Spaces:
Sleeping
Sleeping
Create app-as_automodel.py
Browse files- app-as_automodel.py +81 -0
app-as_automodel.py
ADDED
@@ -0,0 +1,81 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# from huggingface_hub import InferenceClient
|
2 |
+
import gradio as gr
|
3 |
+
import time
|
4 |
+
from transformers import AutoModelForCausalLM, AutoTokenizer
|
5 |
+
|
6 |
+
model_id = "mistralai/Mixtral-8x7B-Instruct-v0.1"
|
7 |
+
tokenizer = AutoTokenizer.from_pretrained(model_id)
|
8 |
+
|
9 |
+
model = AutoModelForCausalLM.from_pretrained(model_id)
|
10 |
+
|
11 |
+
|
12 |
+
# client = InferenceClient("mistralai/Mixtral-8x7B-Instruct-v0.1")
|
13 |
+
#client = InferenceClient("meta-llama/Llama-2-70b-chat-hf")
|
14 |
+
|
15 |
+
def split_list(lst, chunk_size):
|
16 |
+
return [lst[i:i + chunk_size] for i in range(0, len(lst), chunk_size)]
|
17 |
+
|
18 |
+
def format_prompt(message, history, system_prompt):
|
19 |
+
prompt = f"<s>[INST] <<SYS>>{system_prompt}<</SYS>> [/INST] </s>" if system_prompt else "<s>"
|
20 |
+
for user_prompt, bot_response in history:
|
21 |
+
prompt += f"[INST] {user_prompt} [/INST]"
|
22 |
+
prompt += f" {bot_response}</s> "
|
23 |
+
prompt += f"[INST] {message} [/INST]"
|
24 |
+
return prompt
|
25 |
+
|
26 |
+
def format_sys_prompt(username, botname, botpersonality, botexamplechat, chatsituation):
|
27 |
+
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}"
|
28 |
+
|
29 |
+
def generate(
|
30 |
+
prompt, history, username, botname, botpersonality, botexamplechat, chatsituation, shouldoverridehistory, historyoverride, max_new_tokens=1024, temperature=1.2, top_p=0.95, repetition_penalty=1.0,
|
31 |
+
):
|
32 |
+
print(history)
|
33 |
+
print(historyoverride)
|
34 |
+
temperature = float(temperature)
|
35 |
+
if temperature < 1e-2:
|
36 |
+
temperature = 1e-2
|
37 |
+
top_p = float(top_p)
|
38 |
+
|
39 |
+
generate_kwargs = dict(
|
40 |
+
temperature=temperature,
|
41 |
+
max_new_tokens=max_new_tokens,
|
42 |
+
top_p=top_p,
|
43 |
+
repetition_penalty=repetition_penalty,
|
44 |
+
do_sample=True,
|
45 |
+
seed=round(time.time()),
|
46 |
+
)
|
47 |
+
|
48 |
+
if shouldoverridehistory:
|
49 |
+
history = split_list(historyoverride[0], 2)
|
50 |
+
|
51 |
+
print(history)
|
52 |
+
|
53 |
+
formatted_prompt = format_prompt(prompt, history, format_sys_prompt(username, botname, botpersonality, botexamplechat, chatsituation))
|
54 |
+
|
55 |
+
inputs = tokenizer(formatted_prompt, return_tensors="pt")
|
56 |
+
outputs = model.generate(formatted_prompt, **generate_kwargs)
|
57 |
+
|
58 |
+
return tokenizer.decode(outputs[0], skip_special_tokens=True)
|
59 |
+
|
60 |
+
|
61 |
+
mychatbot = gr.Chatbot(
|
62 |
+
avatar_images=["./user.png", "./user.png"], bubble_full_width=False, show_label=False, show_copy_button=True, likeable=True,)
|
63 |
+
|
64 |
+
demo = gr.ChatInterface(fn=generate,
|
65 |
+
chatbot=mychatbot,
|
66 |
+
title="Joystick's Mixtral Chat-optimized interface",
|
67 |
+
retry_btn="🔁 Regenerate",
|
68 |
+
undo_btn="↩️ Undo",
|
69 |
+
additional_inputs=[
|
70 |
+
gr.Textbox(label="Name of user", lines=1, value="Jake"),
|
71 |
+
gr.Textbox(label="Name of bot", lines=1, value="Janet"),
|
72 |
+
gr.Textbox(label="Personality of bot", lines=3, value="Janet's a lovely person. A woman, blue eyed, glasses, smart and looks stunning."),
|
73 |
+
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?">'),
|
74 |
+
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."),
|
75 |
+
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),
|
76 |
+
gr.List(label="History", value=None, row_count=(1, "fixed"), headers=None),
|
77 |
+
gr.Slider(label="Max new tokens", maximum=2048, value=512)
|
78 |
+
]
|
79 |
+
)
|
80 |
+
|
81 |
+
demo.queue().launch(show_api=True)
|