Update app.py

app.py

@@ -7,6 +7,7 @@ nvmlInit()
 gpu_h = nvmlDeviceGetHandleByIndex(0)
 ctx_limit = 1024
 gen_limit = 500
+gen_limit_long = 800
 title = "RWKV-x060-World-7B-v2.1-20240507-ctx4096"
 
 os.environ["RWKV_JIT_ON"] = '1'
@@ -23,8 +24,8 @@ from rwkv.utils import PIPELINE, PIPELINE_ARGS
 pipeline = PIPELINE(model, "rwkv_vocab_v20230424")
 
 args = model.args
-eng_name = 'rwkv-x060-eng_single_round_qa-7B-20240430-ctx1024'
-chn_name = 'rwkv-x060-chn_single_round_qa-7B-20240511-ctx1024'
+eng_name = 'rwkv-x060-eng_single_round_qa-7B-20240516-ctx2048'
+chn_name = 'rwkv-x060-chn_single_round_qa-7B-20240516-ctx2048'
 eng_file = hf_hub_download(repo_id="BlinkDL/temp-latest-training-models", filename=f"{eng_name}.pth")
 chn_file = hf_hub_download(repo_id="BlinkDL/temp-latest-training-models", filename=f"{chn_name}.pth")
 state_eng_raw = torch.load(eng_file)
@@ -223,22 +224,22 @@ examples = [
 ]
 
 examples_eng = [
-    ["How can I craft an engaging story featuring vampires on Mars?", gen_limit, 1, 0.2, 0.3, 0.3],
-    ["Compare the business models of Apple and Google.", gen_limit, 1, 0.2, 0.3, 0.3],
-    ["In JSON format, list the top 5 tourist attractions in Paris.", gen_limit, 1, 0.2, 0.3, 0.3],
-    ["Write an outline for a fantasy novel where dreams can alter reality.", gen_limit, 1, 0.2, 0.3, 0.3],
-    ["Can fish get thirsty?", gen_limit, 1, 0.2, 0.3, 0.3],
-    ["Write a Bash script to check disk usage and send alerts if it's too high.", gen_limit, 1, 0.2, 0.3, 0.3],
-    ["Write a simple website in HTML. When a user clicks the button, it shows a random joke from a list of 4 jokes.", gen_limit, 1, 0.2, 0.3, 0.3],
+    ["How can I craft an engaging story featuring vampires on Mars?", gen_limit_long, 1, 0.2, 0.3, 0.3],
+    ["Compare the business models of Apple and Google.", gen_limit_long, 1, 0.2, 0.3, 0.3],
+    ["In JSON format, list the top 5 tourist attractions in Paris.", gen_limit_long, 1, 0.2, 0.3, 0.3],
+    ["Write an outline for a fantasy novel where dreams can alter reality.", gen_limit_long, 1, 0.2, 0.3, 0.3],
+    ["Can fish get thirsty?", gen_limit_long, 1, 0.2, 0.3, 0.3],
+    ["Write a Bash script to check disk usage and send alerts if it's too high.", gen_limit_long, 1, 0.2, 0.3, 0.3],
+    ["Write a simple website in HTML. When a user clicks the button, it shows a random joke from a list of 4 jokes.", gen_limit_long, 1, 0.2, 0.3, 0.3],
 ]
 
 examples_chn = [
-    ["怎样写一个在火星上的吸血鬼的有趣故事？", gen_limit, 1, 0.2, 0.3, 0.3],
-    ["比较苹果和谷歌的商业模式。", gen_limit, 1, 0.2, 0.3, 0.3],
-    ["鱼会口渴吗？", gen_limit, 1, 0.2, 0.3, 0.3],
-    ["以 JSON 格式解释冰箱是如何工作的。", gen_limit, 1, 0.2, 0.3, 0.3],
-    ["编写一个Bash脚本来检查磁盘使用情况，如果使用量过高则发送警报。", gen_limit, 1, 0.2, 0.3, 0.3],
-    ["用HTML编写一个简单的网站。当用户点击按钮时，从4个笑话的列表中随机显示一个笑话。", gen_limit, 1, 0.2, 0.3, 0.3],
+    ["怎样写一个在火星上的吸血鬼的有趣故事？", gen_limit_long, 1, 0.2, 0.3, 0.3],
+    ["比较苹果和谷歌的商业模式。", gen_limit_long, 1, 0.2, 0.3, 0.3],
+    ["鱼会口渴吗？", gen_limit_long, 1, 0.2, 0.3, 0.3],
+    ["以 JSON 格式解释冰箱是如何工作的。", gen_limit_long, 1, 0.2, 0.3, 0.3],
+    ["编写一个Bash脚本来检查磁盘使用情况，如果使用量过高则发送警报。", gen_limit_long, 1, 0.2, 0.3, 0.3],
+    ["用HTML编写一个简单的网站。当用户点击按钮时，从4个笑话的列表中随机显示一个笑话。", gen_limit_long, 1, 0.2, 0.3, 0.3],
 ]
 
 ##########################################################################
@@ -271,7 +272,7 @@ with gr.Blocks(title=title) as demo:
         with gr.Row():
             with gr.Column():
                 prompt = gr.Textbox(lines=2, label="Prompt", value="How can I craft an engaging story featuring vampires on Mars?")
-                token_count = gr.Slider(10, gen_limit, label="Max Tokens", step=10, value=gen_limit)
+                token_count = gr.Slider(10, gen_limit_long, label="Max Tokens", step=10, value=gen_limit_long)
                 temperature = gr.Slider(0.2, 2.0, label="Temperature", step=0.1, value=1.0)
                 top_p = gr.Slider(0.0, 1.0, label="Top P", step=0.05, value=0.2)
                 presence_penalty = gr.Slider(0.0, 1.0, label="Presence Penalty", step=0.1, value=0.3)
@@ -291,7 +292,7 @@ with gr.Blocks(title=title) as demo:
         with gr.Row():
             with gr.Column():
                 prompt = gr.Textbox(lines=2, label="Prompt", value="怎样写一个在火星上的吸血鬼的有趣故事？")
-                token_count = gr.Slider(10, gen_limit, label="Max Tokens", step=10, value=gen_limit)
+                token_count = gr.Slider(10, gen_limit_long, label="Max Tokens", step=10, value=gen_limit_long)
                 temperature = gr.Slider(0.2, 2.0, label="Temperature", step=0.1, value=1.0)
                 top_p = gr.Slider(0.0, 1.0, label="Top P", step=0.05, value=0.2)
                 presence_penalty = gr.Slider(0.0, 1.0, label="Presence Penalty", step=0.1, value=0.3)