Paper Citation Record · LEDGER

FlashRecovery: Fast and Low-Cost Recovery from Failures for Large-Scale Training of LLMs

As of 18 August 2026, this Paper Citation Record lists 29 of 29 outbound references and 1 inbound Pith citation observation for arXiv:2509.03047.

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

pith.paper-citation-record.v1
2509.03047 v1

Coverage vector

measured 29 of 29 reference resolution

Typed states for the displayed outbound observations.

Source: paper_references, paper_reference_links, observed 2026-08-15T16:37:58.541062Z

measured 30 of 30 standing notices

One-hop event checks from named stored sources.

Source: scholarly_work_events, retraction_status_cache, observed 2026-08-18T06:34:40.430872+00:00

measured 1 of 1 inbound itemization

Pith citing papers itemized under the disclosed page cap.

Source: paper_references, paper_reference_links, observed 2026-05-08T18:19:41.302164Z

measured 0 of 1 external citation measurements

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

Source: arxiv_reference, observed 2026-05-09T06:35:39.100050Z

Reference resolution

29 of 29 outbound references displayed

verified exact1
verified fuzzy14
unresolved14
parse uncertain0
malformed identifier0
metadata mismatch0

External citation measurements

No source-named external measurement is stored.

Outbound references

Observation 5945f027-025d-45a8-a696-ee649e690a1b · outbound

This paper cites GPT-4 Technical Report.

FlashRecovery: Fast and Low-Cost Recovery from Failures for Large-Scale Training of LLMs GPT-4 Technical Report

Reference 1

Resolution: unresolved; no resolver link, observed 2026-08-15T16:37:58.416767Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation d416538b-b179-4ec1-9326-55b5a5d5d332 · outbound

This paper cites Is your code generated by chatgpt really correct? rigorous evaluation of large language models for code generation,.

FlashRecovery: Fast and Low-Cost Recovery from Failures for Large-Scale Training of LLMs Is your code generated by chatgpt really correct? rigorous evaluation of large language models for code generation,

Reference 2

Resolution: unresolved; no resolver link, observed 2026-08-15T16:37:58.422548Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 23de1fb2-58d3-49ec-9c53-0be202d09fcd · outbound

This paper cites DeepSpeed4Science Initiative: Enabling Large-Scale Scientific Discovery through Sophisticated AI System Technologies.

FlashRecovery: Fast and Low-Cost Recovery from Failures for Large-Scale Training of LLMs DeepSpeed4Science Initiative: Enabling Large-Scale Scientific Discovery through Sophisticated AI System Technologies

Reference 3

Resolution: unresolved; no resolver link, observed 2026-08-15T16:37:58.427111Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 7394f22f-0879-4349-a63c-76fa634708cb · outbound

This paper cites Scaling Laws for Neural Language Models.

FlashRecovery: Fast and Low-Cost Recovery from Failures for Large-Scale Training of LLMs Scaling Laws for Neural Language Models

Reference 4

Resolution: unresolved; no resolver link, observed 2026-08-15T16:37:58.431587Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 2a14e99e-cf18-4ad1-a406-8d315e75dd62 · outbound

This paper cites Zero: Memory optimizations toward training trillion parameter models,.

FlashRecovery: Fast and Low-Cost Recovery from Failures for Large-Scale Training of LLMs Zero: Memory optimizations toward training trillion parameter models,

Reference 5

Resolution: unresolved; no resolver link, observed 2026-08-15T16:37:58.436305Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation cd7afc78-b4d5-40a2-b8c3-0aca8c42ff7b · outbound

This paper cites Colossal-ai: A unified deep learning system for large-scale parallel training,.

FlashRecovery: Fast and Low-Cost Recovery from Failures for Large-Scale Training of LLMs Colossal-ai: A unified deep learning system for large-scale parallel training,

Reference 6

Resolution: unresolved; no resolver link, observed 2026-08-15T16:37:58.440176Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 0bb3527d-0304-4d70-b0bc-1a87b6882214 · outbound

This paper cites Megatron-LM: Training Multi-Billion Parameter Language Models Using Model Parallelism.

FlashRecovery: Fast and Low-Cost Recovery from Failures for Large-Scale Training of LLMs Megatron-LM: Training Multi-Billion Parameter Language Models Using Model Parallelism

Reference 7

Resolution: unresolved; no resolver link, observed 2026-08-15T16:37:58.444382Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 81eb126e-fcd0-4f55-99bb-2aee298e55fe · outbound

This paper cites Bloom: A 176b-parameter open-access multilingual language model,.

FlashRecovery: Fast and Low-Cost Recovery from Failures for Large-Scale Training of LLMs Bloom: A 176b-parameter open-access multilingual language model,

Reference 8

Resolution: verified fuzzy; raw_fallback, observed 2026-08-15T16:37:58.930683Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-18T06:34:40.430872+00:00.

Observation 697d7c74-9ef7-41db-baf2-9d65b3ef7695 · outbound

This paper cites Metaseq: Opt-175 logbook,.

FlashRecovery: Fast and Low-Cost Recovery from Failures for Large-Scale Training of LLMs Metaseq: Opt-175 logbook,

Reference 9

Resolution: verified fuzzy; raw_fallback, observed 2026-08-15T16:37:58.915447Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-18T06:34:40.430872+00:00.

Observation 99ad5573-e904-4e12-a6dd-3d2cbd54a51b · outbound

This paper cites The Llama 3 Herd of Models.

FlashRecovery: Fast and Low-Cost Recovery from Failures for Large-Scale Training of LLMs The Llama 3 Herd of Models

Reference 10

Resolution: unresolved; no resolver link, observed 2026-08-15T16:37:58.457341Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 4a8d17c5-fd21-4d1b-8e63-d7dd21deea3b · outbound

This paper cites Evaluating disaster recovery plans using the cloud,.

FlashRecovery: Fast and Low-Cost Recovery from Failures for Large-Scale Training of LLMs Evaluating disaster recovery plans using the cloud,

Reference 11

Resolution: verified fuzzy; raw_fallback, observed 2026-08-15T16:37:58.902569Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-18T06:34:40.430872+00:00.

Observation ffec4c98-d280-4001-ace9-a7874146a622 · outbound

This paper cites Just-in-time checkpointing: Low cost error recovery from deep learning training failures,.

FlashRecovery: Fast and Low-Cost Recovery from Failures for Large-Scale Training of LLMs Just-in-time checkpointing: Low cost error recovery from deep learning training failures,

Reference 12

Resolution: verified fuzzy; raw_fallback, observed 2026-08-15T16:37:58.889992Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-18T06:34:40.430872+00:00.

Observation 365391e2-5239-4387-994b-9bdfdc4dc666 · outbound

This paper cites Efficient large-scale language model training on gpu clusters using megatron-lm,.

FlashRecovery: Fast and Low-Cost Recovery from Failures for Large-Scale Training of LLMs Efficient large-scale language model training on gpu clusters using megatron-lm,

Reference 13

Resolution: verified fuzzy; raw_fallback, observed 2026-08-15T16:37:58.876966Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-18T06:34:40.430872+00:00.

Observation 1f9c979b-47d3-45ee-bce1-34b477f45b6b · outbound

This paper cites Gpipe: Efficient training of giant neural networks using pipeline parallelism,.

FlashRecovery: Fast and Low-Cost Recovery from Failures for Large-Scale Training of LLMs Gpipe: Efficient training of giant neural networks using pipeline parallelism,

Reference 14

Resolution: verified fuzzy; raw_fallback, observed 2026-08-15T16:37:58.863136Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-18T06:34:40.430872+00:00.

Observation 753678ce-7c28-4060-a522-56537fec87eb · outbound

This paper cites Reducing activation recomputation in large transformer models,.

FlashRecovery: Fast and Low-Cost Recovery from Failures for Large-Scale Training of LLMs Reducing activation recomputation in large transformer models,

Reference 15

Resolution: unresolved; no resolver link, observed 2026-08-15T16:37:58.478336Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation a54368dd-c1f5-4f29-91a4-7f5e4f5482f9 · outbound

This paper cites TRANSOM: An Efficient Fault-Tolerant System for Training LLMs.

FlashRecovery: Fast and Low-Cost Recovery from Failures for Large-Scale Training of LLMs TRANSOM: An Efficient Fault-Tolerant System for Training LLMs

Reference 16

Resolution: unresolved; no resolver link, observed 2026-08-15T16:37:58.482820Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation f4186c31-4c0e-4fa4-9eae-8c48b4e53c22 · outbound

This paper cites Unicron: Economizing Self-Healing LLM Training at Scale.

FlashRecovery: Fast and Low-Cost Recovery from Failures for Large-Scale Training of LLMs Unicron: Economizing Self-Healing LLM Training at Scale

Reference 17

Resolution: unresolved; no resolver link, observed 2026-08-15T16:37:58.489021Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 9fd2580d-70aa-4df0-ba9f-ccb23c9aeb28 · outbound

This paper cites Megascale: Scaling large language model training to more than 10,000 gpus,.

FlashRecovery: Fast and Low-Cost Recovery from Failures for Large-Scale Training of LLMs Megascale: Scaling large language model training to more than 10,000 gpus,

Reference 18

Resolution: verified fuzzy; raw_fallback, observed 2026-08-15T16:37:58.842634Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-18T06:34:40.430872+00:00.

Observation 870a91f6-2235-4c16-9ad9-63767c299acd · outbound

This paper cites MoC-System: Efficient Fault Tolerance for Sparse Mixture-of-Experts Model Training.

FlashRecovery: Fast and Low-Cost Recovery from Failures for Large-Scale Training of LLMs MoC-System: Efficient Fault Tolerance for Sparse Mixture-of-Experts Model Training

Reference 19

Resolution: verified exact; local_arxiv, observed 2026-08-15T16:37:58.624673Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-18T06:34:40.430872+00:00.

Observation 9d05668e-6d71-4f22-80a7-4fda4f38d769 · outbound

This paper cites Efficient Training of Large Language Models on Distributed Infrastructures: A Survey.

FlashRecovery: Fast and Low-Cost Recovery from Failures for Large-Scale Training of LLMs Efficient Training of Large Language Models on Distributed Infrastructures: A Survey

Reference 20

Resolution: unresolved; no resolver link, observed 2026-08-15T16:37:58.502783Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation c6593d39-28fc-48a9-91ce-6c52a1f311d8 · outbound

This paper cites Datastates-llm: Lazy asynchronous checkpointing for large language models,.

FlashRecovery: Fast and Low-Cost Recovery from Failures for Large-Scale Training of LLMs Datastates-llm: Lazy asynchronous checkpointing for large language models,

Reference 21

Resolution: verified fuzzy; raw_fallback, observed 2026-08-15T16:37:58.828497Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-18T06:34:40.430872+00:00.

Observation 3400d5c3-3a11-44cd-b6f3-bceb8f94e5a2 · outbound

This paper cites Checkfreq: Frequent, fine-grained dnn checkpointing,.

FlashRecovery: Fast and Low-Cost Recovery from Failures for Large-Scale Training of LLMs Checkfreq: Frequent, fine-grained dnn checkpointing,

Reference 22

Resolution: verified fuzzy; raw_fallback, observed 2026-08-15T16:37:58.815194Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-18T06:34:40.430872+00:00.

Observation 693b984d-08ca-408a-9685-53e4135faba7 · outbound

This paper cites A cost-efficient failure-tolerant scheme for distributed dnn training,.

FlashRecovery: Fast and Low-Cost Recovery from Failures for Large-Scale Training of LLMs A cost-efficient failure-tolerant scheme for distributed dnn training,

Reference 23

Resolution: verified fuzzy; raw_fallback, observed 2026-08-15T16:37:58.796893Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-18T06:34:40.430872+00:00.

Observation fcf4fbf8-d4a0-4694-9b70-a1390445ff05 · outbound

This paper cites Check-n-run: A checkpointing system for training deep learning recommendation models,.

FlashRecovery: Fast and Low-Cost Recovery from Failures for Large-Scale Training of LLMs Check-n-run: A checkpointing system for training deep learning recommendation models,

Reference 24

Resolution: verified fuzzy; raw_fallback, observed 2026-08-15T16:37:58.780370Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-18T06:34:40.430872+00:00.

Observation 72e414b3-4563-4f4b-b95e-a6f5e742febb · outbound

This paper cites Deepfreeze: Towards scalable asynchronous checkpointing of deep learning models,.

FlashRecovery: Fast and Low-Cost Recovery from Failures for Large-Scale Training of LLMs Deepfreeze: Towards scalable asynchronous checkpointing of deep learning models,

Reference 25

Resolution: verified fuzzy; raw_fallback, observed 2026-08-15T16:37:58.765530Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-18T06:34:40.430872+00:00.

Observation 1b9b73ab-1fcd-48c7-9497-36ee3aeaa9d4 · outbound

This paper cites Universal Checkpointing: A Flexible and Efficient Distributed Checkpointing System for Large-Scale DNN Training with Reconfigurable Parallelis.

FlashRecovery: Fast and Low-Cost Recovery from Failures for Large-Scale Training of LLMs Universal Checkpointing: A Flexible and Efficient Distributed Checkpointing System for Large-Scale DNN Training with Reconfigurable Parallelis

Reference 26

Resolution: unresolved; no resolver link, observed 2026-08-15T16:37:58.527618Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 4f10ff25-b57c-45cf-a70a-35652420a2f8 · outbound

This paper cites ByteCheckpoint: A Unified Checkpointing System for Large Foundation Model Development.

FlashRecovery: Fast and Low-Cost Recovery from Failures for Large-Scale Training of LLMs ByteCheckpoint: A Unified Checkpointing System for Large Foundation Model Development

Reference 27

Resolution: unresolved; no resolver link, observed 2026-08-15T16:37:58.532754Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation e09f494b-7704-4f35-a878-674f3cffa4ed · outbound

This paper cites Swift: Expedited failure recovery for large-scale dnn training,.

FlashRecovery: Fast and Low-Cost Recovery from Failures for Large-Scale Training of LLMs Swift: Expedited failure recovery for large-scale dnn training,

Reference 28

Resolution: verified fuzzy; raw_fallback, observed 2026-08-15T16:37:58.744392Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-18T06:34:40.430872+00:00.

Observation a7aeba1a-b462-4e97-9311-11c64d7ec529 · outbound

This paper cites Parcae: Proactive,liveput-optimized dnn training on preemptible instances,.

FlashRecovery: Fast and Low-Cost Recovery from Failures for Large-Scale Training of LLMs Parcae: Proactive,liveput-optimized dnn training on preemptible instances,

Reference 29

Resolution: verified fuzzy; raw_fallback, observed 2026-08-15T16:37:58.731224Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-18T06:34:40.430872+00:00.

Pith citing papers

Observation 7ac5040a-21ae-43fc-a248-ea58525c51d2 · inbound

Towards Robust LLM Post-Training: Automatic Failure Management for Reinforcement Fine-Tuning cites this paper.

Towards Robust LLM Post-Training: Automatic Failure Management for Reinforcement Fine-Tuning FlashRecovery: Fast and Low-Cost Recovery from Failures for Large-Scale Training of LLMs

Reference 9

Resolution: verified exact; arxiv_id, observed 2026-05-09T06:35:39.101920Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-18T06:34:40.430872+00:00.