Add Training-Service for local sft and unify sft for local and online

lazyllm/cli/run.py

@@ -1,7 +1,10 @@
 import sys
 import argparse
 import json
+
+import lazyllm
 from lazyllm.engine.lightengine import LightEngine
+from lazyllm.tools.train_service.serve import TrainServer
 
 # lazyllm run xx.json / xx.dsl / xx.lazyml
 # lazyllm run chatbot --model=xx --framework=xx --source=xx
@@ -47,9 +50,17 @@ def graph(json_file):
         res = engine.run(eid, query)
         print(f'answer: {res}')
 
+def training_service():
+    train_server = TrainServer()
+    local_server = lazyllm.ServerModule(train_server, launcher=lazyllm.launcher.EmptyLauncher(sync=False))
+    local_server.start()
+    local_server()
+    local_server.wait()
+
 def run(commands):
     if not commands:
-        print('Usage:\n  lazyllm run graph.json\n  lazyllm run chatbot\n  lazyllm run rag\n')
+        print('Usage:\n  lazyllm run graph.json\n  lazyllm run chatbot\n  '
+              'lazyllm run rag\n  lazyllm run training_service\n')
 
     parser = argparse.ArgumentParser(description='lazyllm deploy command')
     parser.add_argument('command', type=str, help='command')
@@ -75,6 +86,8 @@ def run(commands):
             rag(llm, args.documents)
     elif args.command.endswith('.json'):
         graph(args.command)
+    elif args.command == 'training_service':
+        training_service()
     else:
         print('lazyllm run is not ready yet.')
         sys.exit(0)

lazyllm/components/finetune/alpacalora.py

@@ -40,7 +40,7 @@ def __init__(self,
                  **kw
                  ):
         if not merge_path:
-            save_path = os.path.join(os.getcwd(), target_path)
+            save_path = os.path.join(lazyllm.config['train_target_root'], target_path)
             target_path, merge_path = os.path.join(save_path, "lazyllm_lora"), os.path.join(save_path, "lazyllm_merge")
             os.system(f'mkdir -p {target_path} {merge_path}')
         super().__init__(

lazyllm/components/finetune/collie.py

@@ -39,7 +39,7 @@ def __init__(self,
                  **kw
                  ):
         if not merge_path:
-            save_path = os.path.join(os.getcwd(), target_path)
+            save_path = os.path.join(lazyllm.config['train_target_root'], target_path)
             target_path, merge_path = os.path.join(save_path, "lazyllm_lora"), os.path.join(save_path, "lazyllm_merge")
             os.system(f'mkdir -p {target_path} {merge_path}')
         super().__init__(

lazyllm/components/finetune/llamafactory.py

@@ -2,6 +2,8 @@
 import yaml
 import json
 import tempfile
+import random
+from datetime import datetime
 
 import lazyllm
 from lazyllm import launchers, ArgsDict, thirdparty, CaseInsensitiveDict
@@ -30,7 +32,7 @@ def __init__(self,
             if os.path.exists(defatult_path):
                 base_model = defatult_path
         if not merge_path:
-            save_path = os.path.join(os.getcwd(), target_path)
+            save_path = os.path.join(lazyllm.config['train_target_root'], target_path)
             target_path, merge_path = os.path.join(save_path, "lazyllm_lora"), os.path.join(save_path, "lazyllm_merge")
             os.system(f'mkdir -p {target_path} {merge_path}')
         super().__init__(
@@ -73,9 +75,10 @@ def __init__(self,
         self.export_dict['export_dir'] = merge_path
         self.export_dict['template'] = self.template_dict['template']
 
-        self.temp_folder = os.path.join(os.getcwd(), '.temp')
+        self.temp_folder = os.path.join(lazyllm.config['temp_dir'], 'llamafactory_config')
         if not os.path.exists(self.temp_folder):
             os.makedirs(self.temp_folder)
+        self.log_file_path = None
 
     def get_template_name(self, base_model):
         try:
@@ -144,8 +147,12 @@ def cmd(self, trainset, valset=None) -> str:
         updated_template_str = yaml.dump(dict(self.template_dict), default_flow_style=False)
         self.temp_yaml_file = self.build_temp_yaml(updated_template_str)
 
+        formatted_date = datetime.now().strftime("%Y-%m-%d_%H-%M-%S")
+        random_value = random.randint(1000, 9999)
+        self.log_file_path = f'{self.target_path}/train_log_{formatted_date}_{random_value}.log'
+
         cmds = f'llamafactory-cli train {self.temp_yaml_file}'
-        cmds += f' 2>&1 | tee {self.target_path}/llm_$(date +"%Y-%m-%d_%H-%M-%S").log'
+        cmds += f' 2>&1 | tee {self.log_file_path}'
         if self.temp_export_yaml_file:
             cmds += f' && llamafactory-cli export {self.temp_export_yaml_file}'
         return cmds

lazyllm/components/stable_diffusion/stable_diffusion3.py

@@ -20,7 +20,7 @@ def __init__(self, base_sd, source=None, embed_batch_size=30, trust_remote_code=
         self.trust_remote_code = trust_remote_code
         self.sd = None
         self.init_flag = lazyllm.once_flag()
-        self.save_path = save_path if save_path else os.path.join(os.getcwd(), '.temp/sd3')
+        self.save_path = save_path or os.path.join(lazyllm.config['temp_dir'], 'sd3')
         if init:
             lazyllm.call_once(self.init_flag, self.load_sd)
 

lazyllm/components/text_to_speech/bark.py

@@ -17,7 +17,7 @@ def __init__(self, base_path, source=None, trust_remote_code=True, save_path=Non
         self.processor, self.bark = None, None
         self.init_flag = lazyllm.once_flag()
         self.device = 'cpu'
-        self.save_path = save_path if save_path else os.path.join(os.getcwd(), '.temp/bark')
+        self.save_path = save_path or os.path.join(lazyllm.config['temp_dir'], 'bark')
         if init:
             lazyllm.call_once(self.init_flag, self.load_bark)
 

lazyllm/components/text_to_speech/chattts.py

@@ -17,7 +17,7 @@ def __init__(self, base_path, source=None, save_path=None, init=False):
         self.init_flag = lazyllm.once_flag()
         self.device = 'cpu'
         self.seed = 1024
-        self.save_path = save_path if save_path else os.path.join(os.getcwd(), '.temp/chattts')
+        self.save_path = save_path or os.path.join(lazyllm.config['temp_dir'], 'chattts')
         if init:
             lazyllm.call_once(self.init_flag, self.load_tts)
 

lazyllm/components/text_to_speech/musicgen.py

@@ -14,7 +14,7 @@ def __init__(self, base_path, source=None, save_path=None, init=False):
         self.base_path = ModelManager(source).download(base_path)
         self.model = None
         self.init_flag = lazyllm.once_flag()
-        self.save_path = save_path if save_path else os.path.join(os.getcwd(), '.temp/musicgen')
+        self.save_path = save_path or os.path.join(lazyllm.config['temp_dir'], 'musicgen')
         if init:
             lazyllm.call_once(self.init_flag, self.load_tts)
 

lazyllm/configs.py

@@ -91,4 +91,6 @@ def refresh(self, targets: Union[str, List[str]] = None) -> None:
                 ).add('repr_ml', bool, False, 'REPR_USE_ML'
                 ).add('rag_store', str, 'none', 'RAG_STORE'
                 ).add('gpu_type', str, 'A100', 'GPU_TYPE'
+                ).add('train_target_root', str, os.path.join(os.getcwd(), 'save_ckpt'), 'TRAIN_TARGET_ROOT'
+                ).add('temp_dir', str, os.path.join(os.getcwd(), '.temp'), 'TEMP_DIR'
                 )