SoominSion

aire-llm-mistral7b-arc-final

Deploy Dedicated

Dedicated Endpoints

Run this model inference on single tenant GPU with unmatched speed and reliability at scale.

Learn more

Container

Run this model inference with full control and performance in your environment.

Learn more

Get help setting up a custom Dedicated Endpoints.

Talk with our engineer to get a quote for reserved GPU instances with discounts.

Base Model

mistralai/Mistral-7B-v0.1
Local base revision used in experiments: 27d67f1b5f57dc0953326b2601d68371d40ea8da

Evaluation

Evaluation used EleutherAI lm-evaluation-harness on arc_challenge with 25 few-shot examples.

Metric	Value	Stderr
acc	0.6681	0.0138
acc_norm	0.7056	0.0133

These numbers are from the local saved checkpoint evaluation. A separate Hub reload evaluation should be run after upload to confirm reproducibility.

Training Summary

The model was trained with QLoRA using 4-bit NF4 quantization and fp16 compute. The training pipeline combined science-domain short-answer SFT, ARC label+answer-text realignment, and light eval-matched text-only DPO ranking correction.

Loading

python
from peft import PeftModel
from transformers import AutoModelForCausalLM, AutoTokenizer

base = "mistralai/Mistral-7B-v0.1"
adapter = "SoominSion/aire-llm-mistral7b-arc-final"

tokenizer = AutoTokenizer.from_pretrained(adapter, use_fast=False)
model = AutoModelForCausalLM.from_pretrained(base, torch_dtype="auto")
model = PeftModel.from_pretrained(model, adapter)