WinterGYC
commited on
Commit
·
38f2963
1
Parent(s):
d409029
Init
Browse files- README.md +1 -1
- app.py +118 -0
- requirements.txt +5 -0
README.md
CHANGED
|
@@ -1,6 +1,6 @@
|
|
| 1 |
---
|
| 2 |
title: Baichuan 13B Chat Int8 Gradio
|
| 3 |
-
emoji:
|
| 4 |
colorFrom: indigo
|
| 5 |
colorTo: blue
|
| 6 |
sdk: gradio
|
|
|
|
| 1 |
---
|
| 2 |
title: Baichuan 13B Chat Int8 Gradio
|
| 3 |
+
emoji: ⚡
|
| 4 |
colorFrom: indigo
|
| 5 |
colorTo: blue
|
| 6 |
sdk: gradio
|
app.py
ADDED
|
@@ -0,0 +1,118 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
from transformers import AutoModel, AutoTokenizer
|
| 2 |
+
import gradio as gr
|
| 3 |
+
import mdtex2html
|
| 4 |
+
from utils import load_model_on_gpus
|
| 5 |
+
|
| 6 |
+
model = AutoModelForCausalLM.from_pretrained(
|
| 7 |
+
"baichuan-inc/Baichuan-13B-Chat",
|
| 8 |
+
torch_dtype=torch.float16,
|
| 9 |
+
device_map="auto",
|
| 10 |
+
trust_remote_code=True
|
| 11 |
+
)
|
| 12 |
+
model.generation_config = GenerationConfig.from_pretrained(
|
| 13 |
+
"baichuan-inc/Baichuan-13B-Chat"
|
| 14 |
+
)
|
| 15 |
+
tokenizer = AutoTokenizer.from_pretrained(
|
| 16 |
+
"baichuan-inc/Baichuan-13B-Chat",
|
| 17 |
+
use_fast=False,
|
| 18 |
+
trust_remote_code=True
|
| 19 |
+
)
|
| 20 |
+
model = model.quantize(8).cuda()
|
| 21 |
+
model = model.eval()
|
| 22 |
+
|
| 23 |
+
"""Override Chatbot.postprocess"""
|
| 24 |
+
|
| 25 |
+
|
| 26 |
+
def postprocess(self, y):
|
| 27 |
+
if y is None:
|
| 28 |
+
return []
|
| 29 |
+
for i, (message, response) in enumerate(y):
|
| 30 |
+
y[i] = (
|
| 31 |
+
None if message is None else mdtex2html.convert((message)),
|
| 32 |
+
None if response is None else mdtex2html.convert(response),
|
| 33 |
+
)
|
| 34 |
+
return y
|
| 35 |
+
|
| 36 |
+
|
| 37 |
+
gr.Chatbot.postprocess = postprocess
|
| 38 |
+
|
| 39 |
+
|
| 40 |
+
def parse_text(text):
|
| 41 |
+
"""copy from https://github.com/GaiZhenbiao/ChuanhuChatGPT/"""
|
| 42 |
+
lines = text.split("\n")
|
| 43 |
+
lines = [line for line in lines if line != ""]
|
| 44 |
+
count = 0
|
| 45 |
+
for i, line in enumerate(lines):
|
| 46 |
+
if "```" in line:
|
| 47 |
+
count += 1
|
| 48 |
+
items = line.split('`')
|
| 49 |
+
if count % 2 == 1:
|
| 50 |
+
lines[i] = f'<pre><code class="language-{items[-1]}">'
|
| 51 |
+
else:
|
| 52 |
+
lines[i] = f'<br></code></pre>'
|
| 53 |
+
else:
|
| 54 |
+
if i > 0:
|
| 55 |
+
if count % 2 == 1:
|
| 56 |
+
line = line.replace("`", "\`")
|
| 57 |
+
line = line.replace("<", "<")
|
| 58 |
+
line = line.replace(">", ">")
|
| 59 |
+
line = line.replace(" ", " ")
|
| 60 |
+
line = line.replace("*", "*")
|
| 61 |
+
line = line.replace("_", "_")
|
| 62 |
+
line = line.replace("-", "-")
|
| 63 |
+
line = line.replace(".", ".")
|
| 64 |
+
line = line.replace("!", "!")
|
| 65 |
+
line = line.replace("(", "(")
|
| 66 |
+
line = line.replace(")", ")")
|
| 67 |
+
line = line.replace("$", "$")
|
| 68 |
+
lines[i] = "<br>"+line
|
| 69 |
+
text = "".join(lines)
|
| 70 |
+
return text
|
| 71 |
+
|
| 72 |
+
|
| 73 |
+
def predict(input, chatbot, max_length, top_p, temperature, history, past_key_values):
|
| 74 |
+
chatbot.append((parse_text(input), ""))
|
| 75 |
+
for response, history, past_key_values in model.stream_chat(tokenizer, input, history, past_key_values=past_key_values,
|
| 76 |
+
return_past_key_values=True,
|
| 77 |
+
max_length=max_length, top_p=top_p,
|
| 78 |
+
temperature=temperature):
|
| 79 |
+
chatbot[-1] = (parse_text(input), parse_text(response))
|
| 80 |
+
|
| 81 |
+
yield chatbot, history, past_key_values
|
| 82 |
+
|
| 83 |
+
|
| 84 |
+
def reset_user_input():
|
| 85 |
+
return gr.update(value='')
|
| 86 |
+
|
| 87 |
+
|
| 88 |
+
def reset_state():
|
| 89 |
+
return [], [], None
|
| 90 |
+
|
| 91 |
+
|
| 92 |
+
with gr.Blocks() as demo:
|
| 93 |
+
gr.HTML("""<h1 align="center">BaiChuan-13B-Int8</h1>""")
|
| 94 |
+
|
| 95 |
+
chatbot = gr.Chatbot()
|
| 96 |
+
with gr.Row():
|
| 97 |
+
with gr.Column(scale=4):
|
| 98 |
+
with gr.Column(scale=12):
|
| 99 |
+
user_input = gr.Textbox(show_label=False, placeholder="Input...", lines=10).style(
|
| 100 |
+
container=False)
|
| 101 |
+
with gr.Column(min_width=32, scale=1):
|
| 102 |
+
submitBtn = gr.Button("Submit", variant="primary")
|
| 103 |
+
with gr.Column(scale=1):
|
| 104 |
+
emptyBtn = gr.Button("Clear History")
|
| 105 |
+
max_length = gr.Slider(0, 32768, value=8192, step=1.0, label="Maximum length", interactive=True)
|
| 106 |
+
top_p = gr.Slider(0, 1, value=0.8, step=0.01, label="Top P", interactive=True)
|
| 107 |
+
temperature = gr.Slider(0, 1, value=0.95, step=0.01, label="Temperature", interactive=True)
|
| 108 |
+
|
| 109 |
+
history = gr.State([])
|
| 110 |
+
past_key_values = gr.State(None)
|
| 111 |
+
|
| 112 |
+
submitBtn.click(predict, [user_input, chatbot, max_length, top_p, temperature, history, past_key_values],
|
| 113 |
+
[chatbot, history, past_key_values], show_progress=True)
|
| 114 |
+
submitBtn.click(reset_user_input, [], [user_input])
|
| 115 |
+
|
| 116 |
+
emptyBtn.click(reset_state, outputs=[chatbot, history, past_key_values], show_progress=True)
|
| 117 |
+
|
| 118 |
+
demo.queue().launch(share=False, inbrowser=True)
|
requirements.txt
ADDED
|
@@ -0,0 +1,5 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
accelerate
|
| 2 |
+
colorama
|
| 3 |
+
cpm_kernels
|
| 4 |
+
sentencepiece
|
| 5 |
+
transformers_stream_generator
|