Pith Number

pith:VXY2QJRM

pith:2026:VXY2QJRMYB5BCEU2CDI2UN2VCC

not attested not anchored not stored refs resolved

Diagnosing Training Inference Mismatch in LLM Reinforcement Learning

Geoffrey Fox, Neiwen Ling, Peng Wu, Tianle Zhong, Tianshu Yu, Xiao Yu, Yifan Pi, Zijun Wei

Small token-level numerical disagreements can independently cause training collapse in LLM reinforcement learning.

arxiv:2605.14220 v1 · 2026-05-14 · cs.LG · cs.AI · cs.CL

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

Add to your LaTeX paper

\usepackage{pith}
\pithnumber{VXY2QJRMYB5BCEU2CDI2UN2VCC}

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

small token-level numerical disagreements can independently cause training collapse

C2weakest assumption

The VeXact diagnostic setting successfully isolates TIM from off-policy drift and stabilization mechanisms without introducing new artifacts.

C3one line summary

Training-inference mismatch in separated rollout and optimization stages of LLM RL can independently cause training collapse.

References

53 extracted · 53 resolved · 6 Pith anchors

[1] Every Step Evolves: Scaling Reinforcement Learning for Trillion-Scale Thinking Model , author=. 2025 , eprint= 2025

[2] Llms can learn to reason via off-policy rl

[3] Thinking Machines Lab: Connectionism , year =

[4] DeepSeek-V3 Technical Report , author=. 2025 , eprint= 2025

[5] DeepSeek-V2: A Strong, Economical, and Efficient Mixture-of-Experts Language Model , author=. 2024 , eprint= 2024

Receipt and verification

First computed	2026-05-17T23:39:10.831276Z
Builder	pith-number-builder-2026-05-17-v1
Signature	Pith Ed25519 (`pith-v1-2026-05`) · public key
Schema	pith-number/v1.0

Canonical hash

adf1a8262cc07a11129a10d1aa3755109abcb75eeeff24231bd3373b15b6f058

Aliases

arxiv: 2605.14220 · arxiv_version: 2605.14220v1 · doi: 10.48550/arxiv.2605.14220 · pith_short_12: VXY2QJRMYB5B · pith_short_16: VXY2QJRMYB5BCEU2 · pith_short_8: VXY2QJRM

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/VXY2QJRMYB5BCEU2CDI2UN2VCC \
  | 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: adf1a8262cc07a11129a10d1aa3755109abcb75eeeff24231bd3373b15b6f058

Canonical record JSON

{
  "metadata": {
    "abstract_canon_sha256": "7194c730f30bb55b8b96467914e6d0ecc1a85d0c86623a37e0c1509ba2261ac8",
    "cross_cats_sorted": [
      "cs.AI",
      "cs.CL"
    ],
    "license": "http://arxiv.org/licenses/nonexclusive-distrib/1.0/",
    "primary_cat": "cs.LG",
    "submitted_at": "2026-05-14T00:27:35Z",
    "title_canon_sha256": "b3906a1c2350895f3131cb1dd66a543835acac8b02f98ba394dcc8c2edf47924"
  },
  "schema_version": "1.0",
  "source": {
    "id": "2605.14220",
    "kind": "arxiv",
    "version": 1
  }
}