Pith Number

pith:URSNDDDQ

pith:2026:URSNDDDQZODJVNCAIOQO6TBQP7

not attested not anchored not stored refs resolved

Beyond GRPO and On-Policy Distillation: An Empirical Sparse-to-Dense Reward Principle for Language-Model Post-Training

Alborz Geramifard, Hejian Sang, Ran He, Yuanda Xu, Zhengze Zhou, Zhipeng Wang

A four-stage workflow with sparse-reward RL on a teacher followed by on-policy distillation outperforms direct GRPO on LLM math reasoning.

arxiv:2605.12483 v3 · 2026-05-12 · cs.LG · 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{URSNDDDQZODJVNCAIOQO6TBQP7}

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

On verifiable math with a Qwen3-1.7B deployment student, the workflow reaches 79.3% MATH and 25.2% AIME 2024 (avg@16), versus 75.9% and 19.8% for direct GRPO on the same student.

C2weakest assumption

The teacher model must itself be reward-shaped (condition C1) and lie within a small KL divergence of the student (condition C2) for the on-policy distillation stage to provide informative dense implicit rewards.

C3one line summary

A four-stage sparse-to-dense reward workflow for LLM post-training reaches 79.3% on MATH and 25.2% on AIME 2024 with a 1.7B student, outperforming direct GRPO by enforcing dense implicit rewards from a shaped teacher.

References

30 extracted · 30 resolved · 21 Pith anchors

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

[2] Rubric-based On-policy Distillation · arXiv:2605.07396

[3] The Llama 3 Herd of Models · arXiv:2407.21783

[4] Self-Distillation Zero: Self-Revision Turns Binary Rewards into Dense Supervision · arXiv:2604.12002

[5] Distilling the Knowledge in a Neural Network · arXiv:1503.02531

Cited by

1 paper in Pith

TRIAGE: Role-Typed Credit Assignment for Agentic Reinforcement Learning

Receipt and verification

First computed	2026-05-20T00:01:44.012985Z
Builder	pith-number-builder-2026-05-17-v1
Signature	Pith Ed25519 (`pith-v1-2026-05`) · public key
Schema	pith-number/v1.0

Canonical hash

a464d18c70cb869ab44043a0ef4c307fdf6452fff054af0d6714aceff5b7685a

Aliases

arxiv: 2605.12483 · arxiv_version: 2605.12483v3 · doi: 10.48550/arxiv.2605.12483 · pith_short_12: URSNDDDQZODJ · pith_short_16: URSNDDDQZODJVNCA · pith_short_8: URSNDDDQ

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/URSNDDDQZODJVNCAIOQO6TBQP7 \
  | 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: a464d18c70cb869ab44043a0ef4c307fdf6452fff054af0d6714aceff5b7685a

Canonical record JSON

{
  "metadata": {
    "abstract_canon_sha256": "25c163c0ee0c1fd8e36b6a83351eb068b56d1735f62dec0a11da2c45c2d631a0",
    "cross_cats_sorted": [
      "cs.AI"
    ],
    "license": "http://creativecommons.org/licenses/by/4.0/",
    "primary_cat": "cs.LG",
    "submitted_at": "2026-05-12T17:57:48Z",
    "title_canon_sha256": "49c215ec67f8dcbf3197b08cf6dc71c97317ce38431a7d9b5787242c65a18466"
  },
  "schema_version": "1.0",
  "source": {
    "id": "2605.12483",
    "kind": "arxiv",
    "version": 3
  }
}