Quantifying Hyperparameter Transfer: Embedding Layer Learning Rate as Key Driver of μP Benefits
This paper develops a three-metric framework to quantify hyperparameter transfer quality across model scales, targeting the problem of extrapolating optimal hyperparameters from small to large LLMs. The central empirical finding is that the well-known advantage of Maximal Update Parameterization (μP) over standard parameterization (SP) with AdamW largely reduces to a single factor: the embedding layer learning rate. In SP, the embedding layer acts as a training bottleneck causing instabilities; scaling its learning rate by model width to match μP substantially stabilizes training and improves transfer. The paper also characterizes how weight decay affects scaling law fit quality versus extrapolation robustness in opposite directions.
Related guides (2)
Related events (8)
Complete-muE: Optimal Hyperparameter Transfer and Scaling for MoE Models
Complete-muE is a framework for transferring hyperparameters across dense FFN and Mixture-of-Experts (MoE) transformer architectures, addressing limitations of existing tools like μP and SDE that cannot handle simultaneous architecture and token-per-expert changes. It uses a two-bridge system: Bridge I maps dense FFN to Dense MoE via active-width μP with normalized router scale, and Bridge II maps Dense MoE to sparse MoE via activated-expert scaling with a first-order SDE correction. The practical outcome is a 'tune dense once, transfer to all' recipe that enables near-optimal hyperparameter reuse across MoE configurations without costly re-tuning. Experiments on language model and diffusion model pretraining confirm stable hyperparameter optima across architectures and parameter counts.
On the Scaling of PEFT: Towards Million Personal Models of Trillion Parameters
This paper reframes parameter-efficient fine-tuning (PEFT) not merely as a cheaper alternative to full fine-tuning, but as a substrate for persistent, instance-specific personal models layered atop shared foundation models. The authors analyze three scaling axes: Scale Up (stronger base models amplifying adapter utility), Scale Down (minimum viable adapter size), and Scale Out (managing millions of concurrent adapted instances). They introduce MinT as an infrastructure reference for adapter identity, versioning, provenance, evaluation, and serving at scale.
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.
CLP: Lightweight collocation-length predictor achieves zero-loss multi-token inference speedup
Researchers propose CLP (Collocation-Length Predictor), a span-level decision layer for accelerating LLM inference via multi-token prediction without quality degradation. The key insight is 'Backbone-as-Architect': the backbone LM head always generates the first token while MTP heads handle only subsequent tokens, eliminating head-backbone competition that causes repetitive outputs in prior methods. CLP uses a single linear layer (~4.6K–7.7K parameters) versus 1M-parameter gate networks in prior work, achieving 1.14x–1.29x speedup on Qwen2.5 models with near-zero repetition ratio. The paper also establishes that shorter prediction horizons improve MTP head accuracy on larger models, offering a scaling-aware design principle.
Scaling Laws for Reward Model Overoptimization
OpenAI published research investigating how reward model overoptimization scales with policy and reward model size in RLHF pipelines. The work characterizes the relationship between KL divergence from the initial policy and gold-standard reward, finding predictable degradation patterns as optimization pressure increases. This provides empirical grounding for understanding Goodhart's Law dynamics in language model fine-tuning and has implications for designing safer, more robust RLHF training regimes.
Shannon Scaling Law: A Noisy-Channel Framework for LLM Capacity and Non-Monotonic Training Phenomena
Researchers propose the Shannon Scaling Law, a theoretical framework that models LLM training as information transmission over a noisy channel using the Shannon-Hartley theorem. By mapping model parameters to channel bandwidth and training tokens to signal power, the framework introduces a fundamental SNR-based capacity limit that explains non-monotonic phenomena like catastrophic overtraining and quantization-induced degradation that classical power-law scaling laws cannot capture. Validated on Pythia and OLMo2 under Gaussian noise, quantization, and fine-tuning perturbations, the law achieves strong R² scores and successfully extrapolates from 6.9B to 12B parameter models trained on up to 307B tokens. The framework outperforms both classical and perturbation-aware scaling laws, predicting U-shaped performance degradation when SNR is insufficient.
PEFT-Arena: Benchmarking Parameter-Efficient Finetuning via Stability-Plasticity Trade-offs
PEFT-Arena is a new benchmark that evaluates parameter-efficient finetuning methods jointly on downstream task performance and retention of pretrained general capabilities, framing the problem as a stability-plasticity dilemma. Across methods tested under comparable parameter budgets, orthogonal finetuning achieves the best Pareto frontier. The paper provides geometric analyses in both weight space (spectral/singular-value structure) and activation space (representation distortion metrics) to explain why different PEFT methods differ in forgetting behavior. A practical finding is that final SFT checkpoints often overshoot an optimal retention operating point, motivating path-wise rewinding as a post-hoc correction.
Forecasting Downstream LLM Performance With Token-Level Proxy Metrics
Researchers propose proxy metrics constructed from token-level statistics (entropy, top-k accuracy, expert token rank) drawn from a candidate model's next-token distribution over expert-written solutions, as a cheaper and more reliable alternative to cross-entropy loss or direct downstream evaluation. Across three settings—cross-family model selection, pretraining data selection, and training-time forecasting—the proxies consistently outperform baselines, achieving mean Spearman Rho of 0.81 vs. 0.36 for cross-entropy loss on model ranking, and reducing compute for data selection by roughly 10,000×. The method enables downstream performance extrapolation across an 18× compute horizon with roughly half the error of existing alternatives, suggesting expert trajectories are broadly useful signals throughout the model development lifecycle.