Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,54 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
+
from transformers import pipeline, AutoTokenizer
|
3 |
+
import torch
|
4 |
+
import os
|
5 |
+
|
6 |
+
|
7 |
+
def load_model(model_name):
|
8 |
+
return pipeline("text-generation", model=model_name, device_map="auto", torch_dtype=torch.bfloat16, trust_remote_code=True, token=os.environ["token"])
|
9 |
+
|
10 |
+
def generate(
|
11 |
+
model_name,
|
12 |
+
user_input,
|
13 |
+
temperature=0.4,
|
14 |
+
top_p=0.95,
|
15 |
+
top_k=50,
|
16 |
+
max_new_tokens=256,
|
17 |
+
):
|
18 |
+
pipe = load_model(model_name)
|
19 |
+
|
20 |
+
# Set tokenize correctly. Otherwise ticking the box breaks it.
|
21 |
+
if model_name == "M4-ai/tau-1.8B":
|
22 |
+
prompt = user_input
|
23 |
+
else:
|
24 |
+
prompt = f"<|im_start|>user\n{user_input}<|im_end|>\n<|im_start|>assistant\n"
|
25 |
+
outputs = pipe(prompt, max_new_tokens=max_new_tokens, do_sample=True,
|
26 |
+
temperature=temperature, top_k=top_k, top_p=top_p, repetition_penalty=1.1)
|
27 |
+
return outputs[0]["generated_text"]
|
28 |
+
|
29 |
+
model_choices = ["LeroyDyer/Mixtral_Instruct","LeroyDyer/Mixtral_Chat","LeroyDyer/Mixtral_Chat_X","LeroyDyer/Mixtral_Chat_X_128k",
|
30 |
+
"LeroyDyer/Mixtral_AI_Instruct_X","LeroyDyer/Mixtral_AI_Thinker",
|
31 |
+
"LeroyDyer/Mixtral_AI_128k_BioMedical","Mixtral_BioMedical","LeroyDyer/Mixtral_Samantha",
|
32 |
+
"LeroyDyer/Mixtral_AI_Base_128k","LeroyDyer/Mixtral_Base","LeroyDyer/Mixtral_AI_1.0",
|
33 |
+
"LeroyDyer/Mixtral_AI_Vision_128K","LeroyDyer/Mixtral_AI_Base","LeroyDyer/Mixtral_AI_Medic_Base",
|
34 |
+
"LeroyDyer/Mixtral_AI_Vision_128K_X","LeroyDyer/Mixtral_AI_Vision_V1_128",
|
35 |
+
"LeroyDyer/Mixtral_AI_Cyber", "LeroyDyer/Mixtral_AI_Cyber_1.0","LeroyDyer/Mixtral_AI_Cyber_2.0",
|
36 |
+
]
|
37 |
+
# What at the best options?
|
38 |
+
g = gr.Interface(
|
39 |
+
fn=generate,
|
40 |
+
inputs=[
|
41 |
+
gr.components.Dropdown(choices=model_choices, label="Model", value=model_choices[0], interactive=True),
|
42 |
+
gr.components.Textbox(lines=2, label="Prompt", value="Write me a Python program that calculates the factorial of a given number."),
|
43 |
+
gr.components.Slider(minimum=0, maximum=1, value=0.4, label="Temperature"),
|
44 |
+
gr.components.Slider(minimum=0, maximum=1, value=0.95, label="Top p"),
|
45 |
+
gr.components.Slider(minimum=0, maximum=100, step=1, value=50, label="Top k"),
|
46 |
+
gr.components.Slider(minimum=1, maximum=2048, step=1, value=1024, label="Max tokens"),
|
47 |
+
],
|
48 |
+
outputs=[gr.Textbox(lines=10, label="Output")],
|
49 |
+
title="SpydazWeb AI (LeroyDyer) Language Models",
|
50 |
+
description="Each has thier own personality's ",
|
51 |
+
concurrency_limit=1
|
52 |
+
)
|
53 |
+
|
54 |
+
g.launch(max_threads=4)
|