โก EvoNet-3B-Reasoning (Edge AI Edition)
EvoNet-3B-Reasoning lร phiรชn bแบฃn rรบt gแปn (Lightweight) thuแปc hแป sinh thรกi EvoNet Security Audit System. ฤฦฐแปฃc xรขy dแปฑng trรชn nแปn tแบฃng Llama-3.2-3B-Instruct mแปi nhแบฅt cแปงa Meta vร tรญch hแปฃp tฦฐ duy suy luแบญn logic, mรด hรฌnh nร y ฤฦฐแปฃc thiแบฟt kแบฟ ฤแบทc biแปt cho cรกc thiแบฟt bแป cรณ phแบงn cแปฉng giแปi hแบกn (Edge Devices, Mobile, Low-VRAM GPUs).
Dรน chแป cรณ kรญch thฦฐแปc 3 tแปท tham sแป, mรด hรฌnh vแบซn giแปฏ ฤฦฐแปฃc khแบฃ nฤng phรขn tรญch bแบฃo mแบญt tแปซng bฦฐแปc (Step-by-step reasoning) vร giao tiแบฟp tiแบฟng Viแปt trรดi chแบฃy.
๐ ฤiแปm nแปi bแบญt
- Siรชu nhแบน & Siรชu tแปc: Vแปi kรญch thฦฐแปc cแปฑc nhแป, cรณ thแป chแบกy mฦฐแปฃt mร trรชn cรกc mรกy tรญnh cรก nhรขn khรดng cรณ GPU chuyรชn dแปฅng, hoแบทc tรญch hแปฃp trแปฑc tiแบฟp vร o แปฉng dแปฅng di ฤแปng.
- Tฦฐ duy phรขn tรญch (Reasoning): Tรญch hแปฃp LoRA Reasoning giรบp mรด hรฌnh suy nghฤฉ trฦฐแปc khi trแบฃ lแปi, cแบฃi thiแปn ฤแป chรญnh xรกc trong cรกc bร i toรกn nhแบญn diแปn mรฃ ฤแปc.
- Hแป trแปฃ Tiแบฟng Viแปt: Khแบฃ nฤng sinh ngแปฏ cแบฃnh vร giแบฃi thรญch lแปi bแบฃo mแบญt bแบฑng tiแบฟng Viแปt tแปฑ nhiรชn.
โ ๏ธ Giแปi hแบกn (Limitation Notice)
Do giแปi hแบกn vแบญt lรฝ vแป sแป lฦฐแปฃng tham sแป (3B so vแปi 8B), mรด hรฌnh cรณ thแป gแบทp hiแปn tฦฐแปฃng Hallucination (แบฃo giรกc) trong cรกc kแปch bแบฃn phรขn tรญch lแป hแปng quรก sรขu hoแบทc giแบฃi thรญch sai cฦก chแบฟ hoแบกt ฤแปng cแปงa mแปt sแป payload phแปฉc tแบกp. ๐ Khuyแบฟn nghแป: Sแปญ dแปฅng phiรชn bแบฃn EvoNet-8B-Reasoning cho cรกc mรดi trฦฐแปng Production ฤรฒi hแปi ฤแป chรญnh xรกc tuyแปt ฤแปi.
๐ ๏ธ Chi tiแบฟt kแปน thuแบญt
- Kiแบฟn trรบc: Llama 3.2
- Kรญch thฦฐแปc: 3 Billion Parameters
- Base Model:
unsloth/Llama-3.2-3B-Instruct - Ngฦฐแปi phรกt triแปn: Phong Huแปณnh (EvoNet)
๐ป Hฦฐแปng dแบซn sแปญ dแปฅng (Python)
Mรด hรฌnh cรณ thแป chแบกy dแป
dร ng trรชn bแบฅt kแปณ GPU nร o (thแบญm chรญ lร CPU) vแปi thฦฐ viแปn transformers.
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch
model_id = "EvoNet/EvoNet-3B-Reasoning"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
model_id,
torch_dtype=torch.float16,
device_map="auto" # Chแบกy siรชu mฦฐแปฃt trรชn cรกc GPU nhแป
)
messages = [
{"role": "system", "content": "Bแบกn lร EvoNet Pentester. Hรฃy suy nghฤฉ tแปซng bฦฐแปc trฦฐแปc khi trแบฃ lแปi."},
{"role": "user", "content": "Phรขn tรญch payload sau: `admin' AND (SELECT 1 FROM (SELECT SLEEP(5))A) AND '1'='1`. ฤรขy lร lแปi gรฌ?"}
]
prompt_text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = tokenizer(prompt_text, return_tensors="pt").to(model.device)
outputs = model.generate(**inputs, max_new_tokens=1024, temperature=0.7)
print(tokenizer.decode(outputs[0][inputs['input_ids'].shape[-1]:], skip_special_tokens=True))
- Downloads last month
- 9