Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
@@ -1,117 +1,8 @@
|
|
1 |
import gradio as gr
|
2 |
-
from datasets import load_dataset
|
3 |
|
4 |
-
import os
|
5 |
-
import spaces
|
6 |
-
from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer
|
7 |
-
import torch
|
8 |
-
from threading import Thread
|
9 |
-
from sentence_transformers import SentenceTransformer
|
10 |
-
from datasets import load_dataset
|
11 |
-
import time
|
12 |
|
13 |
-
token = os.environ["HF_TOKEN"]
|
14 |
-
model = AutoModelForCausalLM.from_pretrained(
|
15 |
-
"google/gemma-7b-it",
|
16 |
-
# torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32,
|
17 |
-
torch_dtype=torch.float16,
|
18 |
-
token=token,
|
19 |
-
)
|
20 |
-
tokenizer = AutoTokenizer.from_pretrained("google/gemma-7b-it", token=token)
|
21 |
-
device = torch.device("cuda")
|
22 |
-
model = model.to(device)
|
23 |
-
RAG = SentenceTransformer("mixedbread-ai/mxbai-embed-large-v1")
|
24 |
-
TOP_K = 1
|
25 |
-
HEADER = "\n# RESOURCES:\n"
|
26 |
-
# prepare data
|
27 |
-
# since data is too big we will only select the first 3K lines
|
28 |
-
|
29 |
-
data = load_dataset("not-lain/wikipedia-small-3000-embedded", split="train")
|
30 |
-
|
31 |
-
# index dataset
|
32 |
-
data.add_faiss_index("embedding")
|
33 |
-
|
34 |
-
|
35 |
-
def search(query: str, k: int = TOP_K):
|
36 |
-
embedded_query = RAG.encode(query)
|
37 |
-
scores, retrieved_examples = data.get_nearest_examples(
|
38 |
-
"embedding", embedded_query, k=k
|
39 |
-
)
|
40 |
-
return retrieved_examples
|
41 |
-
|
42 |
-
|
43 |
-
def prepare_prompt(query, retrieved_examples):
|
44 |
-
prompt = (
|
45 |
-
f"Query: {query}\nContinue to answer the query in short sentences by using the Search Results:\n"
|
46 |
-
)
|
47 |
-
urls = []
|
48 |
-
titles = retrieved_examples["title"][::-1]
|
49 |
-
texts = retrieved_examples["text"][::-1]
|
50 |
-
urls = retrieved_examples["url"][::-1]
|
51 |
-
titles = titles[::-1]
|
52 |
-
for i in range(TOP_K):
|
53 |
-
prompt += f"* {texts[i]}\n"
|
54 |
-
return prompt, zip(titles, urls)
|
55 |
-
|
56 |
-
|
57 |
-
@spaces.GPU(duration=150)
|
58 |
def talk(message, history):
|
59 |
-
|
60 |
-
print("message ", message)
|
61 |
-
print("searching dataset ...")
|
62 |
-
retrieved_examples = search(message)
|
63 |
-
print("preparing prompt ...")
|
64 |
-
message, metadata = prepare_prompt(message, retrieved_examples)
|
65 |
-
resources = HEADER
|
66 |
-
print("preparing metadata ...")
|
67 |
-
for title, url in metadata:
|
68 |
-
resources += f"[{title}]({url}), "
|
69 |
-
print("preparing chat template ...")
|
70 |
-
chat = []
|
71 |
-
for item in history:
|
72 |
-
chat.append({"role": "user", "content": item[0]})
|
73 |
-
cleaned_past = item[1].split(HEADER)[0]
|
74 |
-
chat.append({"role": "assistant", "content": cleaned_past})
|
75 |
-
chat.append({"role": "user", "content": message})
|
76 |
-
messages = tokenizer.apply_chat_template(
|
77 |
-
chat, tokenize=False, add_generation_prompt=True
|
78 |
-
)
|
79 |
-
print("chat template prepared, ", messages)
|
80 |
-
print("tokenizing input ...")
|
81 |
-
# Tokenize the messages string
|
82 |
-
model_inputs = tokenizer([messages], return_tensors="pt").to(device)
|
83 |
-
streamer = TextIteratorStreamer(
|
84 |
-
tokenizer, timeout=10.0, skip_prompt=True, skip_special_tokens=True
|
85 |
-
)
|
86 |
-
generate_kwargs = dict(
|
87 |
-
model_inputs,
|
88 |
-
streamer=streamer,
|
89 |
-
max_new_tokens=1024,
|
90 |
-
do_sample=True,
|
91 |
-
top_p=0.95,
|
92 |
-
top_k=1000,
|
93 |
-
temperature=0.75,
|
94 |
-
num_beams=1,
|
95 |
-
)
|
96 |
-
print("initializing thread ...")
|
97 |
-
t = Thread(target=model.generate, kwargs=generate_kwargs)
|
98 |
-
t.start()
|
99 |
-
time.sleep(1)
|
100 |
-
# Initialize an empty string to store the generated text
|
101 |
-
partial_text = ""
|
102 |
-
i = 0
|
103 |
-
while t.is_alive():
|
104 |
-
try:
|
105 |
-
for new_text in streamer:
|
106 |
-
if new_text is not None:
|
107 |
-
partial_text += new_text
|
108 |
-
yield partial_text
|
109 |
-
except Exception as e:
|
110 |
-
print(f"retry number {i}\n LOGS:\n")
|
111 |
-
i+=1
|
112 |
-
print(e, e.args)
|
113 |
-
partial_text += resources
|
114 |
-
yield partial_text
|
115 |
|
116 |
|
117 |
TITLE = "# RAG"
|
@@ -127,6 +18,8 @@ Resources used to build this project :
|
|
127 |
* chatbot : https://huggingface.co/google/gemma-7b-it
|
128 |
|
129 |
If you want to support my work consider clicking on the heart react button ❤️🤗
|
|
|
|
|
130 |
"""
|
131 |
|
132 |
|
|
|
1 |
import gradio as gr
|
|
|
2 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
3 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
4 |
def talk(message, history):
|
5 |
+
return "hi"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
6 |
|
7 |
|
8 |
TITLE = "# RAG"
|
|
|
18 |
* chatbot : https://huggingface.co/google/gemma-7b-it
|
19 |
|
20 |
If you want to support my work consider clicking on the heart react button ❤️🤗
|
21 |
+
|
22 |
+
(testing the ui)
|
23 |
"""
|
24 |
|
25 |
|