Upload 20251103_03-44-07/backup/config_orig/modal.toml with huggingface_hub
Browse files
20251103_03-44-07/backup/config_orig/modal.toml
ADDED
|
@@ -0,0 +1,91 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[model]
|
| 2 |
+
type = "wan"
|
| 3 |
+
|
| 4 |
+
[deploy]
|
| 5 |
+
mode = "wan"
|
| 6 |
+
|
| 7 |
+
[debug]
|
| 8 |
+
only_test_lora = false
|
| 9 |
+
|
| 10 |
+
[core]
|
| 11 |
+
gpu_type = "H100"
|
| 12 |
+
gpu_count = 1
|
| 13 |
+
timeout_hours = 3
|
| 14 |
+
gpu_type_test = "A100-40GB"
|
| 15 |
+
gpu_count_test = 1
|
| 16 |
+
timeout_hours_test = 1
|
| 17 |
+
[[core.volumes]]
|
| 18 |
+
name = "wan-training-results"
|
| 19 |
+
path = "/root/diffusion-pipe/wan-lora"
|
| 20 |
+
|
| 21 |
+
[[core.volumes]]
|
| 22 |
+
name = "hf-cache"
|
| 23 |
+
path = "/root/.cache/huggingface"
|
| 24 |
+
|
| 25 |
+
[[core.volumes]]
|
| 26 |
+
name = "comfy-output"
|
| 27 |
+
path = "/root/comfy/ComfyUI/output"
|
| 28 |
+
|
| 29 |
+
[training]
|
| 30 |
+
resume = false
|
| 31 |
+
resume_folder = ""
|
| 32 |
+
dataset_dir = "Dataset"
|
| 33 |
+
epochs = 10
|
| 34 |
+
micro_batch_size = 2
|
| 35 |
+
gradient_accum_steps = 4
|
| 36 |
+
warmup_steps = 50
|
| 37 |
+
learning_rate = 0.0001
|
| 38 |
+
weight_decay = 0.01
|
| 39 |
+
lora_rank = 16
|
| 40 |
+
lora_dtype = "bfloat16"
|
| 41 |
+
save_every_n_epochs = 5
|
| 42 |
+
checkpoint_mode = "epochs"
|
| 43 |
+
checkpoint_frequency = 5
|
| 44 |
+
pipeline_stages = 1
|
| 45 |
+
gradient_clipping = 1.0
|
| 46 |
+
activation_checkpointing = true
|
| 47 |
+
eval_every_n_epochs = 1
|
| 48 |
+
caching_batch_size = 2
|
| 49 |
+
steps_per_print = 1
|
| 50 |
+
|
| 51 |
+
[inference]
|
| 52 |
+
api_endpoint = "https://nervopastorale--lora-training-comfyapi-endpoint.modal.run"
|
| 53 |
+
workflow_path = "config/wan_img2video_lora_workflow_api.json"
|
| 54 |
+
test_enabled = false
|
| 55 |
+
test_strength = 1.0
|
| 56 |
+
test_height = 830
|
| 57 |
+
test_width = 480
|
| 58 |
+
test_frames = 81
|
| 59 |
+
test_steps = 30
|
| 60 |
+
test_folder = ""
|
| 61 |
+
test_mode = "latest"
|
| 62 |
+
test_epochs = []
|
| 63 |
+
test_latest_n = 1
|
| 64 |
+
test_prompts = [ "The video shows a water bottle being crushed by a hydraulic press. The blue water bottle is placed between two circular metal plates, which sit on a larger metal surface. The hydraulic press descends from above, slowly compressing the water bottle. As the hydraulic press head moves down, the water bottle begins to deform and leak water. By the end of the video, the water bottle is completely crushed. The background is a green metal wall with some yellow and black hazard markings.", "The video shows a red car being crushed by a hydraulic press. The red car is placed in white background. The hydraulic MCH01 press descends from above, slowly compressing the red car. As the MCH01 hydraulic press head moves down, the red car begins to deform and debris from the red car splashed. By the end of the video, the red car is completely crushed. The background is a green metal wall with some yellow and black hazard markings.",]
|
| 65 |
+
|
| 66 |
+
[hf]
|
| 67 |
+
auto_upload = false
|
| 68 |
+
private_repo = true
|
| 69 |
+
upload_test_outputs = true
|
| 70 |
+
upload_tensorboard = false
|
| 71 |
+
force_redownload = false
|
| 72 |
+
skip_existing = true
|
| 73 |
+
upload_endpoint = "https://nervopastorale--lora-training-upload.modal.run"
|
| 74 |
+
download_endpoint = "https://nervopastorale--lora-training-download.modal.run"
|
| 75 |
+
repo_upload = ""
|
| 76 |
+
|
| 77 |
+
[training.dataset]
|
| 78 |
+
resolutions = [ 720,]
|
| 79 |
+
enable_ar_bucket = false
|
| 80 |
+
min_ar = 1.68
|
| 81 |
+
max_ar = 1.88
|
| 82 |
+
num_ar_buckets = 1
|
| 83 |
+
frame_buckets = [ 1, 41, 81,]
|
| 84 |
+
num_repeats = 2
|
| 85 |
+
|
| 86 |
+
[models.training]
|
| 87 |
+
transformer = "https://huggingface.co/Kijai/HunyuanVideo_comfy/blob/main/hunyuan_video_720_cfgdistill_fp8_e4m3fn.safetensors"
|
| 88 |
+
vae = "https://huggingface.co/Kijai/HunyuanVideo_comfy/blob/main/hunyuan_video_vae_bf16.safetensors"
|
| 89 |
+
clip = "https://huggingface.co/openai/clip-vit-large-patch14/tree/main"
|
| 90 |
+
llm = "https://huggingface.co/Kijai/llava-llama-3-8b-text-encoder-tokenizer/tree/main"
|
| 91 |
+
ckpt_path = "https://huggingface.co/Wan-AI/Wan2.1-I2V-14B-480P"
|