusRoot/cogvideox-sat/2b/t5-v1_1-xxl
|
| 75 |
+
max_length: 226
|
| 76 |
+
first_stage_config:
|
| 77 |
+
target: vae_modules.autoencoder.VideoAutoencoderInferenceWrapper
|
| 78 |
+
params:
|
| 79 |
+
cp_size: 1
|
| 80 |
+
ckpt_path: /path/to/FluidNexusRoot/cogvideox-sat/5b/vae/3d-vae.pt
|
| 81 |
+
ignore_keys:
|
| 82 |
+
- loss
|
| 83 |
+
loss_config:
|
| 84 |
+
target: torch.nn.Identity
|
| 85 |
+
regularizer_config:
|
| 86 |
+
target: vae_modules.regularizers.DiagonalGaussianRegularizer
|
| 87 |
+
encoder_config:
|
| 88 |
+
target: vae_modules.cp_enc_dec.ContextParallelEncoder3D
|
| 89 |
+
params:
|
| 90 |
+
double_z: true
|
| 91 |
+
z_channels: 16
|
| 92 |
+
resolution: 256
|
| 93 |
+
in_channels: 3
|
| 94 |
+
out_ch: 3
|
| 95 |
+
ch: 128
|
| 96 |
+
ch_mult:
|
| 97 |
+
- 1
|
| 98 |
+
- 2
|
| 99 |
+
- 2
|
| 100 |
+
- 4
|
| 101 |
+
attn_resolutions: []
|
| 102 |
+
num_res_blocks: 3
|
| 103 |
+
dropout: 0.0
|
| 104 |
+
gather_norm: true
|
| 105 |
+
decoder_config:
|
| 106 |
+
target: vae_modules.cp_enc_dec.ContextParallelDecoder3D
|
| 107 |
+
params:
|
| 108 |
+
double_z: true
|
| 109 |
+
z_channels: 16
|
| 110 |
+
resolution: 256
|
| 111 |
+
in_channels: 3
|
| 112 |
+
out_ch: 3
|
| 113 |
+
ch: 128
|
| 114 |
+
ch_mult:
|
| 115 |
+
- 1
|
| 116 |
+
- 2
|
| 117 |
+
- 2
|
| 118 |
+
- 4
|
| 119 |
+
attn_resolutions: []
|
| 120 |
+
num_res_blocks: 3
|
| 121 |
+
dropout: 0.0
|
| 122 |
+
gather_norm: false
|
| 123 |
+
loss_fn_config:
|
| 124 |
+
target: sgm.modules.diffusionmodules.loss.VideoDiffusionLoss
|
| 125 |
+
params:
|
| 126 |
+
fixed_frames: 3
|
| 127 |
+
offset_noise_level: 0
|
| 128 |
+
sigma_sampler_config:
|
| 129 |
+
target: sgm.modules.diffusionmodules.sigma_sampling.DiscreteSampling
|
| 130 |
+
params:
|
| 131 |
+
uniform_sampling: true
|
| 132 |
+
num_idx: 1000
|
| 133 |
+
discretization_config:
|
| 134 |
+
target: sgm.modules.diffusionmodules.discretizer.ZeroSNRDDPMDiscretization
|
| 135 |
+
params:
|
| 136 |
+
shift_scale: 1.0
|
| 137 |
+
sampler_config:
|
| 138 |
+
target: sgm.modules.diffusionmodules.sampling.VPSDEDPMPP2MSampler
|
| 139 |
+
params:
|
| 140 |
+
num_steps: 50
|
| 141 |
+
verbose: true
|
| 142 |
+
discretization_config:
|
| 143 |
+
target: sgm.modules.diffusionmodules.discretizer.ZeroSNRDDPMDiscretization
|
| 144 |
+
params:
|
| 145 |
+
shift_scale: 1.0
|
| 146 |
+
guider_config:
|
| 147 |
+
target: sgm.modules.diffusionmodules.guiders.DynamicCFG
|
| 148 |
+
params:
|
| 149 |
+
scale: 6
|
| 150 |
+
exp: 5
|
| 151 |
+
num_steps: 50
|
| 152 |
+
args:
|
| 153 |
+
checkpoint_activations: true
|
| 154 |
+
model_parallel_size: 1
|
| 155 |
+
experiment_name: lora_cogvidx5b_scalarflow_all_i2v_fix3samples
|
| 156 |
+
mode: finetune
|
| 157 |
+
load: /path/to/FluidNexusRoot/cogvideox-sat/5b/transformer
|
| 158 |
+
no_load_rng: true
|
| 159 |
+
train_iters: 10000
|
| 160 |
+
eval_iters: 1
|
| 161 |
+
eval_interval: 1000
|
| 162 |
+
eval_batch_size: 1
|
| 163 |
+
save: /path/to/FluidNexusRoot/cogvideox_lora_ckpts
|
| 164 |
+
summary_dir: /path/to/FluidNexusRoot/cogvideox_lora_runs
|
| 165 |
+
save_interval: 500
|
| 166 |
+
log_interval: 100
|
| 167 |
+
train_data:
|
| 168 |
+
- /dev/shm/ScalarFlow_cogvideox_dataset
|
| 169 |
+
valid_data:
|
| 170 |
+
- /dev/shm/ScalarFlow_cogvideox_dataset_sub_90
|
| 171 |
+
split: 1,0,0
|
| 172 |
+
num_workers: 8
|
| 173 |
+
force_train: true
|
| 174 |
+
only_log_video_latents: false
|
| 175 |
+
data:
|
| 176 |
+
target: data_video.SFTDataset
|
| 177 |
+
params:
|
| 178 |
+
video_size:
|
| 179 |
+
- 480
|
| 180 |
+
- 720
|
| 181 |
+
fps: 8
|
| 182 |
+
max_num_frames: 49
|
| 183 |
+
skip_frms_num: 0
|
| 184 |
+
deepspeed:
|
| 185 |
+
train_micro_batch_size_per_gpu: 2
|
| 186 |
+
gradient_accumulation_steps: 1
|
| 187 |
+
steps_per_print: 50
|
| 188 |
+
gradient_clipping: 0.1
|
| 189 |
+
zero_optimization:
|
| 190 |
+
stage: 2
|
| 191 |
+
cpu_offload: false
|
| 192 |
+
contiguous_gradients: false
|
| 193 |
+
overlap_comm: true
|
| 194 |
+
reduce_scatter: true
|
| 195 |
+
reduce_bucket_size: 1000000000
|
| 196 |
+
allgather_bucket_size: 1000000000
|
| 197 |
+
load_from_fp32_weights: false
|
| 198 |
+
zero_allow_untested_optimizer: true
|
| 199 |
+
bf16:
|
| 200 |
+
enabled: true
|
| 201 |
+
fp16:
|
| 202 |
+
enabled: false
|
| 203 |
+
loss_scale: 0
|
| 204 |
+
loss_scale_window: 400
|
| 205 |
+
hysteresis: 2
|
| 206 |
+
min_loss_scale: 1
|
| 207 |
+
optimizer:
|
| 208 |
+
type: sat.ops.FusedEmaAdam
|
| 209 |
+
params:
|
| 210 |
+
lr: 0.001
|
| 211 |
+
betas:
|
| 212 |
+
- 0.9
|
| 213 |
+
- 0.95
|
| 214 |
+
eps: 1.0e-08
|
| 215 |
+
weight_decay: 0.0001
|
| 216 |
+
activation_checkpointing:
|
| 217 |
+
partition_activations: false
|
| 218 |
+
contiguous_memory_optimization: false
|
| 219 |
+
wall_clock_breakdown: false
|
zero123_finetune_logs/2024-04-24T14-04-16_scalar_flow/checkpoints/step=000015499.ckpt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:79155ecf6818fa9c813d3eaf12d91a13a5d57b6c8845ca1ec3b1e0a72c828a49
|
| 3 |
+
size 15465966043
|
zero123_finetune_logs/2024-10-28T18-26-21_fluid_nexus_ball/checkpoints/step=000079999.ckpt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:54b7685a157f2b602312cf438106f8d1269bb3e8cfcfe8b5b11c45230c14818d
|
| 3 |
+
size 15465966616
|
zero123_finetune_logs/2024-10-30T06-31-08_fluid_nexus_smoke/checkpoints/step=000051999.ckpt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:774267ba698138e2208fcb6b047797e51d6159a003131c9e07c67ec9de0710d3
|
| 3 |
+
size 15465974168
|