Paper Citation Record · LEDGER

Beyond Zero Initialization: Investigating the Impact of Non-Zero Initialization on LoRA Fine-Tuning Dynamics

As of 8 August 2026, this Paper Citation Record lists 14 of 14 outbound references and 2 inbound Pith citation observations for arXiv:2505.23194.

A citation records a reference. It does not transfer a finding from one paper to another.

pith.paper-citation-record.v1
2505.23194 v2

Coverage vector

measured 14 of 14 reference resolution

Typed states for the displayed outbound observations.

Source: paper_references, paper_reference_links, observed 2026-08-07T13:04:51.397456Z

measured 16 of 16 standing notices

One-hop event checks from named stored sources.

Source: scholarly_work_events, retraction_status_cache, observed 2026-08-08T06:32:00.761636+00:00

measured 2 of 2 inbound itemization

Pith citing papers itemized under the disclosed page cap.

Source: paper_references, paper_reference_links, observed 2026-08-01T17:32:34.475671Z

measured 0 of 1 external citation measurements

A source-named dated measurement, never combined with another source.

Source: arxiv_reference, observed 2026-06-28T23:02:45.979783Z

Reference resolution

14 of 14 outbound references displayed

verified exact0
verified fuzzy11
unresolved3
parse uncertain0
malformed identifier0
metadata mismatch0

External citation measurements

No source-named external measurement is stored.

Outbound references

Observation 0e6c22e2-c542-4b3e-b8b9-eb9ecb912e2c · outbound

This paper cites Let Ut = ft(x) − y.

Beyond Zero Initialization: Investigating the Impact of Non-Zero Initialization on LoRA Fine-Tuning Dynamics Let Ut = ft(x) − y

Reference 1

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T13:04:52.327336Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-08T06:32:00.761636+00:00.

Observation 746fe005-bd47-4808-973b-86a62e1d367c · outbound

This paper cites In general, µP specifies the parameter weights to be randomly initialized at a scale of Θ(n−1/2), with weight updates at a scale of Θ(n−1).

Beyond Zero Initialization: Investigating the Impact of Non-Zero Initialization on LoRA Fine-Tuning Dynamics In general, µP specifies the parameter weights to be randomly initialized at a scale of Θ(n−1/2), with weight updates at a scale of Θ(n−1)

Reference 2

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T13:04:52.730487Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-08T06:32:00.761636+00:00.

Observation 4a5746d7-93e9-4521-a468-ba60d607330b · outbound

This paper cites LoRA-One: One-Step Full Gradient Could Suffice for Fine-Tuning Large Language Models, Provably and Efficiently.

Beyond Zero Initialization: Investigating the Impact of Non-Zero Initialization on LoRA Fine-Tuning Dynamics LoRA-One: One-Step Full Gradient Could Suffice for Fine-Tuning Large Language Models, Provably and Efficiently

Reference 6

Resolution: unresolved; no resolver link, observed 2026-08-07T13:04:50.710627Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation da733f07-1735-4891-97c2-becbda4983a0 · outbound

This paper cites (18) Thus, we conclude that: ( γ[ηA] + γ[ηB] = −1, γ[A0] ≤ γ[ηA], γ [B0] ≤ γ[ηB].

Beyond Zero Initialization: Investigating the Impact of Non-Zero Initialization on LoRA Fine-Tuning Dynamics (18) Thus, we conclude that: ( γ[ηA] + γ[ηB] = −1, γ[A0] ≤ γ[ηA], γ [B0] ≤ γ[ηB]

Reference 9

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T13:04:52.495919Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-08T06:32:00.761636+00:00.

Observation 19afe712-addb-4d06-804d-89fd7bdea9ac · outbound

This paper cites The hyperparameter λ represents the value of ηB/ηA (i.e., ηA = η, ηB = λη).

Beyond Zero Initialization: Investigating the Impact of Non-Zero Initialization on LoRA Fine-Tuning Dynamics The hyperparameter λ represents the value of ηB/ηA (i.e., ηA = η, ηB = λη)

Reference 11

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T13:04:51.635176Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-08T06:32:00.761636+00:00.

Observation 0d499bb7-cf9c-4ee3-a3c8-9e9d26c349d4 · outbound

This paper cites β is tuned among {1, 2, 4, 8, 16}.

Beyond Zero Initialization: Investigating the Impact of Non-Zero Initialization on LoRA Fine-Tuning Dynamics β is tuned among {1, 2, 4, 8, 16}

Reference 13

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T13:04:51.813326Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-08T06:32:00.761636+00:00.

Observation 5163e5ea-d792-43ae-bc6f-7fca9ea87ca6 · outbound

This paper cites During fine-tuning, the learning rate is tuned among the set {3e − 4, 4e − 4, 5e − 4, 6e − 4, 7e − 4, 8e − 4, 9e − 4, 1e − 3, 2e − 3, 3e − 3}, with W0, Win, Wout frozen.

Beyond Zero Initialization: Investigating the Impact of Non-Zero Initialization on LoRA Fine-Tuning Dynamics During fine-tuning, the learning rate is tuned among the set {3e − 4, 4e − 4, 5e − 4, 6e − 4, 7e − 4, 8e − 4, 9e − 4, 1e − 3, 2e − 3, 3e − 3}, with W0, Win, Wout frozen

Reference 64

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T13:04:51.961281Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-08T06:32:00.761636+00:00.

Observation b778f28d-8e7d-4a9a-bf9d-1c83bedf7f3a · outbound

This paper cites The power of scale for parameter-efficient prompt tuning.

Beyond Zero Initialization: Investigating the Impact of Non-Zero Initialization on LoRA Fine-Tuning Dynamics The power of scale for parameter-efficient prompt tuning

Reference 2015

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T13:04:53.369672Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-08T06:32:00.761636+00:00.

Observation bc2ffb81-9242-45d4-a637-3cec422479d2 · outbound

This paper cites The model dimensions (d, n, r) are set to (784, 4096, 32).

Beyond Zero Initialization: Investigating the Impact of Non-Zero Initialization on LoRA Fine-Tuning Dynamics The model dimensions (d, n, r) are set to (784, 4096, 32)

Reference 2017

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T13:04:52.132044Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-08T06:32:00.761636+00:00.

Observation aeba0c84-e1f1-4834-a932-526881f9046a · outbound

This paper cites Loftq: Lora-fine-tuning-aware quantiza- tion for large language models.

Beyond Zero Initialization: Investigating the Impact of Non-Zero Initialization on LoRA Fine-Tuning Dynamics Loftq: Lora-fine-tuning-aware quantiza- tion for large language models

Reference 2021

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T13:04:53.222241Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-08T06:32:00.761636+00:00.

Observation 6f41f601-663e-4045-ad59-4ddcab091e89 · outbound

This paper cites an unresolved cited work.

Beyond Zero Initialization: Investigating the Impact of Non-Zero Initialization on LoRA Fine-Tuning Dynamics Unresolved cited work

Reference 2022

Resolution: unresolved; raw_fallback, observed 2026-08-07T13:04:53.546125Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-08T06:32:00.761636+00:00.

Observation 3f9971aa-6445-425d-a790-5658c547f5ce · outbound

This paper cites Mixed-Precision Embeddings for Large-Scale Recommendation Models.

Beyond Zero Initialization: Investigating the Impact of Non-Zero Initialization on LoRA Fine-Tuning Dynamics Mixed-Precision Embeddings for Large-Scale Recommendation Models

Reference 2023

Resolution: unresolved; no resolver link, observed 2026-08-07T13:04:50.439569Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation be83a3ac-7c09-4b1b-b4e3-c61fe20834d8 · outbound

This paper cites H., and Gao, J.

Beyond Zero Initialization: Investigating the Impact of Non-Zero Initialization on LoRA Fine-Tuning Dynamics H., and Gao, J

Reference 2024

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T13:04:53.062113Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-08T06:32:00.761636+00:00.

Observation 88c6770b-7689-4bc5-98cc-d8d55de77042 · outbound

This paper cites Background A.1.

Beyond Zero Initialization: Investigating the Impact of Non-Zero Initialization on LoRA Fine-Tuning Dynamics Background A.1

Reference 2025

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T13:04:52.909014Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-08T06:32:00.761636+00:00.

Pith citing papers

Observation c78e9202-d87b-4ac9-ba9e-f1012cdec212 · inbound

Balanced LoRA: Removing Parameter Invariance to Accelerate Convergence cites this paper.

Balanced LoRA: Removing Parameter Invariance to Accelerate Convergence Beyond Zero Initialization: Investigating the Impact of Non-Zero Initialization on LoRA Fine-Tuning Dynamics

Reference 4

Resolution: verified exact; arxiv_id, observed 2026-06-28T23:02:45.981519Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-08T06:32:00.761636+00:00.

Observation 190627c4-f21a-4fdc-8d18-fb2e58d8305f · inbound

PoLoRA: A Preconditioned Orthogonalized LoRA Optimizer cites this paper.

PoLoRA: A Preconditioned Orthogonalized LoRA Optimizer Beyond Zero Initialization: Investigating the Impact of Non-Zero Initialization on LoRA Fine-Tuning Dynamics

Reference 33

Resolution: unresolved; no resolver link, observed 2026-08-01T17:32:34.475671Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.