Views
No views yet
1from transformers import AutoModelForCausalLM, AutoTokenizer
2import torch
3
4model_id = "lablab-ai-amd-developer-hackathon/security-builder-14b"
5tokenizer = AutoTokenizer.from_pretrained(model_id)
6model = AutoModelForCausalLM.from_pretrained(model_id, torch_dtype=torch.float16, device_map="auto")
7
8### 💬 Example Usage (JSON Mode)
9messages = [
10 {"role": "user", "content": "Fix the buffer overflow and return JSON with keys: fixed_code, explanation, cwe_mitigated."}
11]
12prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
13inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
14
15with torch.no_grad():
16 output = model.generate(**inputs, max_new_tokens=512, temperature=0.1)
17
18import json
19print(json.loads(tokenizer.decode(output[0][inputs.input_ids.shape[1]:], skip_special_tokens=True)))| Parameter | Value |
|---|---|
| Base Model | Qwen2.5-Coder-14B-Instruct |
| Fine-tuning | LoRA (r=64, alpha=128, dropout=0.05) |
| Training Data | Custom secure coding & patch dataset |
| Epochs | 3 |
| Precision | float16 (ROCm-optimized) |
| Format | Safetensors (6 shards, ~28GB) |
| VRAM Required | ~38-42 GB |