Modelpublic

Farhan45876/mavro-ai-lora

sourceHugging Faceapache-2.0updated 23d agoView on Hugging Face

0likes72downloads

Model Card

Mavro AI (LoRA)

Product: Mavro AI Model family: Mavro LLM (LoRA on Qwen2.5-3B-Instruct) Created by: Synereos — AI research & technology company Not the maker: Espera Mavro (global sourcing platform; separate organization)

What this is

PEFT/LoRA adapter fine-tuned for Espera Mavro shopping agent behavior:

—Catalog-first search → 1688 → Amazon/Alibaba/AliExpress sourcing rules
—Tool calling, payment safety, anti-hallucination
—Branding identity: Mavro AI by Synereos

Load

python

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

base_id = "Qwen/Qwen2.5-3B-Instruct"
tok = AutoTokenizer.from_pretrained(base_id, trust_remote_code=True)
base = AutoModelForCausalLM.from_pretrained(
    base_id, torch_dtype=torch.float16, device_map="auto", trust_remote_code=True
)
model = PeftModel.from_pretrained(base, "Farhan45876/mavro-ai-lora")

Training

—Platform: Kaggle T4×2
—Method: QLoRA 4-bit + LoRA r=16
—Data: Mavro V3 commerce + branding JSONL (~2k examples)