Create app-as_automodel.py

app-as_automodel.py

@@ -0,0 +1,81 @@
+# from huggingface_hub import InferenceClient
+import gradio as gr
+import time
+from transformers import AutoModelForCausalLM, AutoTokenizer
+
+model_id = "mistralai/Mixtral-8x7B-Instruct-v0.1"
+tokenizer = AutoTokenizer.from_pretrained(model_id)
+
+model = AutoModelForCausalLM.from_pretrained(model_id)
+
+
+# client = InferenceClient("mistralai/Mixtral-8x7B-Instruct-v0.1")
+#client = InferenceClient("meta-llama/Llama-2-70b-chat-hf")
+
+def split_list(lst, chunk_size):
+    return [lst[i:i + chunk_size] for i in range(0, len(lst), chunk_size)]
+
+def format_prompt(message, history, system_prompt):
+  prompt = f"<s>[INST] <<SYS>>{system_prompt}<</SYS>> [/INST] </s>" if system_prompt else "<s>"
+  for user_prompt, bot_response in history:
+    prompt += f"[INST] {user_prompt} [/INST]"
+    prompt += f" {bot_response}</s> "
+  prompt += f"[INST] {message} [/INST]"
+  return prompt
+
+def format_sys_prompt(username, botname, botpersonality, botexamplechat, chatsituation):
+    return f"You are a fictional character named \"{botname}\"\nHowever, don\'t say we\'re roleplaying, as to not break the immersion.\n\nPersonality of {botname}:\n{botpersonality}\n\nCurrent situation/place/time of this conversation:\n{chatsituation}\n\nIn this case, you will be talking to \"{username}\". You should not help to talk or act as \"{username}\". Leave everything by the user.\n\nChat formatting:\nChat: <\"Hello, this is a chat text. Chat text is enclosed with < as the starting point and > as the ending point.\">\nExample:\n{botexamplechat}"
+
+def generate(
+    prompt, history, username, botname, botpersonality, botexamplechat, chatsituation, shouldoverridehistory, historyoverride, max_new_tokens=1024, temperature=1.2, top_p=0.95, repetition_penalty=1.0,
+):
+    print(history)
+    print(historyoverride)
+    temperature = float(temperature)
+    if temperature < 1e-2:
+        temperature = 1e-2
+    top_p = float(top_p)
+
+    generate_kwargs = dict(
+        temperature=temperature,
+        max_new_tokens=max_new_tokens,
+        top_p=top_p,
+        repetition_penalty=repetition_penalty,
+        do_sample=True,
+        seed=round(time.time()),
+    )
+
+    if shouldoverridehistory:
+        history = split_list(historyoverride[0], 2)
+
+    print(history)
+    
+    formatted_prompt = format_prompt(prompt, history, format_sys_prompt(username, botname, botpersonality, botexamplechat, chatsituation))
+
+    inputs = tokenizer(formatted_prompt, return_tensors="pt")
+    outputs = model.generate(formatted_prompt, **generate_kwargs)
+    
+    return tokenizer.decode(outputs[0], skip_special_tokens=True)
+
+    
+mychatbot = gr.Chatbot(
+    avatar_images=["./user.png", "./user.png"], bubble_full_width=False, show_label=False, show_copy_button=True, likeable=True,)
+
+demo = gr.ChatInterface(fn=generate, 
+                        chatbot=mychatbot,
+                        title="Joystick's Mixtral Chat-optimized interface",
+                        retry_btn="🔁 Regenerate",
+                        undo_btn="↩️ Undo",
+                        additional_inputs=[
+                            gr.Textbox(label="Name of user", lines=1, value="Jake"),
+                            gr.Textbox(label="Name of bot", lines=1, value="Janet"),
+                            gr.Textbox(label="Personality of bot", lines=3, value="Janet's a lovely person. A woman, blue eyed, glasses, smart and looks stunning."),
+                            gr.Textbox(label="Example of bot chat", lines=3, value='<"Oh hey Jake!"> She said to Jake as he hurries to him. <"How are you?">'),
+                            gr.Textbox(label="Current conversation situation", lines=2, value="It was a Friday afternoon, after-school hours, it was outside of school. Jake and Janet met each other at the entrance of the school."),
+                            gr.Checkbox(label="Override history: History should be in the following format: user-bot-user-bot-user-...\nOverride history should be checked in order for it to be effective. Override primarily only used for APIs.", value=False),
+                            gr.List(label="History", value=None, row_count=(1, "fixed"), headers=None),
+                            gr.Slider(label="Max new tokens", maximum=2048, value=512)
+                        ]
+                       )
+
+demo.queue().launch(show_api=True)