update

app.py

@@ -3,7 +3,8 @@ import requests
 import time
 from ast import literal_eval
 
-    
+
+@st.cache
 def infer(
     prompt, 
     model_name, 
@@ -15,7 +16,7 @@ def infer(
     seed=42,
     stop="\n"
 ):
-
+    
     model_name_map = {
         "GPT-JT-6B-v1": "Together-gpt-JT-6B-v1",
     }
@@ -33,39 +34,34 @@ def infer(
     assert 0.0 <= top_p <= 1.0
     
     if temperature == 0.0:
-        temperature = 1.0
-        top_k = 1
-    
-    result = await st.session_state.together_web3.language_model_inference(
-        from_dict(
-            data_class=LanguageModelInferenceRequest,
-            data={
-                "model": model_name_map[model_name],
-                "max_tokens": max_new_tokens,
-                "prompt": prompt,
-                "n": num_completions,
-                "temperature": temperature,
-                "top_k": top_k,
-                "top_p": top_p,
-                "stop": stop,
-                "seed": seed,
-                "echo": False,
-            }
-        ),
-    )
-    
-    generated_text = result.choices[0].text
+        temperature = 0.01
+
+    my_post_dict = {
+        "model": "Together-gpt-JT-6B-v1",
+        "prompt": prompt,
+        "top_p": top_p,
+        "top_k": top_k,
+        "temperature": temperature,
+        "max_tokens": max_new_tokens,
+        "stop": stop,
+    }
+    response = requests.get("https://staging.together.xyz/api/inference", params=my_post_dict).json()
+    generated_text = response['output']['choices'][0]['text']
     
     for stop_word in stop:
-        if stop_word in result:
+        if stop_word in generated_text:
             generated_text = generated_text[:generated_text.find(stop_word)]
+            
+    st.session_state.updated = True
     
     return generated_text
 
+
 def set_preset():
     if st.session_state.preset == "Classification":
         
-        st.session_state.prompt = '''Please classify the given sentence.
+        if not st.session_state.updated:
+            st.session_state.prompt = '''Please classify the given sentence.
 Possible labels:
 1. <label_0>
 2. <label_1>
@@ -77,15 +73,18 @@ Input: <sentence_1>
 Label:'''
         st.session_state.temperature = "0.0"
         st.session_state.top_p = "1.0"
+        st.session_state.max_new_tokens = "10"
         
     elif st.session_state.preset == "Generation":
         
-        st.session_state.prompt = '''Please write a story given keywords.
+        if not st.session_state.updated:
+            st.session_state.prompt = '''Please write a story given keywords.
 
 Input: bear, honey
-Story:'''
-        st.session_state.temperature = "1.0"
-        st.session_state.top_p = "0.5"
+Story: Once upon a time,'''
+        st.session_state.temperature = "0.8"
+        st.session_state.top_p = "0.9"
+        st.session_state.max_new_tokens = "100"
     
     else:
         pass
@@ -97,10 +96,10 @@ def main():
         st.session_state.preset = "Classification"
 
     if 'prompt' not in st.session_state:
-        st.session_state.prompt = "Please answer the following question:\n\nQuestion: Where is Zurich?\nAnswer:"
+        st.session_state.prompt = "Please answer the following question:\n\nQuestion: In which country is Zurich located?\nAnswer:"
 
     if 'temperature' not in st.session_state:
-        st.session_state.temperature = "0.0"
+        st.session_state.temperature = "0.8"
 
     if 'top_p' not in st.session_state:
         st.session_state.top_p = "1.0"
@@ -108,8 +107,11 @@ def main():
     if 'top_k' not in st.session_state:
         st.session_state.top_k = "40"
         
-    if 'together_web3' not in st.session_state:
-        st.session_state.together_web3 = TogetherWeb3()
+    if 'max_new_tokens' not in st.session_state:
+        st.session_state.max_new_tokens = "10"
+        
+    if 'updated' not in st.session_state:
+        st.session_state.updated = False
         
 
     st.title("GPT-JT")
@@ -118,7 +120,7 @@ def main():
 
     with col1:
         model_name = st.selectbox("Model", ["GPT-JT-6B-v1"])
-        max_new_tokens = st.text_input('Max new tokens', "10")
+        max_new_tokens = st.text_input('Max new tokens', st.session_state.max_new_tokens)
         temperature = st.text_input('temperature', st.session_state.temperature)
         top_k = st.text_input('top_k', st.session_state.top_k)
         top_p = st.text_input('top_p', st.session_state.top_p)
@@ -130,19 +132,20 @@ def main():
 
     with col2:
 
-        preset = st.radio(
-            "Recommended Configurations", 
-            ('Classification', 'Generation'), 
-            on_change=set_preset,
-            key="preset",
-            horizontal=True
-        )
+        # preset = st.radio(
+        #     "Recommended Templates", 
+        #     ('Classification', 'Generation'), 
+        #     on_change=set_preset,
+        #     key="preset",
+        #     horizontal=True
+        # )
 
-        prompt = st.text_area(
+        prompt_area = st.empty()
+        prompt = prompt_area.text_area(
             "Prompt",
             value=st.session_state.prompt,
             max_chars=4096,
-            height=400,
+            height=300,
         )
 
         generated_area = st.empty()
@@ -153,12 +156,10 @@ def main():
         if button_submit:
             generated_area.text(prompt)
             report_text = infer(
-                prompt, model_name=model_name, max_new_tokens=max_new_tokens, temperature=temperature, top_p=top_p, top_k=top_k,
+                st.session_state.prompt, model_name=model_name, max_new_tokens=max_new_tokens, temperature=temperature, top_p=top_p, top_k=top_k,
                 num_completions=num_completions, seed=seed, stop=literal_eval("'''"+stop+"'''"),
             )
             generated_area.text(prompt + report_text)
-
-        
         
 if __name__ == '__main__':
-    main()
+    main()

requirements.txt

@@ -0,0 +1 @@
+together_web3