File size: 6,405 Bytes
bd023d4
 
 
 
 
 
 
 
 
 
 
 
 
015111a
bd023d4
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
015111a
 
 
 
 
 
bd023d4
 
 
015111a
bd023d4
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
015111a
bd023d4
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
"""
Inference script for testing the fine-tuned telecom intent model.
Loads LoRA adapters and generates network configurations from natural language intents.

Usage on Kaggle:
    python inference.py --intent "Deploy a low latency slice for autonomous drones in the harbor zone"

Or run with a file of intents:
    python inference.py --input_file intents.txt --output_file configs.json
"""

import argparse
import json
import os
import re
import sys

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel

# ============================================================================
# CONFIGURATION
# ============================================================================

BASE_MODEL = "Qwen/Qwen2.5-7B-Instruct"  # must match train.py
ADAPTER_PATH = "./qwen2.5-7b-telecom-intent-lora"  # output from train.py
MAX_NEW_TOKENS = 1024
TEMPERATURE = 0.1  # low for deterministic config generation
TOP_P = 0.95


def load_model(adapter_path: str, base_model: str):
    """Load base model + LoRA adapters."""
    adapter_path = os.path.abspath(adapter_path)
    if not os.path.isdir(adapter_path):
        print(f"ERROR: Adapter path not found: {adapter_path}")
        print("Run train.py first to generate adapters.")
        sys.exit(1)

    print(f"Loading base model: {base_model}")
    model = AutoModelForCausalLM.from_pretrained(
        base_model,
        dtype=torch.float16,
        device_map="auto",
        trust_remote_code=True,
    )

    print(f"Loading LoRA adapters: {adapter_path}")
    model = PeftModel.from_pretrained(model, adapter_path)
    model.eval()

    tokenizer = AutoTokenizer.from_pretrained(
        base_model,
        trust_remote_code=True,
        padding_side="left",
    )
    if tokenizer.pad_token is None:
        tokenizer.pad_token = tokenizer.eos_token

    print("Model ready!")
    return model, tokenizer


def generate_config(model, tokenizer, intent_text: str) -> str:
    """Generate a network configuration from a natural language intent."""
    messages = [
        {
            "role": "system",
            "content": (
                "You are a 5G/6G network orchestrator. "
                "Given a natural language network intent, output a valid, "
                "spec-compliant JSON network configuration. "
                "Do not include any explanation — only the JSON configuration."
            ),
        },
        {"role": "user", "content": intent_text},
    ]

    prompt = tokenizer.apply_chat_template(
        messages,
        tokenize=False,
        add_generation_prompt=True,
    )

    inputs = tokenizer(prompt, return_tensors="pt").to(model.device)

    with torch.no_grad():
        outputs = model.generate(
            **inputs,
            max_new_tokens=MAX_NEW_TOKENS,
            temperature=TEMPERATURE,
            top_p=TOP_P,
            do_sample=True,
            pad_token_id=tokenizer.pad_token_id,
            eos_token_id=tokenizer.eos_token_id,
        )

    generated = tokenizer.decode(outputs[0], skip_special_tokens=True)

    # Extract only the assistant's response (after the prompt)
    response = generated[len(prompt):].strip()

    # Try to extract JSON if wrapped in markdown
    json_match = re.search(r"```(?:json)?\s*(.*?)\s*```", response, re.DOTALL)
    if json_match:
        response = json_match.group(1)

    return response


def validate_json(text: str) -> tuple[bool, dict | None]:
    """Try to parse response as JSON. Returns (success, parsed)."""
    try:
        text = text.strip()
        start = text.find("{")
        end = text.rfind("}")
        if start != -1 and end != -1 and end > start:
            text = text[start:end + 1]
        parsed = json.loads(text)
        return True, parsed
    except json.JSONDecodeError:
        return False, None


def main():
    parser = argparse.ArgumentParser(description="Telecom Intent Inference")
    parser.add_argument(
        "--intent",
        type=str,
        default=None,
        help="Single natural language intent string",
    )
    parser.add_argument(
        "--input_file",
        type=str,
        default=None,
        help="File with one intent per line",
    )
    parser.add_argument(
        "--output_file",
        type=str,
        default="generated_configs.json",
        help="Output JSON file for batch results",
    )
    parser.add_argument(
        "--adapter_path",
        type=str,
        default=ADAPTER_PATH,
        help="Path to LoRA adapters",
    )
    parser.add_argument(
        "--base_model",
        type=str,
        default=BASE_MODEL,
        help="Base model name",
    )
    args = parser.parse_args()

    model, tokenizer = load_model(args.adapter_path, args.base_model)

    intents = []
    if args.intent:
        intents = [args.intent]
    elif args.input_file:
        with open(args.input_file, "r") as f:
            intents = [line.strip() for line in f if line.strip()]
    else:
        # Interactive mode
        print("\nInteractive mode. Type 'quit' to exit.")
        while True:
            user_input = input("\nIntent> ")
            if user_input.lower() in ("quit", "exit", "q"):
                break
            config = generate_config(model, tokenizer, user_input)
            is_valid, parsed = validate_json(config)
            print(f"\n{'=' * 60}")
            print(f"Generated Config (valid={is_valid}):")
            print(f"{'=' * 60}")
            if is_valid:
                print(json.dumps(parsed, indent=2))
            else:
                print(config)
        return

    # Batch processing
    results = []
    valid_count = 0
    for i, intent in enumerate(intents):
        print(f"\n[{i + 1}/{len(intents)}] Processing: {intent[:80]}...")
        config = generate_config(model, tokenizer, intent)
        is_valid, parsed = validate_json(config)
        if is_valid:
            valid_count += 1

        results.append({
            "intent": intent,
            "generated_config": parsed if is_valid else config,
            "json_valid": is_valid,
        })

    # Save results
    with open(args.output_file, "w") as f:
        json.dump(results, f, indent=2)

    print(f"\n{'=' * 60}")
    print(f"Batch complete: {valid_count}/{len(intents)} valid JSON configs")
    print(f"Results saved to: {args.output_file}")


if __name__ == "__main__":
    main()