Skip to content

Restore legacy action training alongside TI2V inference #7

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
zjn20030811 wants to merge 1 commit into
base: main
Choose a base branch
from
Open

Restore legacy action training alongside TI2V inference #7

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
40 changes: 29 additions & 11 deletions scripts/train.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -42,7 +42,16 @@ def __init__(
# Load models
model_configs = self.parse_model_configs(model_paths, model_id_with_origin_paths, fp8_models=fp8_models, offload_models=offload_models, device=device)
tokenizer_config = ModelConfig(model_id="Wan-AI/Wan2.1-T2V-1.3B", origin_file_pattern="google/umt5-xxl/") if enable_text and tokenizer_path is None else (ModelConfig(tokenizer_path) if enable_text and tokenizer_path else None)
self.pipe = build_wan_video_action_pipeline(torch_dtype=torch.bfloat16, device=device, model_configs=model_configs, tokenizer_config=tokenizer_config, args=args)
self.pipe = build_wan_video_action_pipeline(
torch_dtype=torch.bfloat16,
device=device,
model_configs=model_configs,
tokenizer_config=tokenizer_config,
ckpt_path=ckpt_path,
action_dim=args.action_dim,
action_mode=args.action_mode,
text_enabled=enable_text,
)
self.pipe = self.split_pipeline_units(task, self.pipe, trainable_models, lora_base_model)

# Training mode
Expand Down Expand Up @@ -161,9 +170,19 @@ def wan_parser():
if runtime_config["text_enabled"] and "prompt_emb" in runtime_config["data_file_keys"]:
special_operator_map["prompt_emb"] = ResolvePromptEmbPath(base_path=args.dataset_base_path)

with open(args.action_stat_path, "r") as f:
stats = json.load(f)
stat = {args.action_type: stats[args.action_type]} if args.action_type in stats else stats
stat = None
if runtime_config["action_enabled"]:
if args.action_stat_path is None:
raise ValueError(
"--action_stat_path is required when action conditioning is enabled."
)
with open(args.action_stat_path, "r") as f:
stats = json.load(f)
stat = (
{args.action_type: stats[args.action_type]}
if args.action_type in stats
else stats
)

dataset = UnifiedDataset(
base_path=args.dataset_base_path,
Expand All @@ -185,13 +204,12 @@ def wan_parser():
special_operator_map=special_operator_map,
)

pack_paths(
dataset.data,
("video", "start_frame", "end_frame"),
("action", "start_frame", "end_frame"),
)
path_groups = [("video", "start_frame", "end_frame")]
if runtime_config["action_enabled"]:
path_groups.append(("action", "start_frame", "end_frame"))
pack_paths(dataset.data, *path_groups)

if "action" in runtime_config["data_file_keys"]:
if runtime_config["action_enabled"]:
dataset.special_operator_map["action"] = LoadCobotAction(
base_path=args.dataset_base_path,
action_type=args.action_type,
Expand All @@ -205,7 +223,7 @@ def wan_parser():
model_paths=json.dumps(runtime_config["model_paths_list"]),
model_id_with_origin_paths=args.model_id_with_origin_paths,
tokenizer_path=runtime_config["tokenizer_path"],
enable_text=args.enable_text,
enable_text=runtime_config["text_enabled"],
trainable_models=args.trainable_models,
lora_base_model=args.lora_base_model,
lora_target_modules=args.lora_target_modules,
Expand Down
Loading