Pothong/cognitive-qg-llama2-lora

Cognitive QG - LLaMA 2 7B Chat QLoRA

QLoRA fine-tuned LLaMA 2 7B Chat for structured cognitive analysis and Socratic question generation based on Facione's Critical Thinking Framework.

Model Details

• —Base Model: meta-llama/Llama-2-7b-chat-hf
• —Method: QLoRA (4-bit NF4 quantization + LoRA adapters)
• —LoRA Config: r=16, alpha=32, dropout=0.1, target=all linear layers
• —Training: 30 annotated arguments, 15 epochs, lr=1e-4, cosine schedule
• —Framework: Facione's Critical Thinking (Interpretation, Analysis, Inference, Evaluation, Explanation, Self-Regulation)

Task

Given an argument text, the model produces a structured analysis through 6 cognitive phases and generates up to 3 Socratic questions that probe reasoning weaknesses.

Usage

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

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

base_model = AutoModelForCausalLM.from_pretrained(
    "meta-llama/Llama-2-7b-chat-hf",
    quantization_config=bnb_config,
    device_map="auto",
)

model = PeftModel.from_pretrained(base_model, "Pothong/cognitive-qg-llama2-lora")
model.eval()

tokenizer = AutoTokenizer.from_pretrained("meta-llama/Llama-2-7b-chat-hf")

Output Format

The model outputs structured text with sections:

1. 1.INTERPRETATION (stance, knowledge domain)
2. 2.ANALYSIS (core claim, premise, reasoning type)
3. 3.INFERENCE (consequences, alternatives)
4. 4.EVALUATION (strength, credibility, fallacies)
5. 5.EXPLANATION (reasoning structure, justification)
6. 6.SELF-REGULATION (bias detection, revision)
7. 7.SOCRATIC QUESTIONS (up to 3)