Text chat added to api.py

README.md

@@ -256,6 +256,9 @@ python web_demo_hf.py
 python api.py
 ```
 程序会自动下载 sat 模型，默认部署在本地的 8080 端口，通过 POST 方法进行调用。下面是用`curl`请求的例子，一般而言可以也可以使用代码方法进行POST。
+
+1. 图文对话
+
 ```shell
 echo "{\"image\":\"$(base64 path/to/example.jpg)\",\"text\":\"描述这张图片\",\"history\":[]}" > temp.json
 curl -X POST -H "Content-Type: application/json" -d @temp.json http://127.0.0.1:8080
@@ -270,7 +273,31 @@ curl -X POST -H "Content-Type: application/json" -d @temp.json http://127.0.0.1:
   }
 ```
 
+2. 纯文本对话
+
+```shell
+echo "{\"image\":\"\",\"text\":\"你好\",\"history\":[]}" > temp.json
+curl -X POST -H "Content-Type: application/json" -d @temp.json http://127.0.0.1:8080
+```
+
+得到的返回值为
+
+```
+{
+    "result": "你好！请问有什么需要帮助的吗？", 
+    "history": [
+        [
+            "你好", 
+            "你好！请问有什么需要帮助的吗？"
+        ]
+    ], 
+    "status": 200, 
+    "time": "2023-06-16 15:51:08"
+}
+```
+
 我们也提供了使用Huggingface模型的 [api_hf.py](api_hf.py)，用法和sat模型的api一致：
+
 ```shell
 python api_hf.py
 ```

README_en.md

@@ -187,12 +187,23 @@ First, you need to install additional dependencies pip install fastapi uvicorn,
 python api.py
 ```
 The program will automatically download the sat model, and by default it will be deployed on local port 8080 and called through the POST method. Below is an example of a request with curl, but in general you can also use a code method to POST.
+
+1. Image-text chat
+
 ```shell
 echo "{\"image\":\"$(base64 path/to/example.jpg)\",\"text\":\"Describe this picture\",\"history\":[]}" > temp.json
 curl -X POST -H "Content-Type: application/json" -d @temp.json http://127.0.0.1:8080
 ```
 
+2. Text chat
+
+```shell
+echo "{\"image\":\"\",\"text\":\"Hello\",\"history\":[]}" > temp.json
+curl -X POST -H "Content-Type: application/json" -d @temp.json http://127.0.0.1:8080
+```
+
 We also provide an api_hf.py that uses the Huggingface model, which works the same way as the sat model's api:
+
 ```shell
 python api_hf.py
 ```

api.py

@@ -29,8 +29,10 @@ async def visual_glm(request: Request):
     input_para.update(request_data)
 
     is_zh = is_chinese(input_text)
-    input_data = generate_input(input_text, input_image_encoded, history, input_para)
-    input_image, gen_kwargs =  input_data['input_image'], input_data['gen_kwargs']
+    image_is_encoded = False if input_image_encoded == "" else True
+    input_data = generate_input(input_text, input_image_encoded, history, input_para, image_is_encoded)
+    input_image = None if input_image_encoded == "" else input_data["input_image"]
+    gen_kwargs = input_data["gen_kwargs"]
     with torch.no_grad():
         answer, history, _ = chat(None, model, tokenizer, input_text, history=history, image=input_image, \
                             max_length=gen_kwargs['max_length'], top_p=gen_kwargs['top_p'], \