Moodlerz/mistral-planner-aaqa

Mistral-7B — KG-Grounded Decision Planner

LoRA adapter fine-tuned on Mistral-7B-Instruct-v0.3 for knowledge-graph-grounded decision planning.

Task

Given a user query and KG triples retrieved from a knowledge graph, the model decides:

• ANSWER — graph has sufficient evidence
• ASK — graph is partial, clarification needed
• ABSTAIN — topic absent from graph entirely

Training

• 37,264 samples across QuAC, SHaRC, HotpotQA, ContractNLI
• Single-turn and multi-turn conversational settings
• KG triples as structured context

Usage

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

MODEL_ID   = "mistralai/Mistral-7B-Instruct-v0.3"
ADAPTER_ID = "Moodlerz/mistral-planner-aaqa"

tokenizer = AutoTokenizer.from_pretrained(ADAPTER_ID)
base      = AutoModelForCausalLM.from_pretrained(
    MODEL_ID, torch_dtype=torch.bfloat16, device_map="auto"
)
model = PeftModel.from_pretrained(base, ADAPTER_ID)
model.eval()