| output_dir = "/workspace/outputs/Complete" | |
| dataset = "/workspace/configs/Complete/dataset.toml" | |
| epochs = 10 | |
| micro_batch_size_per_gpu = 1 | |
| pipeline_stages = 1 | |
| gradient_accumulation_steps = 2 | |
| gradient_clipping = 1.0 | |
| warmup_steps = 0 | |
| eval_every_n_epochs = 1 | |
| eval_before_first_step = true | |
| eval_micro_batch_size_per_gpu = 1 | |
| eval_gradient_accumulation_steps = 1 | |
| save_every_n_epochs = 1 | |
| checkpoint_every_n_minutes = 120 | |
| activation_checkpointing = true | |
| partition_method = "parameters" | |
| save_dtype = "bfloat16" | |
| caching_batch_size = 4 | |
| steps_per_print = 1 | |
| video_clip_mode = "single_beginning" | |
| [model] | |
| type = "hunyuan-video" | |
| dtype = "bfloat16" | |
| transformer_path = "/workspace/models/hunyuan/hunyuan_video_720_cfgdistill_fp8_e4m3fn.safetensors" | |
| transformer_dtype = "float8" | |
| vae_path = "/workspace/models/hunyuan/hunyuan_video_vae_bf16.safetensors" | |
| llm_path = "/workspace/models/hunyuan/llava-llama-3-8b-text-encoder-tokenizer" | |
| clip_path = "/workspace/models/hunyuan/clip-vit-large-patch14" | |
| timestep_sample_method = "logit_normal" | |
| [adapter] | |
| type = "lora" | |
| rank = 16 | |
| dtype = "bfloat16" | |
| [optimizer] | |
| type = "adamw8bitKahan" | |
| lr = 0.0002 | |
| weight_decay = 0.01 | |
| eps = 1e-08 | |
| betas = [0.9, 0.99] | |
| [monitoring] | |
| enable_wandb = false | |
| log_dir = "/workspace/outputs/Complete" | |