add gitignore

.gitignore

@@ -0,0 +1 @@
+cache/

app.py

@@ -10,16 +10,14 @@ import os
 import gradio as gr
 import requests
 import random
-# from dotenv import load_dotenv
 import googletrans
 translator = googletrans.Translator()
 
-# load_dotenv()
 model = None
 tokenizer = None
 generator = None
 
-os.environ["CUDA_VISIBLE_DEVICES"]="0"
+os.environ["CUDA_VISIBLE_DEVICES"]=""
 
 def load_model(model_name, eight_bit=0, device_map="auto"):
     global model, tokenizer, generator
@@ -32,20 +30,29 @@ def load_model(model_name, eight_bit=0, device_map="auto"):
     gpu_count = torch.cuda.device_count()
     print('gpu_count', gpu_count)
 
+    if torch.cuda.is_available():
+        torch_dtype = torch.float16
+    else:
+        torch_dtype = torch.float32
+        
     print(model_name)
     tokenizer = transformers.LLaMATokenizer.from_pretrained(model_name)
     model = transformers.LLaMAForCausalLM.from_pretrained(
         model_name,
         #device_map=device_map,
         #device_map="auto",
-        torch_dtype=torch.float16,
+        torch_dtype=torch_dtype,
         #max_memory = {0: "14GB", 1: "14GB", 2: "14GB", 3: "14GB",4: "14GB",5: "14GB",6: "14GB",7: "14GB"},
         #load_in_8bit=eight_bit,
         #from_tf=True,
         low_cpu_mem_usage=True,
         load_in_8bit=False,
         cache_dir="cache"
-    ).cuda()
+    )
+    if torch.cuda.is_available():
+        model = model.cuda()
+    else:
+        model = model.cpu()
     generator = model.generate
 
 # chat doctor
@@ -68,7 +75,11 @@ def chatdoctor(input, state):
     print('fulltext: ',fulltext)
 
     generated_text = ""
-    gen_in = tokenizer(fulltext, return_tensors="pt").input_ids.cuda()
+    gen_in = tokenizer(fulltext, return_tensors="pt").input_ids
+    if torch.cuda.is_available():
+        gen_in = gen_in.cuda()
+    else:
+        gen_in = gen_in.cpu()
     in_tokens = len(gen_in)
     print('len token',in_tokens)
     with torch.no_grad():
@@ -97,9 +108,22 @@ def chatdoctor(input, state):
 
 def predict(input, chatbot, state):
     print('predict state: ', state)
-    en_input = translator.translate(input, src='ko', dest='en').text
+    
+    # input에 한국어가 detect 되면 영어로 변경, 아니면 그대로
+    is_kor = True
+    if googletrans.Translator().detect(input).lang == 'ko':
+        en_input = translator.translate(input, src='ko', dest='en').text
+    else:
+        en_input = input
+        is_kor = False
+        
     response = chatdoctor(en_input, state)
-    ko_response = translator.translate(response, src='en', dest='ko').text
+
+    if is_kor:
+        ko_response = translator.translate(response, src='en', dest='ko').text
+    else:
+        ko_response = response
+        
     state.append(response)
     chatbot.append((input, ko_response))
     return chatbot, state