[update] fp16 inference

eval_llm.py

@@ -27,7 +27,7 @@ def init_model(args):
     else:
         model = AutoModelForCausalLM.from_pretrained(args.load_from, trust_remote_code=True)
     get_model_params(model, model.config)
-    return model.eval().to(args.device), tokenizer
+    return model.half().eval().to(args.device), tokenizer
 
 def main():
     parser = argparse.ArgumentParser(description="MiniMind模型推理与对话")

scripts/eval_toolcall.py

@@ -64,7 +64,7 @@ def init_model(args):
     else:
         model = AutoModelForCausalLM.from_pretrained(args.load_from, trust_remote_code=True)
     get_model_params(model, model.config)
-    return model.eval().to(args.device), tokenizer
+    return model.half().eval().to(args.device), tokenizer
 
 
 def parse_tool_calls(text):

scripts/serve_openai_api.py

@@ -44,7 +44,7 @@ def init_model(args):
     else:
         model = AutoModelForCausalLM.from_pretrained(args.load_from, trust_remote_code=True)
     print(f'MiniMind模型参数量: {sum(p.numel() for p in model.parameters()) / 1e6:.2f} M(illion)')
-    return model.eval().to(device), tokenizer
+    return model.half().eval().to(device), tokenizer
 
 
 class ChatRequest(BaseModel):

scripts/web_demo.py

@@ -205,7 +205,7 @@ def load_model_tokenizer(model_path):
         model_path,
         trust_remote_code=True
     )
-    model = model.eval().to(device)
+    model = model.half().eval().to(device)
     return model, tokenizer