lora-scripts代码分析

用fastapi写的前后端，确实是好代码，没用webui了，这样可以直接在内网就打开页面，还是非常实用的，以往用FastAPI封的很简单，这块值得学习。

prepare_environment()->
- prepare_submodules()
- check_dirs()
- validate_requirements()
- setup_windows_bitsandbytes()
- setup_onnxruntime()
run_tag_editor()->
run_tensorboard()->
 
uvicorn.run()->mikazuki.app->
- mikazuki.app.application->
-- app=FastAPI(lifespan)
-- app.include_router(proxy_router) # 可以处理该路由器定义的http请求
-- app.include_router(api_router,prefix="/api")->
--- router = APIRouter()
--- @router.post("/interrogate")
---- async def run_interrogate(req:TaggerInterrogateRequest,background_task)->
---- interrogator = available_interrogators.get(req.interrogator_model, available_interrogators["wd14-convnextv2-v2"])
---- background_tasks.add_task()
----- on_interrogate()->
----- ratings,tags = interrogator.interrogate(image)->
----- processed_tags = Interrogator.postprocess_tags(tag,*postprocess_opts)
 
--- @router.post("/run")
--- async def create_toml_file(request:Request)->
--- json_data = await request.body()->
--- config = json.loads(json_data.decode("utf-8"))->
--- trainer_file = trainer_mapping[model_train_type]->
--- validated, message = train_utils.validate_model(config["pretrained_model_name_or_path"])
--- result = process.run_train(toml_file, trainer_file, gpu_ids, suggest_cpu_threads)
---- python -m accelerate.commands.launch --num_cpu_threads_per_process=4 \
  "/home/image_team/image_team_docker_home/lgd/e_commerce_sd/tools/sd_lora/sd-scripts/train_network.py" \
  --config_file="E:\comprehensive_library\e_commerce_sd\tools\sd_lora\config\lora_sd_sciprts.toml"

主要是两个接口

import sys
import json
import requests
from pathlib import Path
 
root_dir = Path(__file__).resolve().parent.parent
sys.path.append(str(root_dir))
 
interrogator = False
lora_new = True
lora_expert = False
 
input_json_interrogator = {
    "path": "/home/image_team/image_team_docker_home/lgd/common/lora-scripts-main/lion/",
    "interrogator_model": "wd14-convnextv2-v2",
    "threshold": 0.35,
    "additional_tags": "",
    "exclude_tags": "",
    "escape_tag": True,
    "batch_input_recursive": False,
    "batch_output_action_on_conflict": "ignore",
    "replace_underscore": True
}
 
input_json_lora_new = {
    'pretrained_model_name_or_path': '/home/image_team/image_team_docker_home/lgd/e_commerce_sd/stable-diffusion-webui/models/Stable-diffusion/revAnimated_v121.safetensors',
    'train_data_dir': '/home/image_team/image_team_docker_home/lgd/common/lora-scripts-main/lion/',
    'resolution': '512,512',
    'enable_bucket': True,
    'min_bucket_reso': 256,
    'max_bucket_reso': 1024,
    'output_name': 'aki',
    'output_dir': './output',
    'save_model_as': 'safetensors',
    'save_every_n_epochs': 2,
    'max_train_epochs': 10,
    'train_batch_size': 1,
    'network_train_unet_only': False,
    'network_train_text_encoder_only': False,
    'learning_rate': 0.0001,
    'unet_lr': 0.0001,
    'text_encoder_lr': 1e-05,
    'lr_scheduler': 'cosine_with_restarts',
    'optimizer_type': 'AdamW8bit',
    'lr_scheduler_num_cycles': 1,
    'network_module': 'networks.lora',
    'network_dim': 32,
    'network_alpha': 32,
    'logging_dir': './logs',
    'caption_extension': '.txt',
    'shuffle_caption': True,
    'keep_tokens': 0,
    'max_token_length': 255,
    'seed': 1337,
    'prior_loss_weight': 1,
    'clip_skip': 2,
    'mixed_precision': 'fp16',
    'save_precision': 'fp16',
    'xformers': True,
    'cache_latents': True,
    'persistent_data_loader_workers': True,
    'lr_warmup_steps': 0,
    'gpu_ids': ['0', '1', '2', '3']}
 
# input_json_lora_expert = {
#
# }
 
if interrogator:
    results = requests.post("http://10.111.132.198:28000/api/interrogate", json=input_json_interrogator)
    print(results.json())
 
if lora_new:
    results = requests.post("http://10.111.132.198:28000/api/run", json=input_json_lora_new)
    print(results.json())