flaviodell/pet-expert-mistral7b-lora

Pet Expert — Mistral 7B + LoRA

Fine-tuned Mistral 7B Instruct on a synthetic veterinary dataset covering 37 cat and dog breeds from the Oxford-IIIT Pet dataset.

Model details

• Base model: mistralai/Mistral-7B-Instruct-v0.3
• Method: LoRA (r=16, alpha=32) with 4-bit NF4 quantization
• Dataset: Synthetic — 185 instruction/response pairs
• Domain: Veterinary — breed temperament, health, training advice
• Framework: HuggingFace PEFT + bitsandbytes

Results (RAGAS evaluation)

Metric	Value
Faithfulness	0.491
Answer relevancy	0.9364
Evaluated on	37 questions

Related model

This LLM adapter is part of a two-model project. The companion CV classifier is available at: flaviodell/oxford-pets-resnet50

Usage

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

bnb_config = BitsAndBytesConfig(
    load_in_4bit=True,
    bnb_4bit_quant_type="nf4",
    bnb_4bit_compute_dtype=torch.float16
)

tokenizer = AutoTokenizer.from_pretrained("flaviodell/pet-expert-mistral7b-lora")
base_model = AutoModelForCausalLM.from_pretrained(
    "mistralai/Mistral-7B-Instruct-v0.3",
    quantization_config=bnb_config,
    device_map="auto"
)
model = PeftModel.from_pretrained(base_model, "flaviodell/pet-expert-mistral7b-lora")
model.eval()

prompt = (
    "<|system|>\nYou are an expert veterinarian.\n"
    "<|user|>\nWhat are the health concerns for a Persian cat?\n"
    "<|assistant|>\n"
)
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
with torch.no_grad():
    outputs = model.generate(**inputs, max_new_tokens=300)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))