Parameter	Value
Base model	mistralai/Mistral-7B-v0.3
Dataset	Bitext customer support (26k examples)
Method	QLoRA (4-bit NF4 + LoRA rank=16)
Hardware	NVIDIA T4 16GB (Google Colab free tier)
Precision	fp16 (T4 compatible)
Max steps	500
Effective batch	16 (2 × grad_accum 8)
Learning rate	2e-4 cosine

Intents Covered

order_status · refund_request · technical_issue · billing · general_faq · account · shipping · cancellation

Usage

python

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

model_id = "AyeshaImtiaz/supportsphere-mistral-support"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id, torch_dtype=torch.float16, device_map="auto")

prompt = "<s>[INST] I want to return my order and get a refund. [/INST]"
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
output = model.generate(**inputs, max_new_tokens=150, temperature=0.3, do_sample=True)
print(tokenizer.decode(output[0], skip_special_tokens=True))