Commit
·
a9d8d74
1
Parent(s):
fcd223a
step 2
Browse files- app.py +6 -3
- app/core/prompts.py +11 -0
- app/services/base.py +150 -1
- app/services/service_openai.py +89 -0
app.py
CHANGED
@@ -86,7 +86,7 @@ async def forward_request(
|
|
86 |
service = AIServiceFactory.get_service(ai_vendor)
|
87 |
|
88 |
try:
|
89 |
-
json_attributes = await service.extract_attributes_with_validation(
|
90 |
Product, # type: ignore
|
91 |
ai_model,
|
92 |
None,
|
@@ -101,7 +101,7 @@ async def forward_request(
|
|
101 |
shutil.rmtree(request_temp_folder)
|
102 |
|
103 |
gr.Info("Process completed!")
|
104 |
-
return json_attributes
|
105 |
|
106 |
|
107 |
def add_attribute_schema(attributes, attr_name, attr_desc, attr_type, allowed_values):
|
@@ -380,6 +380,9 @@ with gr.Blocks(title="Internal Demo for Attribution") as demo:
|
|
380 |
output_json = gr.Json(
|
381 |
label="Extracted Attributes", value={}, show_indices=False
|
382 |
)
|
|
|
|
|
|
|
383 |
|
384 |
# add_btn.click(
|
385 |
# add_attribute_schema,
|
@@ -390,7 +393,7 @@ with gr.Blocks(title="Internal Demo for Attribution") as demo:
|
|
390 |
submit_btn.click(
|
391 |
forward_request,
|
392 |
inputs=[attributes, product_taxnomy, product_data, ai_model, gallery],
|
393 |
-
outputs=output_json,
|
394 |
)
|
395 |
|
396 |
|
|
|
86 |
service = AIServiceFactory.get_service(ai_vendor)
|
87 |
|
88 |
try:
|
89 |
+
json_attributes, reevaluated = await service.extract_attributes_with_validation(
|
90 |
Product, # type: ignore
|
91 |
ai_model,
|
92 |
None,
|
|
|
101 |
shutil.rmtree(request_temp_folder)
|
102 |
|
103 |
gr.Info("Process completed!")
|
104 |
+
return json_attributes, reevaluated
|
105 |
|
106 |
|
107 |
def add_attribute_schema(attributes, attr_name, attr_desc, attr_type, allowed_values):
|
|
|
380 |
output_json = gr.Json(
|
381 |
label="Extracted Attributes", value={}, show_indices=False
|
382 |
)
|
383 |
+
reevaluated_output_json = gr.Json(
|
384 |
+
label="Extracted Attributes", value={}, show_indices=False
|
385 |
+
)
|
386 |
|
387 |
# add_btn.click(
|
388 |
# add_attribute_schema,
|
|
|
393 |
submit_btn.click(
|
394 |
forward_request,
|
395 |
inputs=[attributes, product_taxnomy, product_data, ai_model, gallery],
|
396 |
+
outputs=[output_json, reevaluated_output_json],
|
397 |
)
|
398 |
|
399 |
|
app/core/prompts.py
CHANGED
@@ -28,6 +28,13 @@ You should use the following product data to assist you, if available:
|
|
28 |
If an attribute appears in both the image and the product data, use the value from the product data.
|
29 |
"""
|
30 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
31 |
|
32 |
class Prompts(BaseSettings):
|
33 |
EXTRACT_INFO_SYSTEM_MESSAGE: str = EXTRACT_INFO_SYSTEM
|
@@ -42,6 +49,10 @@ class Prompts(BaseSettings):
|
|
42 |
|
43 |
GET_PERCENTAGE_HUMAN_MESSAGE: str = GET_PERCENTAGE_HUMAN
|
44 |
|
|
|
|
|
|
|
|
|
45 |
|
46 |
# Create a cached instance of settings
|
47 |
@lru_cache
|
|
|
28 |
If an attribute appears in both the image and the product data, use the value from the product data.
|
29 |
"""
|
30 |
|
31 |
+
REEVALUATE_SYSTEM = "You are an expert in structured data extraction. You will be given an image or a set of images of a product and set of attributes and should reevaluate certainity of the attributes into the given structure."
|
32 |
+
|
33 |
+
REEVALUATE_HUMAN = """Reevaluate the following attributes of the main product (or {product_taxonomy}) shown in the images. Here are the attributes to reevaluate:
|
34 |
+
{product_data}
|
35 |
+
|
36 |
+
If an attribute can have multiple values, do not need to reevaluate the values, just the attribute itself. If an attribute can have only one value, reevaluate the top three values.
|
37 |
+
"""
|
38 |
|
39 |
class Prompts(BaseSettings):
|
40 |
EXTRACT_INFO_SYSTEM_MESSAGE: str = EXTRACT_INFO_SYSTEM
|
|
|
49 |
|
50 |
GET_PERCENTAGE_HUMAN_MESSAGE: str = GET_PERCENTAGE_HUMAN
|
51 |
|
52 |
+
REEVALUATE_SYSTEM_MESSAGE: str = REEVALUATE_SYSTEM
|
53 |
+
|
54 |
+
REEVALUATE_HUMAN_MESSAGE: str = REEVALUATE_HUMAN
|
55 |
+
|
56 |
|
57 |
# Create a cached instance of settings
|
58 |
@lru_cache
|
app/services/base.py
CHANGED
@@ -11,6 +11,116 @@ from app.schemas.schema_tools import (
|
|
11 |
)
|
12 |
|
13 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
14 |
class BaseAttributionService(ABC):
|
15 |
@abstractmethod
|
16 |
async def extract_attributes(
|
@@ -23,6 +133,17 @@ class BaseAttributionService(ABC):
|
|
23 |
) -> Dict[str, Any]:
|
24 |
pass
|
25 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
26 |
@abstractmethod
|
27 |
async def follow_schema(
|
28 |
self, schema: Dict[str, Any], data: Dict[str, Any]
|
@@ -52,9 +173,37 @@ class BaseAttributionService(ABC):
|
|
52 |
# pil_images=pil_images, # temporarily removed to save cost
|
53 |
img_paths=img_paths,
|
54 |
)
|
|
|
55 |
validate_json_data(data, schema)
|
56 |
|
57 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
58 |
|
59 |
async def follow_schema_with_validation(
|
60 |
self, schema: Dict[str, Any], data: Dict[str, Any]
|
|
|
11 |
)
|
12 |
|
13 |
|
14 |
+
example_data = example_data = {
|
15 |
+
"length": {
|
16 |
+
"maxi": 100,
|
17 |
+
"knee_length": 0,
|
18 |
+
"mini": 0,
|
19 |
+
"midi": 0
|
20 |
+
},
|
21 |
+
"style": {
|
22 |
+
"a_line": 0,
|
23 |
+
"bodycon": 0,
|
24 |
+
"shirt_dress": 0,
|
25 |
+
"wrap_dress": 0,
|
26 |
+
"slip": 0,
|
27 |
+
"smock": 0,
|
28 |
+
"corset": 100,
|
29 |
+
"jumper_dress": 0,
|
30 |
+
"blazer_dress": 0,
|
31 |
+
"asymmetric": 0,
|
32 |
+
"shift": 0,
|
33 |
+
"drop_waist": 0,
|
34 |
+
"empire": 0,
|
35 |
+
"modest": 0
|
36 |
+
},
|
37 |
+
"sleeve_length": {
|
38 |
+
"sleeveless": 0,
|
39 |
+
"three_quarters_sleeve": 0,
|
40 |
+
"long_sleeve": 0,
|
41 |
+
"short_sleeve": 0,
|
42 |
+
"strapless": 100
|
43 |
+
},
|
44 |
+
"neckline": {
|
45 |
+
"v_neck": 0,
|
46 |
+
"sweetheart": 100,
|
47 |
+
"round_neck": 0,
|
48 |
+
"halter_neck": 0,
|
49 |
+
"square_neck": 0,
|
50 |
+
"high_neck": 0,
|
51 |
+
"crew_neck": 0,
|
52 |
+
"turtle_neck": 0,
|
53 |
+
"off_the_shoulder": 0,
|
54 |
+
"one_shoulder": 0,
|
55 |
+
"boat_neck": 0
|
56 |
+
},
|
57 |
+
"pattern": {
|
58 |
+
"floral": 0,
|
59 |
+
"stripe": 0,
|
60 |
+
"leopard_print": 0,
|
61 |
+
"plain": 100,
|
62 |
+
"geometric": 0,
|
63 |
+
"logo": 0,
|
64 |
+
"graphic_print": 0,
|
65 |
+
"other": 0
|
66 |
+
},
|
67 |
+
"fabric": {
|
68 |
+
"cotton": 0,
|
69 |
+
"denim": 0,
|
70 |
+
"linen": 0,
|
71 |
+
"satin": 0,
|
72 |
+
"silk": 0,
|
73 |
+
"sequin": 0,
|
74 |
+
"leather": 0,
|
75 |
+
"velvet": 100,
|
76 |
+
"knit": 0,
|
77 |
+
"lace": 0,
|
78 |
+
"suede": 0,
|
79 |
+
"sheer": 0,
|
80 |
+
"polyester": 0,
|
81 |
+
"viscose": 0
|
82 |
+
},
|
83 |
+
"features": {
|
84 |
+
"pockets": 0,
|
85 |
+
"lined": 0,
|
86 |
+
"cut_out": 0,
|
87 |
+
"backless": 0,
|
88 |
+
"none": 100
|
89 |
+
},
|
90 |
+
"closure": {
|
91 |
+
"button": 0,
|
92 |
+
"zip": 0,
|
93 |
+
"press_stud": 0,
|
94 |
+
"clasp": 0
|
95 |
+
},
|
96 |
+
"body_fit": {
|
97 |
+
"petite": 0,
|
98 |
+
"maternity": 0,
|
99 |
+
"regular": 100,
|
100 |
+
"tall": 0,
|
101 |
+
"plus_size": 0
|
102 |
+
},
|
103 |
+
"occasion": {
|
104 |
+
"beach": 0,
|
105 |
+
"casual": 0,
|
106 |
+
"cocktail": 0,
|
107 |
+
"day": 0,
|
108 |
+
"evening": 100,
|
109 |
+
"mother_of_the_bride": 0,
|
110 |
+
"party": 0,
|
111 |
+
"prom": 0,
|
112 |
+
"wedding_guest": 0,
|
113 |
+
"work": 0,
|
114 |
+
"sportswear": 0
|
115 |
+
},
|
116 |
+
"season": {
|
117 |
+
"spring": 0,
|
118 |
+
"summer": 0,
|
119 |
+
"autumn": 0,
|
120 |
+
"winter": 100
|
121 |
+
}
|
122 |
+
}
|
123 |
+
|
124 |
class BaseAttributionService(ABC):
|
125 |
@abstractmethod
|
126 |
async def extract_attributes(
|
|
|
133 |
) -> Dict[str, Any]:
|
134 |
pass
|
135 |
|
136 |
+
@abstractmethod
|
137 |
+
async def reevaluate_atributes(
|
138 |
+
self,
|
139 |
+
attributes_model: Type[BaseModel],
|
140 |
+
ai_model: str,
|
141 |
+
img_urls: List[str],
|
142 |
+
product_taxonomy: str,
|
143 |
+
pil_images: List[Any] = None,
|
144 |
+
) -> Dict[str, Any]:
|
145 |
+
pass
|
146 |
+
|
147 |
@abstractmethod
|
148 |
async def follow_schema(
|
149 |
self, schema: Dict[str, Any], data: Dict[str, Any]
|
|
|
173 |
# pil_images=pil_images, # temporarily removed to save cost
|
174 |
img_paths=img_paths,
|
175 |
)
|
176 |
+
# data = example_data
|
177 |
validate_json_data(data, schema)
|
178 |
|
179 |
+
str_data = str(data)
|
180 |
+
reevaluate_data = await self.reevaluate_atributes(
|
181 |
+
attributes_model,
|
182 |
+
ai_model,
|
183 |
+
img_urls,
|
184 |
+
product_taxonomy if product_taxonomy != "" else "main",
|
185 |
+
str_data,
|
186 |
+
# pil_images=pil_images, # temporarily removed to save cost
|
187 |
+
img_paths=img_paths,
|
188 |
+
)
|
189 |
+
|
190 |
+
init_reevaluate_data = {}
|
191 |
+
for field_name, field in attributes_model.model_fields.items(): # type: ignore
|
192 |
+
print(f"{field_name}: {field.description}")
|
193 |
+
if "single value" in field.description.lower():
|
194 |
+
max_percentage = 0
|
195 |
+
for k, v in reevaluate_data[field_name].items():
|
196 |
+
if v > max_percentage:
|
197 |
+
max_percentage = v
|
198 |
+
init_reevaluate_data[field_name] = k
|
199 |
+
elif "multiple values" in field.description.lower():
|
200 |
+
init_list = []
|
201 |
+
for k, v in reevaluate_data[field_name].items():
|
202 |
+
if v >= 60:
|
203 |
+
init_list.append(k)
|
204 |
+
init_reevaluate_data[field_name] = init_list
|
205 |
+
|
206 |
+
return data, init_reevaluate_data
|
207 |
|
208 |
async def follow_schema_with_validation(
|
209 |
self, schema: Dict[str, Any], data: Dict[str, Any]
|
app/services/service_openai.py
CHANGED
@@ -147,6 +147,95 @@ class OpenAIService(BaseAttributionService):
|
|
147 |
raise VendorError(errors.VENDOR_ERROR_INVALID_JSON)
|
148 |
|
149 |
return parsed_data
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
150 |
|
151 |
@weave.op
|
152 |
async def follow_schema(
|
|
|
147 |
raise VendorError(errors.VENDOR_ERROR_INVALID_JSON)
|
148 |
|
149 |
return parsed_data
|
150 |
+
|
151 |
+
async def reevaluate_atributes(
|
152 |
+
self,
|
153 |
+
attributes_model: Type[BaseModel],
|
154 |
+
ai_model: str,
|
155 |
+
img_urls: List[str],
|
156 |
+
product_taxonomy: str,
|
157 |
+
product_data: str,
|
158 |
+
pil_images: List[Any] = None, # do not remove, this is for weave
|
159 |
+
img_paths: List[str] = None,
|
160 |
+
) -> Dict[str, Any]:
|
161 |
+
|
162 |
+
print("Prompt: ")
|
163 |
+
print(prompts.REEVALUATE_HUMAN_MESSAGE.format(product_taxonomy=product_taxonomy, product_data=product_data))
|
164 |
+
|
165 |
+
text_content = [
|
166 |
+
{
|
167 |
+
"type": "text",
|
168 |
+
"text": prompts.REEVALUATE_HUMAN_MESSAGE.format(
|
169 |
+
product_taxonomy=product_taxonomy,
|
170 |
+
product_data=product_data,
|
171 |
+
),
|
172 |
+
},
|
173 |
+
]
|
174 |
+
if img_urls is not None:
|
175 |
+
base64_data_list = []
|
176 |
+
data_format_list = []
|
177 |
+
|
178 |
+
for img_url in img_urls:
|
179 |
+
base64_data, data_format = get_image_base64_and_type(img_url)
|
180 |
+
base64_data_list.append(base64_data)
|
181 |
+
data_format_list.append(data_format)
|
182 |
+
|
183 |
+
image_content = [
|
184 |
+
{
|
185 |
+
"type": "image_url",
|
186 |
+
"image_url": {
|
187 |
+
"url": f"data:image/{data_format};base64,{base64_data}",
|
188 |
+
},
|
189 |
+
}
|
190 |
+
for base64_data, data_format in zip(base64_data_list, data_format_list)
|
191 |
+
]
|
192 |
+
elif img_paths is not None:
|
193 |
+
image_content = [
|
194 |
+
{
|
195 |
+
"type": "image_url",
|
196 |
+
"image_url": {
|
197 |
+
"url": f"data:image/{get_data_format(img_path)};base64,{get_image_data(img_path)}",
|
198 |
+
},
|
199 |
+
}
|
200 |
+
for img_path in img_paths
|
201 |
+
]
|
202 |
+
|
203 |
+
try:
|
204 |
+
logger.info("Extracting info via OpenAI...")
|
205 |
+
response = await self.client.beta.chat.completions.parse(
|
206 |
+
model=ai_model,
|
207 |
+
messages=[
|
208 |
+
{
|
209 |
+
"role": "system",
|
210 |
+
"content": prompts.REEVALUATE_SYSTEM_MESSAGE,
|
211 |
+
},
|
212 |
+
{
|
213 |
+
"role": "user",
|
214 |
+
"content": text_content + image_content,
|
215 |
+
},
|
216 |
+
],
|
217 |
+
max_tokens=1000,
|
218 |
+
response_format=attributes_model,
|
219 |
+
logprobs=False,
|
220 |
+
# top_logprobs=2,
|
221 |
+
# temperature=0.0,
|
222 |
+
top_p=1e-45,
|
223 |
+
)
|
224 |
+
except openai.BadRequestError as e:
|
225 |
+
error_message = exception_to_str(e)
|
226 |
+
raise BadRequestError(error_message)
|
227 |
+
except Exception as e:
|
228 |
+
raise VendorError(
|
229 |
+
errors.VENDOR_THROW_ERROR.format(error_message=exception_to_str(e))
|
230 |
+
)
|
231 |
+
|
232 |
+
try:
|
233 |
+
content = response.choices[0].message.content
|
234 |
+
parsed_data = json.loads(content)
|
235 |
+
except:
|
236 |
+
raise VendorError(errors.VENDOR_ERROR_INVALID_JSON)
|
237 |
+
|
238 |
+
return parsed_data
|
239 |
|
240 |
@weave.op
|
241 |
async def follow_schema(
|