Spaces:
Sleeping
Sleeping
update known data
Browse files- llm_api/openai_api.py +41 -17
llm_api/openai_api.py
CHANGED
|
@@ -1,33 +1,49 @@
|
|
| 1 |
import json
|
|
|
|
| 2 |
from dotenv import load_dotenv
|
| 3 |
-
from
|
| 4 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 5 |
from .exceptions import RefusalError
|
| 6 |
-
from .
|
| 7 |
|
| 8 |
load_dotenv(override=True)
|
| 9 |
client = OpenAI()
|
| 10 |
|
| 11 |
|
| 12 |
-
def extract_info(img_paths, schema):
|
| 13 |
-
print(
|
| 14 |
text_content = [
|
| 15 |
{
|
| 16 |
"type": "text",
|
| 17 |
"text": EXTRACT_INFO_HUMAN_MESSAGE,
|
| 18 |
},
|
| 19 |
]
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 20 |
image_content = [
|
| 21 |
{
|
| 22 |
"type": "image_url",
|
| 23 |
"image_url": {
|
| 24 |
-
|
| 25 |
},
|
| 26 |
}
|
| 27 |
for img_path in img_paths
|
| 28 |
]
|
| 29 |
response = client.beta.chat.completions.parse(
|
| 30 |
-
|
| 31 |
messages=[
|
| 32 |
{
|
| 33 |
"role": "system",
|
|
@@ -36,7 +52,7 @@ def extract_info(img_paths, schema):
|
|
| 36 |
{
|
| 37 |
"role": "user",
|
| 38 |
"content": text_content + image_content,
|
| 39 |
-
}
|
| 40 |
],
|
| 41 |
max_tokens=1000,
|
| 42 |
response_format=schema,
|
|
@@ -46,25 +62,34 @@ def extract_info(img_paths, schema):
|
|
| 46 |
# top_p=.0000000000000000000001
|
| 47 |
)
|
| 48 |
if response.choices[0].message.refusal:
|
| 49 |
-
raise RefusalError(
|
| 50 |
|
| 51 |
content = response.choices[0].message.content
|
| 52 |
-
parsed_data = json.loads(content)
|
| 53 |
model_data = schema.model_validate(parsed_data)
|
| 54 |
|
| 55 |
return 200, model_data
|
| 56 |
|
| 57 |
|
| 58 |
-
def follow_structure(json_info, schema):
|
| 59 |
-
print(
|
| 60 |
text_content = [
|
| 61 |
{
|
| 62 |
"type": "text",
|
| 63 |
"text": FOLLOW_SCHEMA_HUMAN_MESSAGE.format(json_info=json_info),
|
| 64 |
},
|
| 65 |
]
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 66 |
response = client.beta.chat.completions.parse(
|
| 67 |
-
|
| 68 |
messages=[
|
| 69 |
{
|
| 70 |
"role": "system",
|
|
@@ -73,7 +98,7 @@ def follow_structure(json_info, schema):
|
|
| 73 |
{
|
| 74 |
"role": "user",
|
| 75 |
"content": text_content,
|
| 76 |
-
}
|
| 77 |
],
|
| 78 |
max_tokens=1000,
|
| 79 |
response_format=schema,
|
|
@@ -84,11 +109,10 @@ def follow_structure(json_info, schema):
|
|
| 84 |
)
|
| 85 |
|
| 86 |
if response.choices[0].message.refusal:
|
| 87 |
-
raise RefusalError(
|
| 88 |
|
| 89 |
content = response.choices[0].message.content
|
| 90 |
-
parsed_data = json.loads(content)
|
| 91 |
model_data = schema.model_validate(parsed_data)
|
| 92 |
|
| 93 |
return 200, model_data
|
| 94 |
-
|
|
|
|
| 1 |
import json
|
| 2 |
+
|
| 3 |
from dotenv import load_dotenv
|
| 4 |
+
from openai import BadRequestError, OpenAI
|
| 5 |
+
|
| 6 |
+
from .constants import (
|
| 7 |
+
EXTRACT_INFO_HUMAN_MESSAGE,
|
| 8 |
+
EXTRACT_INFO_SYSTEM_MESSAGE,
|
| 9 |
+
FOLLOW_SCHEMA_HUMAN_MESSAGE,
|
| 10 |
+
FOLLOW_SCHEMA_SYSTEM_MESSAGE,
|
| 11 |
+
)
|
| 12 |
from .exceptions import RefusalError
|
| 13 |
+
from .utils import get_data_format, get_image_data
|
| 14 |
|
| 15 |
load_dotenv(override=True)
|
| 16 |
client = OpenAI()
|
| 17 |
|
| 18 |
|
| 19 |
+
def extract_info(img_paths, schema, known_data=None):
|
| 20 |
+
print("Extracting info via OpenAI...")
|
| 21 |
text_content = [
|
| 22 |
{
|
| 23 |
"type": "text",
|
| 24 |
"text": EXTRACT_INFO_HUMAN_MESSAGE,
|
| 25 |
},
|
| 26 |
]
|
| 27 |
+
|
| 28 |
+
if known_data is not None:
|
| 29 |
+
text_content.append(
|
| 30 |
+
{
|
| 31 |
+
"type": "text",
|
| 32 |
+
"text": f'\nAlso exploit the known data: \n\n"{known_data}"',
|
| 33 |
+
}
|
| 34 |
+
)
|
| 35 |
+
|
| 36 |
image_content = [
|
| 37 |
{
|
| 38 |
"type": "image_url",
|
| 39 |
"image_url": {
|
| 40 |
+
"url": f"data:image/{get_image_data(img_path)};base64,{get_image_data(img_path)}",
|
| 41 |
},
|
| 42 |
}
|
| 43 |
for img_path in img_paths
|
| 44 |
]
|
| 45 |
response = client.beta.chat.completions.parse(
|
| 46 |
+
model="gpt-4o-2024-08-06",
|
| 47 |
messages=[
|
| 48 |
{
|
| 49 |
"role": "system",
|
|
|
|
| 52 |
{
|
| 53 |
"role": "user",
|
| 54 |
"content": text_content + image_content,
|
| 55 |
+
},
|
| 56 |
],
|
| 57 |
max_tokens=1000,
|
| 58 |
response_format=schema,
|
|
|
|
| 62 |
# top_p=.0000000000000000000001
|
| 63 |
)
|
| 64 |
if response.choices[0].message.refusal:
|
| 65 |
+
raise RefusalError("OpenAI refused to respond to the request")
|
| 66 |
|
| 67 |
content = response.choices[0].message.content
|
| 68 |
+
parsed_data = json.loads(content)
|
| 69 |
model_data = schema.model_validate(parsed_data)
|
| 70 |
|
| 71 |
return 200, model_data
|
| 72 |
|
| 73 |
|
| 74 |
+
def follow_structure(json_info, schema, known_data=None):
|
| 75 |
+
print("Following structure via OpenAI...")
|
| 76 |
text_content = [
|
| 77 |
{
|
| 78 |
"type": "text",
|
| 79 |
"text": FOLLOW_SCHEMA_HUMAN_MESSAGE.format(json_info=json_info),
|
| 80 |
},
|
| 81 |
]
|
| 82 |
+
|
| 83 |
+
if known_data is not None:
|
| 84 |
+
text_content.append(
|
| 85 |
+
{
|
| 86 |
+
"type": "text",
|
| 87 |
+
"text": f'\nAlso Exploit the known data: \n\n"{known_data}"',
|
| 88 |
+
}
|
| 89 |
+
)
|
| 90 |
+
|
| 91 |
response = client.beta.chat.completions.parse(
|
| 92 |
+
model="gpt-4o-2024-08-06",
|
| 93 |
messages=[
|
| 94 |
{
|
| 95 |
"role": "system",
|
|
|
|
| 98 |
{
|
| 99 |
"role": "user",
|
| 100 |
"content": text_content,
|
| 101 |
+
},
|
| 102 |
],
|
| 103 |
max_tokens=1000,
|
| 104 |
response_format=schema,
|
|
|
|
| 109 |
)
|
| 110 |
|
| 111 |
if response.choices[0].message.refusal:
|
| 112 |
+
raise RefusalError("OpenAI refused to respond to the request")
|
| 113 |
|
| 114 |
content = response.choices[0].message.content
|
| 115 |
+
parsed_data = json.loads(content)
|
| 116 |
model_data = schema.model_validate(parsed_data)
|
| 117 |
|
| 118 |
return 200, model_data
|
|
|