---
language:
- en
license: other
license_name: apache-2.0-mixed-data-terms
library_name: transformers
pipeline_tag: text-generation
base_model: mistralai/Mistral-7B-v0.1
datasets:
- allenai/ai2_arc
- allenai/openbookqa
- allenai/qasc
- allenai/sciq
tags:
- mistral
- qlora
- preference-learning
- mutual-information-optimization
- mio
- hard-negative-mining
- arc-challenge
- science-qa
- multiple-choice
- fp16
model-index:
- name: AIRE-ARC-Mistral-7B-SFT-MIO-FP16
  results:
  - task:
      type: question-answering
      name: ARC-Challenge 25-shot
    dataset:
      type: allenai/ai2_arc
      name: AI2 ARC-Challenge
      config: ARC-Challenge
      split: test
    metrics:
    - type: acc_norm
      name: Normalized Accuracy
      value: 0.765358361774744
    - type: acc
      name: Accuracy
      value: 0.7474402730375427
---

# AIRE-ARC-Mistral-7B-SFT-MIO-FP16

A standalone FP16 fine-tune of [`mistralai/Mistral-7B-v0.1`](https://huggingface.co/mistralai/Mistral-7B-v0.1) for ARC-Challenge and related science multiple-choice reasoning. The model was trained in two stages:

1. completion-only QLoRA SFT on 27,294 public science-QA training examples;
2. MIO preference fine-tuning on 47,087 train-only pairs, where the rejected answers were the incorrect choices that the SFT model found most plausible.

The final model reached **76.54% `acc_norm`** and **74.74% raw accuracy** on the 1,172-example ARC-Challenge test split with 25-shot evaluation. Model selection was completed on the validation split before the test split was opened.

> **Important:** this is a domain-specialized completion model, not a chat or general instruction model. It improves science and commonsense multiple-choice reasoning, but capability-retention audits found regressions on MMLU Humanities, MMLU Social Sciences, and WikiText perplexity.

## Model summary

| Item | Value |
|---|---|
| Architecture | `MistralForCausalLM` |
| Parameters | 7B |
| Parent model | `mistralai/Mistral-7B-v0.1` |
| Parent revision | `27d67f1b5f57dc0953326b2601d68371d40ea8da` |
| Repository format | Standalone merged FP16 weights |
| Adapter required | No |
| Primary language | English |
| Primary task | Science multiple-choice QA |
| Training hardware | One NVIDIA A100-SXM4 40GB |

The repository contains the tokenizer and merged model weights. PEFT and bitsandbytes are not required for inference. `merge_summary.json` records the merge provenance used for the submitted checkpoint.

## Intended use

Suitable uses include:

- evaluating science and commonsense multiple-choice reasoning;
- studying model-scored hard-negative selection;
- reproducing the reported ARC-Challenge evaluation;
- research on preference optimization for short-answer QA.

The model is not intended for safety-critical use, factual deployment without verification, general chat, or instruction following. No additional safety alignment was performed.

## Prompt format

Training and evaluation use answer-text completions rather than answer-position labels:

```text
Question: {question}
Answer:
```

The expected completion is the answer text, including its leading space. No chat template is used.

## Quick start

```python
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

repo_id = "psymon/mistral-7b-mio-arc-fp16"

tokenizer = AutoTokenizer.from_pretrained(repo_id)
if tokenizer.pad_token_id is None:
    tokenizer.pad_token = tokenizer.eos_token

model = AutoModelForCausalLM.from_pretrained(
    repo_id,
    dtype=torch.float16,
    device_map="auto",
)
model.eval()

prompt = "Question: What is a worldwide increase in temperature called?\nAnswer:"
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)

with torch.inference_mode():
    output = model.generate(
        **inputs,
        max_new_tokens=24,
        do_sample=False,
        pad_token_id=tokenizer.pad_token_id,
        eos_token_id=tokenizer.eos_token_id,
    )

completion = tokenizer.decode(
    output[0, inputs["input_ids"].shape[1]:],
    skip_special_tokens=True,
)
print(completion)
```

For multiple-choice evaluation, compare the conditional log-likelihood of each answer text. Do not convert the choices to `A`, `B`, `C`, and `D` unless the evaluation protocol is changed accordingly.

## Training data

Only public **training splits** were used for optimization. ARC-Challenge validation was used for deduplication, failure analysis, and model selection. ARC-Challenge test was not used during training or model selection.

### SFT corpus

| Source | Examples after filtering | Role |
|---|---:|---|
| ARC-Challenge | 1,117 | SFT and preference candidates |
| ARC-Easy | 2,241 | SFT and preference candidates |
| OpenBookQA | 4,826 | SFT and preference candidates |
| QASC | 7,514 | SFT and preference candidates |
| SciQ | 11,596 | SFT only |
| **Total** | **27,294** | |

The SFT prompt contains the question only; the target is the correct answer text. Choices were retained for data normalization and duplicate checks but were not included in the SFT prompt.

### Preference corpus

The SFT model scored every answer choice for 15,698 train-only prompts from ARC-Challenge, ARC-Easy, OpenBookQA, and QASC. Each score was the answer-token log-likelihood sum divided by the answer's character length. For each question, up to three highest-scoring incorrect choices were paired with the correct answer:

```text
prompt   = "Question: {question}\nAnswer:"
chosen   = " {correct answer text}"
rejected = " {model-selected incorrect answer text}"
```

This produced 47,087 preference pairs. Seven three-choice questions contributed two pairs each; the remaining questions contributed three pairs each. In 3,081 prompts, at least one incorrect choice received a higher selection score than the correct answer.

## Training procedure

### Stage 1: completion-only SFT

| Hyperparameter | Value |
|---|---:|
| Quantization | 4-bit NF4 QLoRA |
| Compute dtype | BF16 |
| LoRA target | all linear layers |
| LoRA rank / alpha / dropout | 16 / 32 / 0.05 |
| Learning rate | `1e-4` |
| Scheduler | cosine |
| Effective batch size | 16 |
| Maximum sequence length | 384 |
| Epochs / optimizer steps | 1 / 1,706 |

Loss was computed only on the answer completion; prompt tokens were excluded from the labels.

### Stage 2: MIO preference fine-tuning

MIO (Mutual Information Optimization) contrasts a chosen and rejected completion against a frozen reference model. This run used the merged SFT checkpoint as the reference and initialized a new LoRA policy from the same weights. Completion log-probabilities were averaged over response tokens, including EOS.

| Hyperparameter | Value |
|---|---:|
| Preference pairs | 47,087 |
| Beta | 0.5 |
| Learning rate | `7.5e-6` |
| LoRA rank / alpha / dropout | 16 / 32 / 0.0 |
| Effective batch size | 32 |
| Maximum sequence length | 256 |
| Epochs / optimizer steps | 1 / 1,472 |

The initial policy/reference log-ratios were exactly zero and the initial loss was `1.386294`, matching `ln(4)`. Reference inference was executed first under `no_grad`; its activations were released before the policy forward pass to fit training on a 40GB A100 without changing the objective.

The result should be interpreted as the effect of the **complete procedure**: SFT reference, model-scored hard negatives, mean completion log-probabilities, and MIO. This experiment did not isolate the causal contribution of each component.

## Evaluation

Evaluation used EleutherAI's `lm-evaluation-harness` protocol with 25 few-shot examples and answer-choice likelihood normalization.

### ARC-Challenge validation

| Model stage | `acc` | `acc_norm` |
|---|---:|---:|
| Mistral-7B-v0.1 | 51.51% | 56.52% |
| Completion-only SFT | 57.86% | 59.20% |
| SFT + MIO | 71.57% | **73.24%** |

From SFT to MIO, 51 normalized predictions changed from incorrect to correct and 9 changed from correct to incorrect, for a net gain of 42 correct answers on 299 validation examples.

### One-time held-out test

| Split | Documents | Few-shot | `acc` | `acc_norm` |
|---|---:|---:|---:|---:|
| ARC-Challenge test | 1,172 | 25 | 74.74% | **76.54%** |

The candidate model was locked before this evaluation. The test split was evaluated once, and no post-test training, hyperparameter tuning, or model reselection was performed.

### Post-test overlap audit

The test split was compared with the training ledger only after final evaluation:

- exact Test/SFT ID overlap: 0;
- exact question plus complete choice-set overlap: 0;
- exact normalized question-stem overlap: 9 test rows;
- character similarity of at least 0.90: 18 test rows.

Removing the 18 near-overlap rows changed test `acc_norm` from 76.5358% to **76.4298%** (-0.1060 percentage points). This is a sensitivity analysis, not proof that semantic contamination is absent; paraphrases and shared underlying facts may remain undetected.

## Capability-retention audit

After model selection and the one-time ARC test, the standalone Hub checkpoint was downloaded again and compared with the base model. These results were not used for training or selection.

| Benchmark | Metric | Base | Final | Change |
|---|---|---:|---:|---:|
| HellaSwag | `acc_norm` | 81.17% | 85.43% | +4.26 pp |
| PIQA | `acc_norm` | 82.48% | 85.75% | +3.26 pp |
| WinoGrande | `acc` | 75.37% | 80.03% | +4.66 pp |
| MMLU Humanities | `acc` | 56.43% | 53.18% | -3.25 pp |
| MMLU Social Sciences | `acc` | 73.81% | 70.46% | -3.35 pp |
| WikiText | word perplexity | 8.0848 | 8.7194 | +7.85% (worse) |

The MMLU regressions partly depend on answer-position labels: accuracy fell much more for answers at positions `B` and `D` than for `A` and `C`. This suggests that output-label calibration changed in addition to any knowledge loss. WikiText does not use answer labels and also regressed, so format calibration alone cannot explain all of the degradation.

## Reproducing the ARC evaluation

```bash
pip install "lm-eval==0.4.12"

python -m lm_eval \
  --model hf \
  --model_args "pretrained=psymon/mistral-7b-mio-arc-fp16,dtype=float16" \
  --tasks arc_challenge \
  --num_fewshot 25 \
  --batch_size 8
```

The reported submission result used the official ARC-Challenge parquet files in an offline custom task, with hashes checked before evaluation. The submitted Colab notebook contains the exact data validation, model-selection lock, and one-time test procedure.

## Limitations

- **Domain specialization:** optimized for short English science-QA completions.
- **Not a chat model:** no instruction-following or conversational alignment was added.
- **Single training seed:** training variance was not estimated across seeds.
- **Public-corpus overlap risk:** post-test string audits cannot exclude semantic overlap or paraphrases.
- **Capability trade-offs:** MMLU and WikiText regressions show that the model did not preserve all base capabilities.
- **Incomplete coverage:** code, mathematics, safety, long-form generation, summarization, and multilingual behavior were not evaluated.
- **No safety guarantee:** outputs may be incorrect, biased, or unsafe.

## License and data terms

The parent model is released under Apache-2.0. The training sources have separate terms:

| Source | Terms shown by the source dataset card |
|---|---|
| ARC | CC BY-SA 4.0 |
| QASC | CC BY 4.0 |
| SciQ | CC BY-NC 3.0 |
| OpenBookQA | No license declared in the pinned Hugging Face dataset card |

Because the SFT stage includes SciQ and the OpenBookQA card does not declare complete licensing information, this model card intentionally uses `license: other` rather than presenting the repository as unconditionally Apache-2.0. The repository is shared for research and coding-test evaluation. Users are responsible for reviewing the parent-model license and each source dataset's terms before redistribution or downstream use, especially commercial use. This section is informational and not legal advice.

## References

- A. Q. Jiang et al., [Mistral 7B](https://arxiv.org/abs/2310.06825), 2023.
- T. Dettmers et al., [QLoRA: Efficient Finetuning of Quantized LLMs](https://arxiv.org/abs/2305.14314), 2023.
- X. Lv et al., [The Hidden Link Between RLHF and Contrastive Learning](https://arxiv.org/abs/2506.22578), ICLR 2026.
- P. Clark et al., [Think you have Solved Question Answering? Try ARC, the AI2 Reasoning Challenge](https://arxiv.org/abs/1803.05457), 2018.