Upload app.py

app.py

@@ -1,15 +1,24 @@
 import gradio as gr
-from transformers import AutoModelForCausalLM, AutoTokenizer
+from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig
 import torch
 
-# 加载DeepSeek-Coder-6.7B模型
-model_name = "deepseek/DeepSeek-Coder-6.7B"
-tokenizer = AutoTokenizer.from_pretrained(model_name)
+# 配置 4-bit 量化以适配 16GB GPU
+quantization_config = BitsAndBytesConfig(
+    load_in_4bit=True,
+    bnb_4bit_compute_dtype=torch.float16,
+    bnb_4bit_quant_type="nf4",
+    bnb_4bit_use_double_quant=True
+)
+
+# 加载 DeepSeek-Coder-6.7B-Instruct 模型
+model_name = "deepseek-ai/deepseek-coder-6.7b-instruct"
+tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)
 model = AutoModelForCausalLM.from_pretrained(
     model_name,
-    torch_dtype=torch.float16,  # 使用半精度以减少显存
-    device_map="auto",         # 自动分配到GPU
-    trust_remote_code=True     # DeepSeek模型可能需要
+    quantization_config=quantization_config,
+    device_map="auto",
+    trust_remote_code=True,
+    low_cpu_mem_usage=True
 )
 
 def respond(
@@ -30,7 +39,7 @@ def respond(
 
     messages.append({"role": "user", "content": message})
 
-    # 格式化输入
+    # 使用聊天模板格式化输入
     input_text = tokenizer.apply_chat_template(messages, tokenize=False)
     inputs = tokenizer(input_text, return_tensors="pt").to("cuda")
 
@@ -41,11 +50,12 @@ def respond(
         temperature=temperature,
         top_p=top_p,
         do_sample=True,
+        pad_token_id=tokenizer.eos_token_id
     )
     response = tokenizer.decode(outputs[0], skip_special_tokens=True)
     yield response
 
-# Gradio界面
+# Gradio 界面
 demo = gr.ChatInterface(
     respond,
     additional_inputs=[