File size: 4,699 Bytes
cc8e96c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
import json

# Query Parser Prompt

QUERY_SYSTEM_PROMPT = """

You are a query-understanding component for an image search engine.



Correct obvious spelling errors in the user's Arabic or English query,

then return ONLY one valid JSON object with this exact structure:



{

  "language": "",

  "corrected_query": "",

  "semantic_query": "",

  "object_terms": [],

  "text_terms": [],

  "attributes": [],

  "relations": [],

  "search_mode": ""

}



Rules:



1. language:

   - Use "ar" for Arabic queries.

   - Use "en" for English queries.



2. corrected_query:

   - Correct spelling and grammar.

   - Keep it in the same language as the original query.

   - Do not translate it.

   - Preserve the user's dialect when possible.



3. semantic_query:

   - Write a clear English description representing the user's meaning.

   - Do not add details that the user did not mention.



4. object_terms:

   - Important visible objects.

   - Use English singular words.



5. text_terms:

   - Exact text that the user wants to find inside images.

   - Preserve capitalization and wording.



6. attributes:

   - Colors and visual characteristics in English.



7. relations:

   - Relationships between objects in English.



8. search_mode must be one of:

   - "semantic"

   - "object"

   - "text"

   - "hybrid"



9. Use "hybrid" when the query includes objects with attributes,

   relations, or more than one search method.



10. Return JSON only.

Do not use markdown and do not include explanations.

"""


# Run Query Parser using Qwen

def run_qwen_text(user_query, vlm_model, vlm_processor):
    """

    Convert the user's natural-language query

    into a structured search query using Qwen.

    """

    messages = [
        {
            "role": "system",
            "content": [
                {
                    "type": "text",
                    "text": QUERY_SYSTEM_PROMPT
                }
            ]
        },
        {
            "role": "user",
            "content": [
                {
                    "type": "text",
                    "text": user_query
                }
            ]
        }
    ]

    inputs = vlm_processor.apply_chat_template(
        messages,
        tokenize=True,
        add_generation_prompt=True,
        return_dict=True,
        return_tensors="pt"
    )

    inputs = inputs.to(vlm_model.device)

    generated_ids = vlm_model.generate(
        **inputs,
        max_new_tokens=300,
        do_sample=False,
        pad_token_id=vlm_processor.tokenizer.eos_token_id
    )

    generated_ids = generated_ids[
        :,
        inputs["input_ids"].shape[1]:
    ]

    output_text = vlm_processor.batch_decode(
        generated_ids,
        skip_special_tokens=True,
        clean_up_tokenization_spaces=False
    )[0]

    return output_text


# Clean Qwen JSON Output

def prepare_json_text(raw_text):
    """

    Remove markdown formatting and extract

    the JSON object from Qwen's response.

    """

    cleaned_text = raw_text.strip()

    cleaned_text = cleaned_text.replace(
        "```json",
        ""
    )

    cleaned_text = cleaned_text.replace(
        "```",
        ""
    )

    cleaned_text = cleaned_text.strip()

    start_index = cleaned_text.find("{")
    end_index = cleaned_text.rfind("}")

    if start_index == -1 or end_index == -1:
        raise ValueError(
            "Qwen output does not contain a valid JSON object."
        )

    return cleaned_text[
        start_index:end_index + 1
    ]


# Parse User Query

def parse_search_query(

    user_query,

    vlm_model,

    vlm_processor

):
    """

    Convert a user's search query into structured fields.

    """

    raw_output = run_qwen_text(
        user_query,
        vlm_model,
        vlm_processor
    )

    cleaned_output = prepare_json_text(
        raw_output
    )

    try:
        parsed_query = json.loads(
            cleaned_output
        )

    except json.JSONDecodeError as error:

        print("Raw Qwen output:")
        print(raw_output)

        raise ValueError(
            f"Qwen returned invalid JSON: {error}"
        )

    required_fields = [
        "language",
        "corrected_query",
        "semantic_query",
        "object_terms",
        "text_terms",
        "attributes",
        "relations",
        "search_mode"
    ]

    for field in required_fields:

        if field not in parsed_query:

            raise ValueError(
                f"Missing query field: {field}"
            )

    return parsed_query