Pith Number

pith:USNR7PI7

pith:2026:USNR7PI7ZDEQNRM6DFJQ3IZ4SX

not attested not anchored not stored refs resolved

JanusPipe: Efficient Pipeline Parallel Training for Machine Learning Interatomic Potentials

Guangming Tan, Hongtao Xu, Hongyu Wang, Mingzhen Li, Weijian Liu, Weile Jia, Yan Wang

JanusPipe enables efficient pipeline parallel training for conservative MLIPs by using SymFold and WaveK to handle their double-backward pattern.

arxiv:2605.18404 v1 · 2026-05-18 · cs.DC

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

Add to your LaTeX paper

\usepackage{pith}
\pithnumber{USNR7PI7ZDEQNRM6DFJQ3IZ4SX}

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

We present JanusPipe, an efficient 3D-parallel (PP/DP/GP) training system tailored for conservative MLIPs. It integrates SymFold to enable memory-efficient pipeline parallelism for conservative MLIPs, and WaveK to reduce pipeline bubbles by balancing the four-phase compute time. Experimental results on 32 GPUs show that JanusPipe improves throughput by 1.51× and 1.45× on average over 1F1B and Hanayo, respectively.

C2weakest assumption

The double-backward execution pattern of conservative MLIPs creates a fundamental mismatch with existing pipeline parallelism systems that SymFold and WaveK can resolve without introducing significant accuracy loss, overhead, or model-specific constraints that limit generality.

C3one line summary

JanusPipe is a new 3D-parallel training system for conservative MLIPs that uses SymFold and WaveK to achieve 1.51x and 1.45x average throughput gains over 1F1B and Hanayo on 32 GPUs.

References

94 extracted · 94 resolved · 10 Pith anchors

[1] A foundation model for atomistic materials chemistry · doi:10.48550/arxiv.2401.00096

[2] Open Materials 2024 (OMat24) Inorganic Materials Dataset and Models 2024 · doi:10.48550/arxiv.2410.12771

[3] MatterSim: A Deep Learning Atomistic Model Across Elements, Temperatures and Pressures · doi:10.48550/arxiv.2405.04967

[4] Forty-second International Conference on Machine Learning , year=

[5] Tuckerman, Mark E. , title =. 2010 , address = 2010

Formal links

2 machine-checked theorem links

Receipt and verification

First computed	2026-05-20T00:05:59.051865Z
Builder	pith-number-builder-2026-05-17-v1
Signature	Pith Ed25519 (`pith-v1-2026-05`) · public key
Schema	pith-number/v1.0

Canonical hash

a49b1fbd1fc8c906c59e19530da33c95cf5a0e31e6cdf40bfce79846c10792f1

Aliases

arxiv: 2605.18404 · arxiv_version: 2605.18404v1 · doi: 10.48550/arxiv.2605.18404 · pith_short_12: USNR7PI7ZDEQ · pith_short_16: USNR7PI7ZDEQNRM6 · pith_short_8: USNR7PI7

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/USNR7PI7ZDEQNRM6DFJQ3IZ4SX \
  | 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: a49b1fbd1fc8c906c59e19530da33c95cf5a0e31e6cdf40bfce79846c10792f1

Canonical record JSON

{
  "metadata": {
    "abstract_canon_sha256": "9fecb13af08e5670227cceed92709b3fb86475c7ed367aa5af0c2b5111645675",
    "cross_cats_sorted": [],
    "license": "http://creativecommons.org/licenses/by/4.0/",
    "primary_cat": "cs.DC",
    "submitted_at": "2026-05-18T13:45:58Z",
    "title_canon_sha256": "5804b1dc7c8ddcdc6deccf583a7fba00c7c12204df1953ab2a3e4393096b31d6"
  },
  "schema_version": "1.0",
  "source": {
    "id": "2605.18404",
    "kind": "arxiv",
    "version": 1
  }
}