This repository contains a merged fp16 model for DepEd Grade 11-12 math tutoring. The LoRA adapter has been merged into the base model weights, so inference does not require PEFT adapters.

Base model: Qwen/Qwen2.5-7B-Instruct
Source adapter: Deign86/deped-math-qwen2.5-7b-checkpoint-700-lora
Training checkpoint: step 700
Target merged repo: Deign86/deped-math-qwen2.5-7b-deped-math-merged
Output format: standalone fp16 Transformers model

Intended Use

DepEd-aligned tutoring support for Grade 11-12 mathematics
Step-by-step instructional responses
Prototyping, evaluation, and classroom assistant workflows

Limitations

Not a replacement for teacher assessment and curriculum validation
May produce incorrect or incomplete reasoning for edge cases
Requires prompt and output review in real classroom settings

Quick Usage

python

from transformers import AutoTokenizer, AutoModelForCausalLM

model_id = "Deign86/deped-math-qwen2.5-7b-deped-math-merged"

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype="auto",
    device_map="auto",
)