Roofline-inspired scaling model predicts Transformer fine-tuning energy consumption across GPU configurations
A new arXiv preprint presents a framework for modeling energy consumption during Transformer training on multiple GPUs, using BERT architectural sweeps to relate measured energy to proxies for compute, memory traffic, and hardware efficiency. The approach adapts roofline modeling with a speedup-based hardware-efficiency factor that accounts for tensor parallelism and fully sharded data parallelism. The resulting scaling law accurately predicts training energy across heterogeneous configurations, targeting sustainable and cost-aware system design.
Related guides (2)
Related events (8)
Energy-based transformers as unified predictors of reading difficulty in computational psycholinguistics
A new arXiv preprint introduces energy-based transformer measures as predictors of human reading difficulty, evaluated across three reading-time corpora (Natural Stories, UCL eye-tracking, UCL self-paced reading). The energy measure outperforms surprisal alone and appears to subsume both surprisal and attention entropy effects, suggesting it could serve as a single unified predictor. The work connects transformer language models to Hopfield networks and dense associative memory literature, marking the first application of energy-based transformer measures in computational psycholinguistics.
Fit More and Train Faster With ZeRO via DeepSpeed and FairScale
This Hugging Face blog post from January 2021 covers integration of ZeRO (Zero Redundancy Optimizer) memory optimization techniques via DeepSpeed and FairScale into the Transformers training ecosystem. ZeRO partitions optimizer states, gradients, and model parameters across GPUs to enable training of much larger models on the same hardware. The post serves as a practical guide for practitioners looking to scale model training without additional infrastructure investment.
Dynamic short convolutions yield 1.33–1.60× compute advantage over standard Transformers
A new arXiv preprint introduces dynamic short convolutions as an architectural primitive for Transformers, using input-dependent filters to combine locality bias with increased expressivity. Experiments across 150M–2B parameter language models show consistent perplexity improvements over standard Transformers and static convolution variants, with scaling-law fits indicating a 1.33× compute advantage when applied to key/query/value vectors and 1.60× when added after every linear layer. The technique also improves linear RNNs (Mamba-2, Gated DeltaNet) and mixture-of-experts architectures, with custom Triton kernels making training practical.
Introducing Optimum: The Optimization Toolkit for Transformers at Scale
Hugging Face announced Optimum, an optimization toolkit designed to accelerate Transformers models on various hardware backends. The toolkit aims to bridge the gap between Transformers model development and hardware-specific optimizations from partners. It provides a unified interface for quantization, pruning, and hardware-accelerated inference across different accelerators.
Test-Time Training End-to-End (TTT-E2E) Retrains Model Weights to Handle Long Inputs
Researchers from Astera Institute, Nvidia, Stanford, UC Berkeley, and UC San Diego introduced TTT-E2E, a method that compresses long context into transformer weights by training the model during inference via meta-learning. The approach uses sliding-window attention restricted to 8,000 tokens and updates only the fully connected layers of the last quarter of the network on each 1,000-token chunk at inference time, keeping per-token generation latency roughly constant as context scales to 128,000 tokens. TTT-E2E slightly outperforms vanilla transformers on next-token prediction loss across long contexts and matches efficient architectures like Mamba 2 and Gated DeltaNet on inference speed, but fails dramatically on Needle-in-a-Haystack retrieval beyond 8,000 tokens and incurs substantially higher training latency. The work reframes long-context handling as a training-inference trade-off rather than an architectural design problem.
Framework for Evaluating Datacenter Power Delivery Hierarchies for AI Workloads
Researchers from Microsoft Azure present a simulation framework for evaluating datacenter power delivery designs under AI-era conditions, where rack power density is projected to approach 1MW per deployment by 2027. The framework combines GPU/compute/storage projection models with production operational data to assess throughput, power, and cost metrics across realistic deployment sequences. Key findings show that multi-resource stranding materially affects deployable capacity and effective capital expenditure, and that the correct planning objective is deployable capacity over time rather than installed megawatts. The work addresses the challenge of designing power hierarchies that remain efficient across multiple hardware generations as AI accelerator density rises.
Accelerating PyTorch Transformers with Intel Sapphire Rapids - Part 2
This Hugging Face blog post covers inference optimization techniques for PyTorch Transformer models on Intel Sapphire Rapids (4th Gen Xeon) CPUs. It likely demonstrates performance gains using hardware-specific features such as AMX (Advanced Matrix Extensions) and BF16 support. The post is part of a series focused on making transformer inference more efficient on Intel server hardware without requiring GPU acceleration.
Hyperfitting Explained: Terminal Geometric Expansion in Final Transformer Layers Drives Diversity Gains
This paper investigates the 'hyperfitting' phenomenon—where fine-tuning LLMs to near-zero loss on small datasets improves open-ended generation and reduces repetition—and demonstrates it is mechanistically distinct from temperature scaling. Entropy-matched control experiments falsify both the temperature-equivalence and static vocabulary reweighting hypotheses, instead localizing the effect to a 'Terminal Expansion' in the final transformer block where feature-space dimensionality expands by ~80.8 dimensions, enabling promotion of deep-tail tokens via context-dependent rank reordering. The authors introduce Late-Stage LoRA, a targeted fine-tuning strategy updating only the final 5 layers, achieving robust generation with minimal parameter updates.