Pith. sign in

REVIEW 3 cited by

Angel-PTM: A Scalable and Economical Large-scale Pre-training System in Tencent

Not yet reviewed by Pith; the record is open.

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2303.02868 v1 pith:MACTQGW3 submitted 2023-03-06 cs.LG cs.DC

Angel-PTM: A Scalable and Economical Large-scale Pre-training System in Tencent

classification cs.LG cs.DC
keywords angel-ptmmodelslarge-scaletencentmemorymodelpre-trainedpre-training
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved
0 comments
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Recent years have witnessed the unprecedented achievements of large-scale pre-trained models, especially the Transformer models. Many products and services in Tencent Inc., such as WeChat, QQ, and Tencent Advertisement, have been opted in to gain the power of pre-trained models. In this work, we present Angel-PTM, a productive deep learning system designed for pre-training and fine-tuning Transformer models. Angel-PTM can train extremely large-scale models with hierarchical memory efficiently. The key designs of Angel-PTM are the fine-grained memory management via the Page abstraction and a unified scheduling method that coordinate the computations, data movements, and communications. Furthermore, Angel-PTM supports extreme model scaling with SSD storage and implements the lock-free updating mechanism to address the SSD I/O bandwidth bottlenecks. Experimental results demonstrate that Angel-PTM outperforms existing systems by up to 114.8% in terms of maximum model scale as well as up to 88.9% in terms of training throughput. Additionally, experiments on GPT3-175B and T5-MoE-1.2T models utilizing hundreds of GPUs verify the strong scalability of Angel-PTM.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 3 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Tencent Advertising Algorithm Challenge 2025: All-Modality Generative Recommendation

    cs.IR 2026-04 accept novelty 7.0

    Releases TencentGR-1M and TencentGR-10M datasets with baselines for all-modality generative recommendation in advertising, including weighted evaluation for conversions.

  2. From PowerSGD to PowerSGD+: Low-Rank Gradient Compression for Distributed Optimization with Convergence Guarantees

    math.OC 2025-09 conditional novelty 6.0

    PowerSGD can provably fail to converge; the proposed PowerSGD+ with periodic SVD subspace resets converges under standard assumptions at O(1/sqrt(NT)).

  3. HunyuanVideo: A Systematic Framework For Large Video Generative Models

    cs.CV 2024-12 unverdicted novelty 5.0

    HunyuanVideo presents a 13B-parameter open-source video generative model with integrated data, architecture, training, and inference systems whose professional evaluations show it outperforming prior SOTA models inclu...