Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
@@ -1,16 +1,61 @@
|
|
1 |
import gradio as gr
|
|
|
|
|
|
|
|
|
|
|
2 |
import os
|
3 |
-
from
|
4 |
token = os.environ["HF_TOKEN"]
|
5 |
-
tokenizer = AutoTokenizer.from_pretrained("google/gemma-2b",token=token)
|
6 |
-
model = AutoModelForCausalLM.from_pretrained("google/gemma-2b",token=token)
|
7 |
-
streamer = TextStreamer(tokenizer,skip_prompt=True)
|
8 |
|
|
|
|
|
|
|
|
|
|
|
9 |
|
10 |
-
def generate(inputs,history):
|
11 |
-
inputs = tokenizer([inputs], return_tensors="pt")
|
12 |
-
yield model.generate(**inputs, streamer=streamer)
|
13 |
|
|
|
14 |
|
15 |
-
|
16 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
import gradio as gr
|
2 |
+
import torch
|
3 |
+
from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer
|
4 |
+
import time
|
5 |
+
import numpy as np
|
6 |
+
from torch.nn import functional as F
|
7 |
import os
|
8 |
+
from threading import Thread
|
9 |
token = os.environ["HF_TOKEN"]
|
|
|
|
|
|
|
10 |
|
11 |
+
model = AutoModelForCausalLM.from_pretrained("google/gemma-2b-it", torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32,token=token)
|
12 |
+
tok = AutoTokenizer.from_pretrained("google/gemma-2b-it",token=token)
|
13 |
+
# using CUDA for an optimal experience
|
14 |
+
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
|
15 |
+
model = model.to(device)
|
16 |
|
|
|
|
|
|
|
17 |
|
18 |
+
start_message = ""
|
19 |
|
20 |
+
def user(message, history):
|
21 |
+
# Append the user's message to the conversation history
|
22 |
+
return "", history + [[message, ""]]
|
23 |
+
|
24 |
+
|
25 |
+
def chat(message, history):
|
26 |
+
chat = []
|
27 |
+
for item in history:
|
28 |
+
chat.append({"role": "user", "content": item[0]})
|
29 |
+
if item[1] is not None:
|
30 |
+
chat.append({"role": "assistant", "content": item[1]})
|
31 |
+
chat.append({"role": "user", "content": message})
|
32 |
+
messages = tok.apply_chat_template(chat, tokenize=False, add_generation_prompt=True)
|
33 |
+
# Tokenize the messages string
|
34 |
+
model_inputs = tok([messages], return_tensors="pt").to(device)
|
35 |
+
streamer = TextIteratorStreamer(
|
36 |
+
tok, timeout=10., skip_prompt=True, skip_special_tokens=True)
|
37 |
+
generate_kwargs = dict(
|
38 |
+
model_inputs,
|
39 |
+
streamer=streamer,
|
40 |
+
max_new_tokens=1024,
|
41 |
+
do_sample=True,
|
42 |
+
top_p=0.95,
|
43 |
+
top_k=1000,
|
44 |
+
temperature=0.75,
|
45 |
+
num_beams=1,
|
46 |
+
)
|
47 |
+
t = Thread(target=model.generate, kwargs=generate_kwargs)
|
48 |
+
t.start()
|
49 |
+
|
50 |
+
# Initialize an empty string to store the generated text
|
51 |
+
partial_text = ""
|
52 |
+
for new_text in streamer:
|
53 |
+
# print(new_text)
|
54 |
+
partial_text += new_text
|
55 |
+
# Yield an empty string to cleanup the message textbox and the updated conversation history
|
56 |
+
yield partial_text
|
57 |
+
|
58 |
+
|
59 |
+
|
60 |
+
demo = gr.ChatInterface(fn=chat, examples=[["Write me a poem about Machine Learning."]], title="gemma 2b-it")
|
61 |
+
demo.launch()
|