Pith Number

pith:DA5PMZWE

pith:2026:DA5PMZWEDSJIBAILJERAP22J73

not attested not anchored not stored refs resolved

Early Data Exposure Improves Robustness to Subsequent Fine-Tuning

Aditi Raghunathan, Gaurav R. Ghosal, Jacob Mitchell Springer, Lawrence Feng, Ziqian Zhong

Mixing some target data into pretraining improves retention of that capability after later fine-tuning on new tasks.

arxiv:2605.12705 v1 · 2026-05-12 · cs.LG

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

Add to your LaTeX paper

\usepackage{pith}
\pithnumber{DA5PMZWEDSJIBAILJERAP22J73}

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

early exposure - mixing post-training data into pretraining - consistently improves the frontier between retained upstream performance and downstream performance. In compute-matched experiments, where the target data must be allocated between pretraining and post-training, we find that the optimum lies at neither extreme.

C2weakest assumption

That the controlled three-stage pipeline and specific model sizes/tasks used here capture the dynamics that matter in larger-scale, real-world training where data distributions and objectives are more complex and less cleanly separated.

C3one line summary

Early mixing of post-training data into pretraining improves retention of acquired capabilities after subsequent fine-tuning in language models.

References

25 extracted · 25 resolved · 7 Pith anchors

[1] SmolLM2: When Smol Goes Big -- Data-Centric Training of a Small Language Model · arXiv:2502.02737

[2] Louis Bethune, David Grangier, Dan Busbridge, Eleonora Gualdoni, Marco Cuturi, and Pierre Ablin

[3] arXiv preprint arXiv:2502.06042 , year=

[4] Lora learns less and forgets less

[5] Scaling laws for predicting downstream performance in llms, 2025

Receipt and verification

First computed	2026-05-18T03:09:49.627455Z
Builder	pith-number-builder-2026-05-17-v1
Signature	Pith Ed25519 (`pith-v1-2026-05`) · public key
Schema	pith-number/v1.0

Canonical hash

183af666c41c9280810b492207eb49fefd2f347f6fbf777d118b30fac1521edf

Aliases

arxiv: 2605.12705 · arxiv_version: 2605.12705v1 · doi: 10.48550/arxiv.2605.12705 · pith_short_12: DA5PMZWEDSJI · pith_short_16: DA5PMZWEDSJIBAIL · pith_short_8: DA5PMZWE

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/DA5PMZWEDSJIBAILJERAP22J73 \
  | 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: 183af666c41c9280810b492207eb49fefd2f347f6fbf777d118b30fac1521edf

Canonical record JSON

{
  "metadata": {
    "abstract_canon_sha256": "e8c0907e9d5fb802a936c88d8ea0be29e73c356d09aaa643f42975db6acbeb5c",
    "cross_cats_sorted": [],
    "license": "http://creativecommons.org/licenses/by/4.0/",
    "primary_cat": "cs.LG",
    "submitted_at": "2026-05-12T20:08:00Z",
    "title_canon_sha256": "6fe29006452258f23bebdc91ad610ebd316c92593f87cbf3ba80d94371051193"
  },
  "schema_version": "1.0",
  "source": {
    "id": "2605.12705",
    "kind": "arxiv",
    "version": 1
  }
}