MinaGabriel/phi2-2.7b-lora-nars-adapter

Phi-2 LoRA Adapter — NARS Reasoning (A/B/C Inference)

This repository contains a LoRA adapter fine-tuned on microsoft/phi-2 to perform structured reasoning in the style of Non-Axiomatic Reasoning (NARS).
The model learns to read a set of premises and a claim, then decide whether the claim is True (A), False (B), or Uncertain (C).

---

Model Summary

Field	Value
Base model	microsoft/phi-2
Adapter type	LoRA (PEFT)
Quantization	4-bit (NF4), bfloat16 compute
Target modules	q_proj, k_proj, v_proj, o_proj
LoRA config	r=16, alpha=32, dropout=0.05
Dataset	MinaGabriel/NARS-Reasoning-v0.1
Task	3-way reasoning classification (A/B/C)
Prompt masking	Prompts masked, answer-only supervision

---

Usage

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

BASE = "microsoft/phi-2"
ADAPTER = "MinaGabriel/phi2-2.7b-lora-nars-adapter"
device = "cuda" if torch.cuda.is_available() else "cpu"

tok = AutoTokenizer.from_pretrained(BASE, use_fast=True)
if tok.pad_token is None:
    tok.pad_token = tok.eos_token

base = AutoModelForCausalLM.from_pretrained(BASE, device_map="auto", torch_dtype=torch.float16)
model = PeftModel.from_pretrained(base, ADAPTER).eval().to(device)

PROMPT = """Premises:
{context}

Claim:
{question}

Choose one option and output its LETTER only.
A) True
B) False
C) Uncertain

Answer:"""

def build_prompt(example):
    return PROMPT.format(context=example["context"], question=example["question"])
 
prompt = build_prompt({
    "context": "All mammals are warm-blooded. All whales are mammals. Moby is a whale.",
    "question": "Moby is warm-blooded."
})
print("Prompt:\n", prompt)
inputs = tok(prompt, return_tensors="pt").to(device)
out = model.generate(**inputs, max_new_tokens=1)
print("Model Output:\n", tok.decode(out[0], skip_special_tokens=True)) #Answer: A