Spaces:
Running
on
Zero
Running
on
Zero
add : llm wrapper
Browse files- llm_wrapper.py +99 -99
llm_wrapper.py
CHANGED
@@ -1,101 +1,101 @@
|
|
1 |
-
|
2 |
-
|
3 |
-
|
4 |
-
|
5 |
-
|
6 |
-
|
7 |
-
|
8 |
-
|
9 |
-
|
10 |
-
|
11 |
-
|
12 |
-
|
13 |
-
|
14 |
-
|
15 |
-
|
16 |
-
|
17 |
-
|
18 |
-
|
19 |
-
#
|
20 |
-
|
21 |
-
|
22 |
-
|
23 |
-
#
|
24 |
-
|
25 |
-
|
26 |
-
|
27 |
-
|
28 |
-
|
29 |
-
|
30 |
-
#
|
31 |
-
|
32 |
-
|
33 |
-
|
34 |
-
#
|
35 |
-
|
36 |
-
|
37 |
-
|
38 |
-
|
39 |
-
|
40 |
-
#
|
41 |
-
|
42 |
-
|
43 |
-
|
44 |
-
|
45 |
-
#
|
46 |
-
|
47 |
-
|
48 |
-
|
49 |
|
50 |
-
|
51 |
-
|
52 |
-
|
53 |
-
|
54 |
-
|
55 |
-
|
56 |
-
|
57 |
-
|
58 |
-
|
59 |
-
|
60 |
-
|
61 |
-
|
62 |
-
|
63 |
-
|
64 |
-
|
65 |
-
|
66 |
-
|
67 |
-
|
68 |
-
|
69 |
-
|
70 |
-
|
71 |
-
|
72 |
-
|
73 |
-
|
74 |
-
|
75 |
-
|
76 |
-
|
77 |
-
|
78 |
-
|
79 |
-
|
80 |
-
|
81 |
-
|
82 |
-
|
83 |
|
84 |
-
|
85 |
-
|
86 |
-
|
87 |
-
#
|
88 |
-
|
89 |
-
|
90 |
-
|
91 |
-
|
92 |
-
|
93 |
-
|
94 |
-
|
95 |
-
|
96 |
-
|
97 |
-
|
98 |
-
|
99 |
-
|
100 |
-
|
101 |
-
|
|
|
1 |
+
import logging
|
2 |
+
from PIL import Image
|
3 |
+
from io import BytesIO
|
4 |
+
import requests, os, json, time
|
5 |
+
|
6 |
+
from google import genai
|
7 |
+
|
8 |
+
prompt_base_path = "src/llm_wrapper/prompt"
|
9 |
+
client = genai.Client(api_key=os.getenv("GEMINI_API_KEY"))
|
10 |
+
|
11 |
+
|
12 |
+
def encode_image(image_source):
|
13 |
+
"""
|
14 |
+
์ด๋ฏธ์ง ๊ฒฝ๋ก๊ฐ URL์ด๋ ๋ก์ปฌ ํ์ผ์ด๋ Pillow Image ๊ฐ์ฒด์ด๋ ๋์ผํ๊ฒ ์ฒ๋ฆฌํ๋ ํจ์.
|
15 |
+
์ด๋ฏธ์ง๋ฅผ ์ด์ด google.genai.types.Part ๊ฐ์ฒด๋ก ๋ณํํฉ๋๋ค.
|
16 |
+
Pillow์์ ์ง์๋์ง ์๋ ํฌ๋งท์ ๋ํด์๋ ์์ธ๋ฅผ ๋ฐ์์ํต๋๋ค.
|
17 |
+
"""
|
18 |
+
try:
|
19 |
+
# ์ด๋ฏธ Pillow ์ด๋ฏธ์ง ๊ฐ์ฒด์ธ ๊ฒฝ์ฐ ๊ทธ๋๋ก ์ฌ์ฉ
|
20 |
+
if isinstance(image_source, Image.Image):
|
21 |
+
image = image_source
|
22 |
+
else:
|
23 |
+
# URL์์ ์ด๋ฏธ์ง ๋ค์ด๋ก๋
|
24 |
+
if isinstance(image_source, str) and (
|
25 |
+
image_source.startswith("http://")
|
26 |
+
or image_source.startswith("https://")
|
27 |
+
):
|
28 |
+
response = requests.get(image_source)
|
29 |
+
image = Image.open(BytesIO(response.content))
|
30 |
+
# ๋ก์ปฌ ํ์ผ์์ ์ด๋ฏธ์ง ์ด๊ธฐ
|
31 |
+
else:
|
32 |
+
image = Image.open(image_source)
|
33 |
+
|
34 |
+
# ์ด๋ฏธ์ง ํฌ๋งท์ด None์ธ ๊ฒฝ์ฐ (๋ฉ๋ชจ๋ฆฌ์์ ์์ฑ๋ ์ด๋ฏธ์ง ๋ฑ)
|
35 |
+
if image.format is None:
|
36 |
+
image_format = "JPEG"
|
37 |
+
else:
|
38 |
+
image_format = image.format
|
39 |
+
|
40 |
+
# ์ด๋ฏธ์ง ํฌ๋งท์ด ์ง์๋์ง ์๋ ๊ฒฝ์ฐ ์์ธ ๋ฐ์
|
41 |
+
if image_format not in Image.registered_extensions().values():
|
42 |
+
raise ValueError(f"Unsupported image format: {image_format}.")
|
43 |
+
|
44 |
+
buffered = BytesIO()
|
45 |
+
# PIL์์ ์ง์๋์ง ์๋ ํฌ๋งท์ด๋ ๋ค์ํ ์ฑ๋์ RGB๋ก ๋ณํ ํ ์ ์ฅ
|
46 |
+
if image.mode in ("RGBA", "P", "CMYK"): # RGBA, ํ๋ ํธ, CMYK ๋ฑ์ RGB๋ก ๋ณํ
|
47 |
+
image = image.convert("RGB")
|
48 |
+
image.save(buffered, format="JPEG")
|
49 |
|
50 |
+
return genai.types.Part.from_bytes(data=buffered.getvalue(), mime_type="image/jpeg")
|
51 |
+
|
52 |
+
except requests.exceptions.RequestException as e:
|
53 |
+
raise ValueError(f"Failed to download the image from URL: {e}")
|
54 |
+
except IOError as e:
|
55 |
+
raise ValueError(f"Failed to process the image file: {e}")
|
56 |
+
except ValueError as e:
|
57 |
+
raise ValueError(e)
|
58 |
+
|
59 |
+
|
60 |
+
def run_gemini(
|
61 |
+
target_prompt: str,
|
62 |
+
prompt_in_path: str,
|
63 |
+
img_in_data: str = None,
|
64 |
+
model: str = "gemini-2.0-flash",
|
65 |
+
) -> str:
|
66 |
+
"""
|
67 |
+
GEMINI API๋ฅผ ๋๊ธฐ ๋ฐฉ์์ผ๋ก ํธ์ถํ์ฌ ๋ฌธ์์ด ์๋ต์ ๋ฐ์ต๋๋ค.
|
68 |
+
retry ๋
ผ๋ฆฌ๋ ์ ๊ฑฐ๋์์ต๋๋ค.
|
69 |
+
"""
|
70 |
+
with open(os.path.join(prompt_base_path, prompt_in_path), "r", encoding="utf-8") as file:
|
71 |
+
prompt_dict = json.load(file)
|
72 |
+
|
73 |
+
system_prompt = prompt_dict["system_prompt"]
|
74 |
+
user_prompt_head = prompt_dict["user_prompt"]["head"]
|
75 |
+
user_prompt_tail = prompt_dict["user_prompt"]["tail"]
|
76 |
+
|
77 |
+
user_prompt_text = "\n".join([user_prompt_head, target_prompt, user_prompt_tail])
|
78 |
+
input_content = [user_prompt_text]
|
79 |
+
|
80 |
+
if img_in_data is not None:
|
81 |
+
encoded_image = encode_image(img_in_data)
|
82 |
+
input_content.append(encoded_image)
|
83 |
|
84 |
+
logging.info("Requested API for chat completion response (sync call)...")
|
85 |
+
start_time = time.time()
|
86 |
+
|
87 |
+
# ๋๊ธฐ ๋ฐฉ์: client.models.generate_content(...)
|
88 |
+
chat_completion = client.models.generate_content(
|
89 |
+
model=model,
|
90 |
+
contents=input_content,
|
91 |
+
)
|
92 |
+
|
93 |
+
chat_output = chat_completion.parsed
|
94 |
+
input_token = chat_completion.usage_metadata.prompt_token_count
|
95 |
+
output_token = chat_completion.usage_metadata.candidates_token_count
|
96 |
+
pricing = input_token / 1000000 * 0.1 * 1500 + output_token / 1000000 * 0.7 * 1500
|
97 |
+
|
98 |
+
logging.info(
|
99 |
+
f"[GEMINI] Request completed (sync). Time taken: {time.time()-start_time:.2f}s / Pricing(KRW): {pricing:.2f}"
|
100 |
+
)
|
101 |
+
return chat_output, chat_completion
|