When the Edge Meets Transformers: Distributed Inference with Transformer Models
Bibliographic record
Abstract
Transformer models achieved significant break-throughs in a wide variety of applications, yet their exorbitant computation costs pose significant challenges when it comes to deploying these models for inference, especially on resource-constrained edge devices. In this paper, we introduce the concept of cross-device distributed inference to transformer models, which accelerates the speed of inference by distributing its workload among multiple edge devices. Unlike previous work designed for multi-GPU environments, the challenge of distributing inference workload on edge devices includes not only limited computation power, but also low bandwidth connections to exchange intermediate results. To address these challenges, we propose Voltage, a distributed inference system tailored for edge devices. By exploiting the inherent parallelizability of the input sequence, Voltage partitions the transformer inference workload based on positions to accelerate the inference speed. We also analyze the relationship between the partition settings and the computation complexity, which allows Voltage to adaptively select the most efficient computation scheme. To demonstrate its effectiveness and generalizability, the performance of Voltage has been evaluated in the context of well-known transformer models, and in a variety of experimental settings. Our results show that Voltage significantly outperforms tensor parallelism by reducing the communication size by 4 x, thereby accelerate the inference speed by up to 32.2 % compared with single device deployment.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.002 | 0.009 |
| Meta-epidemiology (narrow) | 0.001 | 0.001 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.001 | 0.001 |
| Science and technology studies | 0.001 | 0.001 |
| Scholarly communication | 0.002 | 0.005 |
| Open science | 0.002 | 0.002 |
| Research integrity | 0.001 | 0.002 |
| Insufficient payload (model declined to judge) | 0.003 | 0.001 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".