Pith Number

pith:KCYBYGFN

pith:2024:KCYBYGFNDCFVCHIZJQNV3ZXWCY

not attested not anchored not stored refs resolved

ORPO: Monolithic Preference Optimization without Reference Model

James Thorne, Jiwoo Hong, Noah Lee

A simple odds-ratio penalty during supervised fine-tuning suffices to align language models without any reference model or separate alignment stage.

arxiv:2403.07691 v2 · 2024-03-12 · cs.CL · cs.AI

Open paper page JSON Open Graph Bundle Merged state Verified badge What is a Pith Number?

Add to your LaTeX paper

\usepackage{pith}
\pithnumber{KCYBYGFNDCFVCHIZJQNV3ZXWCY}

Prints a linked badge after your title and injects PDF metadata. Compiles on arXiv. Learn more · Embed verified badge

Record completeness

1 Bitcoin timestamp

2 Internet Archive

3 Author claim open · sign in to claim

4 Citations open

5 Replications open

✓ Portable graph bundle live · download bundle · merged state

The bundle contains the canonical record plus signed events. A mirror can host it anywhere and recompute the same current state with the deterministic merge algorithm.

Claims

C1strongest claim

fine-tuning Phi-2 (2.7B), Llama-2 (7B), and Mistral (7B) with ORPO on the UltraFeedback alone surpasses the performance of state-of-the-art language models with more than 7B and 13B parameters: achieving up to 12.20% on AlpacaEval 2.0, 66.19% on IFEval (instruction-level loose), and 7.32 in MT-Bench.

C2weakest assumption

That the odds ratio is a sensible choice for contrasting favored and disfavored generation styles during supervised fine-tuning, and that a minor penalty for disfavored responses is sufficient to achieve preference alignment.

C3one line summary

ORPO performs preference alignment during supervised fine-tuning via a monolithic odds ratio penalty, allowing 7B models to outperform larger state-of-the-art models on alignment benchmarks.

References

291 extracted · 291 resolved · 26 Pith anchors

[2] The Falcon Series of Open Language Models 2023 · arXiv:2311.16867

[3] arXiv preprint arXiv:2310.12036 , year= 2023

[4] Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback 2022 · arXiv:2204.05862

[5] Constitutional AI: Harmlessness from AI Feedback 2022 · arXiv:2212.08073

[6] Alvaro Bartolome, Gabriel Martin, and Daniel Vila. 2023. Notus. https://github.com/argilla-io/notus 2023

Formal links

2 machine-checked theorem links

Cited by

37 papers in Pith

UNA: A Unified Supervised Framework for Efficient LLM Alignment Across Feedback Types

Preference Goal Tuning: Post-Training as Latent Control for Frozen Policies

CrossVLA: Cross-Paradigm Post-Training and Inference Optimization for Vision-Language-Action Models

Towards Context-Invariant Safety Alignment for Large Language Models

Anomaly-Preference Image Generation

Receipt and verification

First computed	2026-05-17T23:38:48.340253Z
Builder	pith-number-builder-2026-05-17-v1
Signature	Pith Ed25519 (`pith-v1-2026-05`) · public key
Schema	pith-number/v1.0

Canonical hash

50b01c18ad188b511d194c1b5de6f6162b717b773d12372a2d1c31efb8ca5f37

Aliases

arxiv: 2403.07691 · arxiv_version: 2403.07691v2 · doi: 10.48550/arxiv.2403.07691 · pith_short_12: KCYBYGFNDCFV · pith_short_16: KCYBYGFNDCFVCHIZ · pith_short_8: KCYBYGFN

Agent API

Resolver JSON Graph JSON Events JSON Schema Signing key

Verify this Pith Number yourself

curl -sH 'Accept: application/ld+json' https://pith.science/pith/KCYBYGFNDCFVCHIZJQNV3ZXWCY \
  | jq -c '.canonical_record' \
  | python3 -c "import sys,json,hashlib; b=json.dumps(json.loads(sys.stdin.read()), sort_keys=True, separators=(',',':'), ensure_ascii=False).encode(); print(hashlib.sha256(b).hexdigest())"
# expect: 50b01c18ad188b511d194c1b5de6f6162b717b773d12372a2d1c31efb8ca5f37

Canonical record JSON

{
  "metadata": {
    "abstract_canon_sha256": "5b402e995dbe6b11df7d0c941b97055f5f558cd0f2637be6ded1421aced958cc",
    "cross_cats_sorted": [
      "cs.AI"
    ],
    "license": "http://arxiv.org/licenses/nonexclusive-distrib/1.0/",
    "primary_cat": "cs.CL",
    "submitted_at": "2024-03-12T14:34:08Z",
    "title_canon_sha256": "e21d7c1c1d1611ef27dee51ddccca28d890122978ef35588b08dc0edca8a9566"
  },
  "schema_version": "1.0",
  "source": {
    "id": "2403.07691",
    "kind": "arxiv",
    "version": 2
  }
}