feat: 实现模型动态注册机制

scripts/demo/generate_ar.py

@@ -2,24 +2,32 @@ from pathlib import Path
 
 import torch
 
-from astrai.config.param_config import ModelParameter
 from astrai.inference import InferenceEngine
+from astrai.model import AutoModel
+from astrai.tokenize import AutoTokenizer
 
 PROJECT_ROOT = Path(__file__).resolve().parents[2]
 PARAMETER_ROOT = Path(PROJECT_ROOT, "params")
 
 
 def generate_text():
-    param = ModelParameter.load(PARAMETER_ROOT, disable_init=True)
-    param.to(device="cuda", dtype=torch.bfloat16)
+    # Load model from pretrained
+    model = AutoModel.from_pretrained(PARAMETER_ROOT)
+    model.to(device="cuda", dtype=torch.bfloat16)
+
+    # Load tokenizer from pretrained
+    tokenizer = AutoTokenizer.from_pretrained(PARAMETER_ROOT / "tokenizer")
 
     query = input(">> ")
 
-    engine = InferenceEngine(param)
+    engine = InferenceEngine(
+        model=model,
+        tokenizer=tokenizer,
+    )
     response = engine.generate(
         prompt=query,
         stream=False,
-        max_tokens=param.config.max_len,
+        max_tokens=2048,
         temperature=0.8,
         top_p=0.95,
         top_k=50,